Symposium on Blockchain for Robotics and AI Systems

>> Hello? Hello, everybody. Thank you for coming We are super lucky because actually today is pretty nice You cannot imagine what happened to me two days ago, when I saw I was coming, I thought there was going to be a massive problem for you to be here, but fortunately, you are here. I’m very dry. So that’s fantastic Welcome to the second edition of the symposium of blockchain for robotics Testing testing

But I would like to give you some kind of vision And I will start with my own story So these — one of these guys here, right, was my Ph.D. advisor The other is a robot clone Right? Let me give you some context. I get my master and Ph.D. in Japan. One of these guys was my supervisor One of these guys is called professor Yoshudo. He was very interested in how he can be in two places at the same time He really wanted to explore this. He was a very simple professor in Japan and he was hired in give lectures in Tokyo and all other the place. He realized he was spending a lot of time traveling, traveling to the places he really didn’t want to be. So he realized that, okay, what happens if somebody could represent me in that place What happens if I would give this in proxy? He decided to build a very realistic robot and send it to the place he didn’t want to be This makes more sense when you see this So what you’re seeing here is basically the two together You can see the robot as a very, very expensive Skype client Now, the robot is sent to the place where he doesn’t want to be. Right? He just operates the robot, like through cameras and sensors Yeah? Basically what will happen, if you see very realistic image of somebody that talks like that person, that moves like that person, and reacts like that person, but it’s not that person, because that person is not there So in a certain sense, he wanted to transfer that persons, right? Let me tell you a story about that So of course, he didn’t want to be in the place that the robot went eventually, but somebody had to move the robot, right? So that was me, right? [laughs] It was not only Japan, but all over the world, and I had to travel with a human torso in my luggage [laughs] So everything security guard knows my face. Right? [laughs] The problem is the future of robotics, we will have a very complex robot that will represent us, one very complex robot, that is expensive and hard to prepare, et cetera. For me, when I was doing my Ph.D I realized that was not my vision for the future of robotics. Mine was more de-centralized

I thought by making robots very simple and easy to repair but putting them into big groups, you can achieve complex tasks. At the same time, you can basically have a lot of nice properties, that one very complex robot doesn’t have What you’re seeing here is very easy example of a swarm of robots. They’re doing something called foraging Basically, robots self-organize in order to find 3D tokens, like in a football field. They can represent sources or data or people. Right? And what they do, basically, these robots, in a very de-centralized — there’s no boss here, no set of command and control operation Find these three tokens. Then they put them into the nest, which is the center of the field. So these nests simulates a human society, for example Once the robot puts, like, a token, the robot gets a recharge in the battery, a rework The robots start to self-organize in a centralized way in order to keep this doing for a long period of time, to achieve sustainable behavior Now that you see this, you think, okay, why is this useful? Why do we care about this? You can start seeing the use of these, right? The interesting thing about these systems is that since they are de-centralized and there’s no single point of failure, I can break one robot or two robots or three robots, and the others will work. This system has robust and has full capabilities by design. Which is very interesting for this kind of like new public infrastructure that we’re trying to envision with robotics But there’s another problem here The world of robotics is very polarized There’s people trying to do research of the theory on the basis of these emerging properties that this system has, and they’re focused on this, but there’s also other people that says, no, these systems are going to deliver packages in five years and we’ll have swarms of self-running cars in cities You know? I realized these two communities are very far apart and there’s nobody trying to bridge these two things, these two big missions of these worlds. This is basically because there’s many things that we didn’t tackle in order to make these systems, which have good capabilities, work, and available. Right? Some of the problems we have for these systems, we don’t have any security standards for these systems. We realize they have good properties, but what happens if at some point in time some of the robots gets hacked or start to misbehave? What will happen to these systems? Will it remain robust? There’s also no good way to understand how these big systems, especially large swarms, can get into agreements and consensus for certain things We don’t have the research based on that, like the boots on the grounds, for these systems And more particularly, we don’t have new business models for these systems. Right? It’s very hard to bridge the gap between academia and the industry So one of the things I realized, like why we have these problems, is because we don’t have good interfaces to these systems. In academia we have a lot of research about how to do human-to-robot interaction with one human and one robot, we don’t have go interface for doing human to groups of robots interaction. It’s because as you scale these systems, it gets more complex. So they are very difficult to audit and very difficult to do this So we started to see this is a need. For example, an article I found, a couple of weeks ago, in which researchers started to say if we have self–driving cars in New York, for example, we know which cars, for example, are in Times Square, we can actually block part of Manhattan You might think hacking a self-driving car is complex, but it’s not It’s extremely simple There’s been people traveling on the highway or driving. Yeah What this article projects a little

bit is we’re trying to create new things but we’re not covering the holes this is providing Do you remember three or four years ago, ransom ware, that encrypts your computer or hard drive and if you don’t pay me $1,000 in bitcoin, you’re not going to get your hard drive back. Well, that was cute, but imagine if you’re driving down the highway, and at some point a pop-up comes up on the dashboard, I’m not going to brake unless you pay me $1,000 in Bitcoin So we tried to tackle these problems Basically this shows for the first time how a group of vehicles can self-police or monitor each other What you’re seeing here is something very simple. You’re seeing these robots, which are very, very simple, trying to go around this checkerboard and trying to sense the tiles, the color of the tiles, and then come to consensus about what’s the majority color, something super simple One robot goes to part of the checkerboard and senses there’s 30% black tiles and 70% white tiles Once you bump with another robot you exchange opinions and it emerges into a big consensus and at the end we all agree that black is much more in color or white is the majority color. Simple stuff. So we started to simulate what would happen if you introduce Byzantine robots, robots hacked to break the consensus, start to lie, and we realized, well, we can just compare the classical approach, that we have for consensus, and a blockchain-based approach This distinction makes for the fact that if, for example, we use the classical approach, you can get, like, messages around, like, this swarm. But robots basically somehow believe what the other robots are saying. Right? With the blockchain approach, you have a record. You get all these votes into a blockchain recording in every single robot and every time you bump into each other you synchronize it. Right? We saw in this graph you have two axes The X axis is the number of Byzantine robots, the number of bad bots, right? And the Y axis is the exit probability. The number of times they consensus the right color As you start to include more bad bots, the success rate drops dramatically. This means if you are in New York, and you have, like, 300 self-driving cars, and then you hack, let’s say, ten, your success probability that you will have a good system, based on pure peer-to-peer communication, like technology, will drop dramatically And it’s because actually this peer-to-peer gives you really good things, but also gives you a lot of open problems Which is If you use blockchain approach, and you start these bots in transactions among the robots, you’ll start to find inconsistencies in the system. You can start to find the fact, actually, if I told you the majority color is white, and then I tell you the majority color is black, and you start to find I got into an inconsistency We all have the same controller, but I’m starting to change my opinion in a very weird way. So you can assign me a reputation And you can weed me out of the system and continue doing whatever you’re doing. What would happen if we add reputations to the robots, based on the information they provide? That’s the idea With that, we can continue the system, the self-sustaining behavior of the system But this has a problem, also It seems like a very simple case, but what would happen if we try to do more complex actions, sequential actions, where the robots have to

assemble stuff and keep order Right? Normally, what happens is that in these kind of actions, in order to maintain these terms, and these nice probabilities, we need to distribute, for example, the blueprint of what the robots need to do. If they need to make a bridge, for example, they need to understand, okay, the piece one should go here and piece two should go here and piece three should go here If I am broken, because you have the same plan, you can continue the plan. But this also has a problem. If we all have the plan, because we need to maintain these capabilities, the fact that actually we have a lot of us, with a lot of plans replicated, also opens new holes If I’m an attacker and I want to hurt that system the best, I just need to get one robot to understand the plan and then act accordingly What we envision, by exploring this blockchain space, do we have tools in this blockchain space to give a blueprint of the robot without actually giving the data? Turns out, yes, we can do it Many of you might know the concept of mercant tree. It’s binary tree, instead of having the data in the notes or the decisions of whatever, you have hashes. All the hashes are stored in the lower level, and then you encrypt this information and you rehash it, rehash it, rehash it, until you get to the root What we did, in this research, is try, like, to substitute the normal transactions in the blockchain A sends B one Bitcoin, with robot actions that belong to our plan Stuck piece number one, stuck piece number two, stuck piece number three, and then we can encrypt this information up to the root, and we can give that root to the robots So how this works is the fact that, like, an operator designs the whole plan in advance, and says, okay, so in order to build this bridge I need action one with piece number one, and action two with piece number two, and action three with piece number three. You encrypt all this information and then you give this tree to the robots The robots, with that tree, do not know what they have to do because everything is encrypted Right? But they know if at some point in time they find the correct combination of robot action and robot sensor input, let’s say, like that, if they’re in front of piece number one, and they say, okay, what should I do with piece number one? Should I stack it? Move it? And they find the right combination and that ends up being the hash of the first leaf or the second leaf or the third leaf, they know they have to do that, even though they don’t know what the other actions really mean. Right? So we tried this with several, like, missions which could be projected in many other, like, things, but for example I’m going to show you here Robots, we encoded one of these trees. One is an obstacles, there’s empty spaces, and there’s the entrance and the exits. We encoded this tree and then we give it to the robots and say when they’re around, at some point in time, if you find a place that belongs to the tree, just stop there So robots do not know what they have to do Robots cannot infer any details about the plan. But they know once they find a good action they just stop it. Right? So in the end, through a lot of wandering around, robots are able to make the maze. Right? The interesting thing about this, if I now capture any of the robots and say, “What do you know about the plan?” They know nothing They know this hash and this hash is correct, is part of the plan, so I just stopped here Right? So you cannot infer where the entrance or exit might be, which is very interesting for security reasons, right? So we try to project this and say, okay, instead of a very simple maze, can we do large-scale missions? What would happen if we encode the millennium falcon in this tree? We conducted this research and understood that it’s within reach of the technology to do that In coding, like the Millennial Falcon, it only requires 230 kilobytes of memory and there’s communication between the robots So we can achieve that But now, I would like to give you a final touch to this

The last thing that we talked about is the fact there’s missing new business models for the systems What I’m going to present to you here is something that would be present in more detail in the paper representation, but it’s something I hope makes you think a little bit What you’re seeing here is a robot, Kak achu He’s assembling robot parts, but he’s painting pictures He’s basically choosing a Japanese kangee from the internet and basically replicating it with a brush So the interesting thing about Kakachu, once the robot starts painting, there’s an auction that he’s starting Auctioners through the internet can auction what they want to pay for that picture. Right? So once the auction is over, there’s a winner. Right? Of the picture. And the winner instead of, like, giving the money to the owner of the robot, just gives money to the robot’s own account So the robot gets the funds of that process The important thing about this is that those funds are used in order, like, for the robot autonomously to buy everything it needs for the next picture So the robot can buy more canvasses or more paint, the electricity, the internet bill, to paint the next picture. Right? With these, we’re trying to understand what will happen if these robot s s So this is something that I thought was interesting So this is a case where I think So to finish, what we are seeing here is the synergy of different worlds that are set apart, that now they can create something bigger than them alone. Right? Of course, we’re here because we are interested in the world of robotics and the world of AI The robots are here and coming This is, in my opinion, this is not us against the robots. It’s us with the robots And we need to find ways in order to coexist and benefit from these, like, coalition But, of course, we cannot just leave them unattended We cannot leave them autonomy for the sake of autonomy. It comes with a price We need a new interface for that autonomy. 20 years ago we didn’t have certain tools, but now we have them, and with the combination of these two worlds we can do something very powerful, but also we need to put it somewhere. We cannot just leave it anywhere This needs to be placed in a society, like, framework in order to make things in a good way, right? And make societies more greener or efficient. Right? Of course, we’re in the Media Lab today so we care about how to deploy these systems and move from academia outwards So with this, I am over. This is done. Thank you very much [Applause] I don’t know if we are running late in the first presentation, but I would like to introduce you to Professor Sandy Pentland. He’s one of the founders of this place and very, very interested in data, as you will see And without further delay, I give the stage to him >> Thanks. Thank you. Glad to see you’re all here This is a good turnout. Nice day

Blockchain robotics, and how we got here, I refer to my group and me, because I’m not known for robotics, so I thought I’d explain why we’re doing this now and where we think it will go and also bring people together in an interesting way What I am known for is variable computing Back in the early ’90s we did some of the first wearable computing, decorating humans with computers and sensors and stuff like that. It was great fun. It produced a lot of very weird-looking people A lot of people said, “I’ll never wear that,” and there were fashion schools that came up with stuff like this In the early ’90s they were wearing things that looked like iPhones before there was wireless and such. The guy wearing the display there actually went on to do Google Glass But there’s a symbiosis between computers and humans I worked for Nissan and designed the framework for their autonomous vehicle. The goal there was to be able to have a cooperation between people and the machine And one of the real challenges of autonomous vehicles is that you’re going to be in an environment where cars are — the other cars have no autonomy. They’re just people And by other cars are by other manufacturers So you have this requirement to be cooperative without necessarily being able to talk to them in a deep code-to-code sort of way I think that’s the type of system we’ll see more and more of We have these wearable elements, not necessarily robots the way we think about it, but we have to cooperate with them and it’s a mixture of people and machines cooperating to get something done and you want to design the system as a whole One of the main things we learned from doing this was that it’s not about the robots. It’s not about the wearable computers. It’s really about the communication between them It’s actually not that difficult to build a lot of these things. At least first order. But it’s very difficult to get them to coordinate and cooperate with each other. And this has been noticed before with respect to people This is a little quote from Adam Smith in the late 1700s Everybody knows what the invisible hand is, right? The invisible hand is a way of people and institutions cooperating with each other without conscious planning, the way Eddie was just talking about. In today’s society we tend to think of this as being a market property, that comes from the market, but that’s not what Adam Smith said Adam Smith said it was peer-to-peer communication, local communication, and local negotiation, that determined the balance of services and the norms for cooperation So not a global thing, a local emergent property. That’s really interesting. Because for a lot of reasons, local emergent properties are more robust, all sorts of things They’re less susceptible to corruption and attack of various sorts So what we study — oh, and actually Karl Marx said the same thing. These two guys, may be the only time they agreed, but there they are So what we study is we study how you can get systems of communication, local communication, where you continually negotiate policies of action to get a desired overall system performance And a good example of this is network bandits or distributed bandit problems Bandits are things that are little autonomous elements, actors, that have a number of different policies, number of different options, that they can choose from They don’t know the rewards associated with each of the options that they can choose So they experiment to find something that is the best way for them to get along in their environment Okay? Sort of like people There’s actually in biology this is called foraging behavior You see animals experimenting to find better food sources and

things like that This is a mathematical model of that. In distributed systems you have the ability to observe and communicate with other people, and that’s good. It’s a very powerful technique You can imagine early humans, if I see you eat the blueberries and then get sick, I’m not going to eat the blueberries, almost no cost to me, just distributed learning There are different ways to do this Most of those are framed as single users, though, in an environment There has not been nearly as much research on distributed things, although, clearly, there is. So we’re interested in that problem because that maps to these sorts of problems Eddie was just talking about You have bunches of agents, whether people or machines, and they have to learn from each other to coordinate actions that they take that have the best utility for all of them Sounds pretty good Some of the things we’re doing is we’re focusing on environments where one of the things that’s wrong with machine learning today and a lot of estimation is that they implicitly assume concentrated distributions. So noise levels, like normal models, or something like that, but actually in these distributed systems you get cascades. You see me do something, you begin to copy it, he begins to copy it, five other people begin to copy, you get this cascade of behavior Current techniques typically don’t work very well with that at all They go heywire in all sorts of ways. We’re focusing on how can you build systems that are robust to this and actually can learn from these sorts of signals? Byzantine agents, Eddie talked about that a little bit. What happens when the agents are trying to mislead you? And this comes in different flavors. It may not be intentional misleading. It may be they have a very different purpose than you do. So they take an action and report this is a very good action; whereas, you would think it’s a very bad action. How do you detect this and compensate for it? This is certainly key to the sorts of problems Privacy, certainly, with people, you can understand how you don’t want your personal data leaking out everywhere, but the same is probably true in many situations with robots, particularly if robots are agents of people. How do you actually do this communication, which proveably preserves people’s privacy? And unreliability, you can look at that as a form of Byzantine You get screwy stuff happening sometimes. You have to be robust We do work in this area. I’m not going to talk a whole lot about it, except to sort of say we have some very strong results in these areas For each of these cases, we find mathematical scenes that are completely de-centralized and robust and differentially private. You wouldn’t have thought you could do this The key idea is that in these communication channels between agents, you’re trying to model the distributions observed by the agents, and you hold out outliers for later consideration. You say, whoa, that looks weird; I’m going to hold that back, and when I get more of them, I can decide whether the guy is trying to trick me, whether this is a cascade, or other sorts of problems So there’s nice mathematical ways to do this This is very new stuff. Avi is a Ph.D. student who is rocking on this. If you’re interested in the math in it, I do point you at that Hopefully, that sort of gets you a framing of the types of things that we do. I can be happy to talk more about this The other thing we do in my group is we build software to support these sorts of things So we build things that are blockchain systems that have offchain data, and methods of doing communication, auditing, and machine learning on top of them And we’ve been very successful This is me wearing a tie, something you never see, the president of the European union invited me to talk about how they should be handling data for privacy and things like that. That’s basically, you now know the story, blockchain, offchain data, and do

analytics on that In these certain coordination systems, you don’t want to share data. The moment you share the raw data, even if it’s anonymized, you doomed yourself. What you can do is you can share answers about your private data with other people For instance, for differentially private distributed bandit problems, it turns out that sharing means of sort of buckets of action, different types of actions, you share the mean payoff, not the specific payoffs, yields and differentiately private scheme It’s pretty good because you can still get optimal convergence on that sort of a thing The second thing, you have to log things on a log chain so you can go back and remember things accurately, and so other people can query you if there’s a problem. You can essentially debug things if you can show that there’s a problem, you can then go up a level and get higher level permissions to be able to go back and look at the data you need to be able to figure out what’s going on So we build stuff like that I’m not going to spend a lot of time on it We’re building systems for Senegal and Columbia that do this sort of thing and give you this auditing ability It’s pretty amazing The bottom line is there’s two threads that we do We’ll be happy to collaborate with people You’ll here a little bit more about this One thread is building this blockchain system for humans and human institutions, but nowadays that includes robots and actors and autonomous all sorts of things How do you actually guarantee like your thermo stat and things are used in the correct way and not hacked and maintain your privacy? Those are the sorts of questions They’re not interesting robots, but they’re very interesting in IT and control problems, particularly when you talk about a distributed thing, autonomous vehicles and that sort of things Hopefully we’ll find ways to work together and I just wanted to welcome you and tell you why we’re interested here Okay? Thank you [Applause] >> The algorithm? >> The meeting you do every year. The annual meeting >> Yes, the annual meeting, that thing there, we do one coming up in January, and we do one here that’s sort of a big open one, and then we also have a meeting that is much more technical meeting which is for our responsers and collaborators That includes, at this point, seven nations and half a dozen large corporations, like Ernst & Young, Intuit, IBM, et cetera So we have twice of those — twice a year If you’re interested, sure, why not? [laughs] Okay Yeah >> SPEAKER: Yeah, so one of the big concerns that everybody has, right, is dominance of the tech world, by a few players And the whole result of AI and all that is about data So it really boils down to who owns and controls the data? Today, we have sort of come from this wild west, which has allowed certain organizations to become very, very large And it has all sorts of dangers I actually don’t think Google or Facebook are going to come and kill my children They might steal their wallets but they’re not going to kill

them. On the other hand, governments have access to this, too, and they might. [laughs] They’ve historically not behaved themselves The best thing at the moment that I’m pushing is called data cooperatives Under the laws of the US and most of the EU you have these cooperative organizations which are owned by the members and are democratic They’re in Switzerland and there’s a whole variety of institutions here that are typically called credit unions and they’re chartered by the government to manage the money, of course, but also the data, just sort of by accidents in the regulation, of the members So you can have something where a group of people have access to having a copy and controlling their own data. Just to sort of give you a sense of the power of that, currently, a lot of websites have these terms and conditions. It says you can’t use it for this, can’t do that, well, having somebody as a legal representative of you overrides those. Imagine you’re in a hospital and you’re in a coma Your lawyer could see your Facebook page regardless of what Facebook says. Right? And keep a copy of all that data Absolutely. Just like no question So can one of these cooperative things. Legally, it is you. So that’s thing one Thing two, here in Boston, or any place, we have all these hospitals They do treatments on you They give you drugs Nobody, including them, knows if they’re any good So when drugs get approved, everybody stops looking, because they might see something bad Okay? So we take these pills and God knows what happens We don’t know if they interact There’s all sorts of stuff we don’t know Because nobody wants to look And they hide behind privacy law They say, “We can’t share that with you.” But if you had a cooperative, you had a right to your medical record If we had 50,000 people who had their medical records here in Boston, those 50,000 people could analyze their medical records o Not giving them up, still control them, but you agree as a cooperative to ask, “What’s the efficacy of this drug? What are the interactions?” So the people could know Once the people could know, it will happen This is a political action, but a vehicle of political power to have a sort of knowledge-based activity The key is you have to have collectives of people Your data, my data, not very valuable Can’t really get much in the way of insights out of it But if we have 50,000 people in a town, we can tell in the government is any good. We can tell if the hospitals are any good We can tell if the bank is behaving. You name it. We can go right down the list This is a little bit like labor union battles were a century ago A century ago you had this being corporations that owned everything, the robber barrens they were called in this country, and they were exploiting their workers because there were no other options You worked under their rules or you didn’t work They banded together, under cooperatives, called labor unions today, and as a cooperative they were able to point out unfair practices and get companies and eventually government Interestingly the companies changed before the government changed. Right? And I think the same thing probably needs to happen with data But it’s through this collective action Sir? >> SPEAKER: How are we seeing today adoption of this model to introduce cooperatives? >> SPEAKER: People are — sort of ongoing discussion. Most companies that are not Google or Facebook or Amazon are really interested in this because they’re feeling completely cut out. Right? Citizens, of course, feel cut out

Governments are worried because they have most of the data that they need to provide citizen services is not data that they own So if you look at the sustainable development goals for the UN, most of those goals require having data from private entities like banks Typically, that’s not the situation in the world today Some countries have made laws about this, but they don’t have practice, regular practice, about that That’s one of the battles that’s going on And interestingly, companies are willing to give up their data, enough to make government better. Give it to government to be able to do better management, but not all of the data. They don’t want to give up personal data, just aggregate data, and that’s actually enough for a government. It’s like census data Companies are willing to contribute to the census, a rich census If you go to edu, which is one of our sites, you can see what you can do with it. It’s really surprising Without any individual-level data Okay? So who is next? [Applause] I think we’re going to move to the next speaker I think in this case, I would like to introduce Professor Marco Dorigo Marco is one of the founders of the swarm technology field He directs the IRIDIA lab in Brussels, which is one of the forefront labs for the swarm organization Without further delay? Thank you, Eduardo. Very nice to be here. Last time I was here was many years ago I’m not a frequent visitor I’m not an expert in anything with robotics I will represent first with robotics what we do And then what we do with blockchain to make robots more secure Not working? Next slide So I think you agree with me that the future will be more and more robots There’s already drones In the future there might be nano robots or possibility the first autonomous vehicle in huge use Next slide. Yeah Okay Let me tell you something about swarm robotics. A swarm is a large number of autonomous robots that communicate in a peer-to-peer way through local interactions between and with the environment, and self-organize to solve problems or perform some tasks All these are in absence of centralized control Next slide So we started robotics out to design such systems. There’s some collective behavior There’s local interactions between the robots and between the environment without any centralized controls Next slide Okay. So the problem I am

interested in, in the last 20 years, approximately, is how the control of the swarms so that robots cooperate to form a task. And it’s scalable, so you don’t need to reprogram the swarm when you want more work done or less work done, and the swarm is tolerant to manufacturing or money issues Next slide Thank you Okay So most cases, we use self-organization, no centralized control And this is very good because it’s However, there’s a problem, next slide, so the problem is that our goal is to program the swarm but we can only program the single robots How do we program the single robots so we can get the swarm to do what we want? The way we do this, is by taking an approach where we design and implement behaviors for the robots. For the single robots And then we test the behavior of the swarm in simulation. We repeat this cycle until we’re happy with the results of our swarm, the way the swarm performs And then we move to the test of the robots, and we recycle again until we’re happy with the final result. The reason to take this approach is that we need or our goal is to program the swarm, however, it takes a lot of time, can break the robot, can cause many type of problems, security problems, also So it’s better to get faster and more efficient to work in simulation; however, once you get good results in simulation, most probably they will not carry over to the real robots nor a lot of different reasons, so you need to add the second cycle Next slide There are many definite collective behaviors thats There’s behavior of how they aggregate and form together, so on There are navigation behaviors, and there are collective decision-making behaviors One we’re being focusing on, in our blockchain for robot swarm research, next slide, so I would say that the way that we program our robots taken this behavior-based approach, and most of the time what we do is we program the robots using simple rules, very often inspired by behaviors that you see in insects or other social animals Just to give you a couple examples, we have been working on self-organized search and retrieval where a certain number of robots you see there are self-organizing, searching this base for an object And they manage to take, over time, different roles, in a self-organized way, up to the moment in which one of these chains of robots reaches the object that has to be retrieved and the other robots use the chain to find the object and then to grasp it and retrieve it and return to the location, the blue object on the far right In a similar way, we’ve done experiments with search and retrieve where we have three types of robots that can move in the environment and they search for an object, and self-organize and retrieve it Now, I want to show you briefly a video of these experiments so you can get a better feeling of what we’re doing Swarmanoid is a heterogenous robotic swarm made up of three types of robot The hand bot is designed to manipulate objects The handbot can also climb but needs help from other robots to move around The footbot is a wield robot with a gripper

Using its gripper, a footbot can form physical connections with other footbots or with the handbot An eye bot can fly and explore large areas It can attach to the ceiling and provide environmental information to the other robots In this film, the Swarmanoid is deployed to find and then retrieve a book Here, the Swarmanoid has already partially explored its environment As the eyebots search, successive eyebots attach to the ceiling, forming a connected network Once an eyebot has found the book, the knowledge propagates back to the deployment area The handbot then requests transport assistance from the footbots Using the eyebot network, the footbots form a ground-based chain, linking the deployment area to the book The footbot/handbot entity then follows this ground-based chain A second handbot prepares for transport The first handbot/footbot entity has rotated and aligned with the bookshelf While climbing, the handbot supports its weight with a cord attached to the ceiling The handbot has control over its angle of rotation around the virtual axis Swarmanoid is a parallel distributed system Parallel activity and redundancy increase its robustness and flexibility The second footbot/handbot could retrieve another book or act as a backup should the first one fail. In this film, the Swarmanoid retrieves a single book However, the true Swarmanoid concept would manifest itself in parallel scenarios and unstructured environments Future Swarmanoids might be able to replace human work in hazardous environments, perform search and rescue missions, or even conduct extra planetary exploration This gives you an idea of what we do with the swarms. As you can imagine, everything is organized What you were seeing at the beginning, one of the reasons for robotics research, this is a little bit wishful thinking, besides that it is true in principle, but when robots break down, they create problems for the others They misbehave and create problems for other robots

We need to find ways to increase the autonomy of the system Now, the main subject of my presentation, in the next slide we show you another shot video in which to explain how we do collective decision-making, with the swarm and the robot Then I will move to the blockchain In our research, we study collective decisions in swarms of simple robots. We take inspiration from the house-hunting behavior of honeybee swarms When house-hunting, honeybees choose their new nest location in a self organized manner. The collective choice they make is the result of simple interactions between the swarm networks In our artificial swarms, collective decisions are also the result of self-organized interactions between individuals The Pinabot is a small robot with little capabilities It can move in a straight line or around the center It has only one sensor with which it can measure the brightness of the ambient light It can also exchange messages with neighboring robots and when receiving the message estimate the distance of the center We consider our site selection problem in a swarm of 100 kilobots Robots are initially located in the nest The area where robots exchange site preferences and take individual decisions From the nest, robots can move either to the red or the blue side The goal of the swarm is to find consensus on the best side In our case, the red side The quality of a site is an abstract numeric value We use infrared beacons placed under the arena surface A swarm has made a decision when as a result of a decision-making strategy, a large majority of robots have the same preference We control the kilobots with a machine that implements our decision-making strategy. In the dissemination state, the robot is in the nest, and its primary goal is to promote its current site preference To do so, the robot repeatedly broadcasts its preference. Before moving to the exploration state, the robot corrects to preferences of its neighbors It then applies the majority rule to update this preference, which determines the site it will explore next In the exploration state, the robot travels towards the chosen site Once there, it randomly explores the area in order to estimate the site quality Eventually, the robot returns to the nest and reinterests the dissemination state. In a way similar to honeybees, the effort each robot takes to promote a particular site is proportional to the quality of that site Specifically, a robot promotes its preferred site for a time that is proportional to its current estimation of the site quality This modulation introduces a positive feedback which in time It can overcome the limitations of the individual robot This system is robust to robots that break down, but what happens if some of these robots start to send wrong messages? This is what prompted us to start studying malicious robots in the context of this type of problem That is what Eduardo presented in the first presentation this morning So going back to the problem I’m interested in, as I said, my main interest today is to show our initial research on how to make the robot more tolerant This is work as I say done in collaboration with Ph.D students in our lab and Eduardo

Castello and a postdoc in my lab It’s clear as soon as we have a robot swarm deployed in the real world it will be subject to attacks There will be some guy that wants to create problems So what we are trying to do is to see whether it’s possible to control these swarms using a typical type of computer program that’s in the smart contract so that they’re robust and tolerant and encountering with messages and similar attacks So I think everybody knows what is a cyber attack or creating a fake ID so that one robot can try to take over the swarm by creating many fake ID’s So you know that blockchain basically creates a trusted system and trusted agent, usually computers What we are doing is to use exactly this same approach using robots in place of computers And we are doing this in the So the goal of the study is to first show that it’s possible to write a smart contract and control the decision-making of the robot swarm. This was not done before So it’s important to check it Then to show that the blockchain-based control makes the robot resistance to these Byzantine robots or messages that try similar attack, and then show the blockchain-based approach outperforms other classic approaches So the this is an example that you’ve seen before this morning two talks ago, with Eduardo We have a swarm that has to collectively estimate the frequency of ties on the ground The ground is covered with ties Compared on the example that was shown by Eduardo this morning, this is not the collective decision on which is the most frequent, but it’s a collective estimation which the swarm has to hand out what the percentage of white ties in the environment So the experiments are run in the simulation, but we have already everything in place to run experiments with the robots, which we still have to do that Next slide So the blockchain-based approach that we use works as follows The robots move randomly in the environment. They explore the environment and measure the frequency of tiles that they move over For example, what is their own local estimate of white tiles Then, I believe, 45 seconds, they send their reading as a transaction out to the pool of transactions And to do so, they pay a certain amount of tokens, within the framework Our robots, while moving randomly in the environment, they mine, and usually the first robot to discover the parcel sends to the blockchain and gets the reward So when a block is added to the blockchain, it’s a transaction refined, and this smart contract computes the means of the estimating the transaction

Here, we use a very, very simple detection mechanism The goal is not to find the best possible outlier detection mechanism, but to show that it works Then, robots are actually computing the mean, are paid back a certain amount that is bigger to what they paid, but is paid This is because only the transactions that were actually used to compute the mean, so not the outlier, are paid back Next slide So you understand here that this mechanism is automatically takes care of a similar attack, because everyone wants to send transaction as to pay, but if there’s bad guys, malicious robot, they’re sending transactions that’s outliers, is not paid back, so it does not have money to perform a similar attack So the project, compared with classical approaches for computing estimate of the frequency in a distributed system, by simple constraints, the robots collect their own readings, and estimates of neighbors, for a period of 45 seconds, and then they update their own estimate with the formula that you see there And while moving around randomly, they distribute their estimate to other robots that happen to be in their neighborhood all the time If a Byzantine version of this, it’s basically the same except that the update of the estimate is done without considering the outliers So in our experiment, we use an information matrix that has the difference between the new estimate of the frequency and the estimate of the robots, and the blockchain size What you see in the next few slides is the structure. You have the representation of the experiments And then you have linear consensus, and blockchain results Here, you see on the X axis, the true presentation of white tiles and then the opposite. What we see from these graphs is that all the approaches are similarly good at finding the estimate of the percentage of white tiles Although, the blockchain approach has slightly higher error These results show it’s possible to implement our results, and then the performance is good enough This is the same results where you are able — the actual presentation of white tiles and the estimate The optimal solution is along the dotted line, dashed line So now what happens when we have a Byzantine robots, we see as soon as a number of byzantine robots on the X axis is increasing, the opposite error increases a lot in both byzantine consensus and linear consensus. The curve here is a little bit misleading, but what you have there are the Ox plots plus the presentation of outliers, and the outliers is every time the system was giving basically 75% error, which is the maximum in this particular case The system is estimating On the other side, when you look at the results for blockchain, when the number of byzantine robots increases, it

increases a little bit, but it stays — it remains slow When we come to consensus type, the two classical approaches, basically, there’s no consensus time, because since there are malicious robots, it will always vote for the wrong outcome There will never be consensus On the other time, with blockchain, we can have consensus How do you know the results? You take one of the robots and you read the estimate, right? But you don’t know which robots are byzantine and which are not Basically, you pick a random robot But you don’t know whether it’s the one estimating or one of the malicious Differently, even within the blockchain, even the robots that are malicious, they share the same blockchain as the others So you can read the wrong estimate even when you have random robot that is malicious I already say when a robot tried to perform a similar attack, it will not work with the blockchain The next slide we see in the results — sorry This thing cannot work with the task approaches It can only work — it cannot work with the blockchain, but it works very well with the other approaches You see here the graph, the arrow, it grows very, very fast with the number of byzantine robots for the consensus. It remains quite low with the blockchain approach Next slide Last, as you know, one of the issues with blockchain is that the memory usage grows with time In our experiments, what we’ve done is measured how the growth faster and we found out at least in our experiment, the framework for our experiment, this is very manageable, because the size of a transaction is 148 bytes And even the impact of robots, that we were using, of 16 gigabytes, or approximately, but this is something that, for sure, has to be taken care in future research So in conclusion, what we have shown is that implement robot swarm behavior using smart construct within a blockchain-based context and using the blockchain-based approach first the robot swarm can achieve consensus even in the presence of byzantine robots because the byzantine robots are identified and discarded from the computation of the estimate We We found our warm is resistant to similar attacks Additionally, since all the act is memorized in the blockchain, the behavior of the robots can be audited in the future and analyzed So there are many, I think, many, many problems and open challenges in this line of research First one is that when a robot swarm moves around, at the moment there’s no guarantee that it remains connected all the time If two disconnect from each other, they could grow different blockchain When these two subswarm connect again, one of the two, the shorter one of the two blockchains, will just be loose It’s a lot of work that has been done by part of the robots in the swarm that is somehow lost So maybe we should ensure connectivity all the time Maybe there are other solutions We don’t know yet Another challenge is how to extend

this to more challenging scenarios, as we consider one smart contract, one particular problem, simple problem Can we extend these to problems where there is more than one task? Another issue is that now current robot swarm, all the robots are the same They all have the same computation of power And in the real world, it may be that the swarms are joined by robots of different intelligence, systems, capacities So maybe that is not the best way to go and maybe we should look at alternatives Finally, what should we implement in the blockchain framework and which should we know? For example, in this case, the collective decision, it’s reasonable to think blockchain approach is okay, but there might be a situation where you need a faster response to the activities of the robots, and which is not compatible with the implementation of the blockchain Thank you very much [Applause] When we started research, we wanted to give better support for implementing the smart contract We already starting to look into alternative frameworks, especially for the issues of using proof of work Quick question. I think it was you addressed computational deployments, and comparing the linear approximation approach to the blockchain approach, has there been any comparative study of total computational requirement and skill? Of the entire system? They’re very similar. The problem with blockchain is that you need more communication Hi. Just to confirm the robot swarms are not the ones contributing to the consensus process Rather, they’re just clients, putting their data on to the Etherium blockchain and there’s nodes on the Etherium blockchain I’m next sure I understand. The way it works is each robot is running is one node in the blockchain framework It’s running Etherium or not Each of the robots are also full nodes? Yes. They’re mining all the time. Profession but are they part of a private network? Yeah, yeah, it’s private It’s not the main Etherium [Applause] So yeah Let’s introduce the next speaker. Thomas Hardjono. He’s one of the first guys I encountered here at MIT. He helped me draft some of the initial papers in this field so I’m really grateful to him. I think he’s going to talk about identity, about data, how to manage these kinds of things Without further delay, let’s welcome the speaker [Applause] Thank you, Eddie, and Professor Marco, and we have some famous names here. I met Aleksandr earlier And flying into beautiful snowy Boston to witness the snow It’s good. I recognize some of you guys from last year. We had the same conference here Hopefully, this will become kind of an annual thing. It’s good to see what people are doing in other fields We tend to, like, focus on our own fields and we forget about everything else. Occasionally it’s fun to see something like this

Compared to your videos, my life is boring [Laughter] It’s just a white board with a lot of maps. Nothing moving Nothing 3D Today I’m going to talk about rethinking computing, particularly in the context of IOT devices, blockchains, so on My naive view of robots is it’s an IoT device that has intelligence. It’s not just a sensor that does one thing. It has capabilities and, therefore, it’s good and it could be dangerous. Right? If you abuse a whole warm of robots, you can get into serious trouble, and cause a lot of harm. You know? Thinking airports, roads, and so on So blockchain technology is still nascent, and if somebody tells you it’s mature and ready to go to production, and please buy my coins or tokens, run for the hills If you were interested, I was reading this recent BBC article about One Coin or Coin One, one of these Ponzi schemes The person who started it is now missing and wanted by several authorities So virtues of blockchains, what got people interested in Bitcoin in particular, you would have these distinct nodes, a physical device in your basement, in order for all the devices to reach a particular state, shared state on the ledger, it would need to run some agreement, consensus protocols. This is proof of work. And they need to be able to do this independently of each other There must not be interdependence within nodes within a blockchain system, by definition The fact that each of the nodes carry a complete set of transaction transactions And actually, well-defined limited operations One reason Bitcoin is successful it has such a limited upcoding You need three operations in Bitcoin and you’re done. It just runs But in Etherium, it’s a fully-fledged programming language. You can do interesting stuff. You can do damage, like the Dow attack and so on So when you look at swarms of robots, how do you know that those things flying in the sky, those things that are running in your house, are healthy? What does healthy mean? So in the context of trusted computing, healthy means the device is running the correct firmware and software and correct hardware and not been tampered with by anybody else and you’d like to have visibility into this world. Literally, could you have a screen where you have yellow, green, red, power for each of your nodes, where red means you’re not sure? What is it running right now? Could you have the ability for the nodes to report, truthfully report, its type Just imagine your drone being a complete PC board, from bias, all the way, everything, it needs to be reportable So one of the key things, I think, is that the idea of a cohesion, the value of a swarm of robots, is dependent on this ability to report the node status, and therefore the health of the entire population This is actually not just true in robots, but just in your boring sort of enterprise network devices Enterprise IT and network guys want to know how healthy each of the nodes are, each of the routers within its domain So could we rethink how we use trusted computing technology to, you know, use the features there to feed into the decision-making based on the policy? So that when you ask for a robot to report on something, in addition to that something, you have the option to ask about its health. Has it been tampered with? Has somebody touched it? Has the firmware been updated? Has somebody tried to update the firmware?

A bit of history for those who are old enough Back in the 1980s there was famous documents came out of the DoD, this one is called the Rainbow book It had a network of trusted computing base It was the totality of protection mechanisms within a network system. The whole point is it needs to be able to implement and carry out a policy that you decide on Now, this is TC segments, 1985 That’s a long time ago This is long before virtualization I think VMware did not exist, if you know VMware Cloud computing was a dream, did not exist So interestingly, some of these definitions are being revisited in the trusted computing group There’s a group of vendors and service providers who have been working in this space more than 20 years, since 1999, trying to address some of these issues So Trust Computing Group people talk about trust and trustless today pretty loosely If you look in a coin desk or some of these media, but it’s not an easy matter In late ’99/2000, the TCG came up with the following definition of trust. Think about it carefully. It needs to perform a well-defined function. Think about the brakes on your car Why do you trust this little thing when you’re driving that it will stop the vehicle? Have you thought about that? You just press it, and it works? Why is that? It works once It works twice It works five times It works a hundred times It works five thousand times It works ten years straight So repeated operations of the same thing, and without deviation, it creates social trust, human trust, in the function The function needs to be well-defined Another one is your door handle It’s a very well-defined function If it does something else, if it does 360, you usually panic Like, what? What is this? So these are easy examples of how you define trust The second property, and think about it, when the TCG defined this, it was actually thinking of a chip set, and what features it needed to have It needs to operate unhindered and shielded So your car brakes need to work unhindered. You can’t have a piece of carpet sticking out You guys know about what happened with the Toyota case? The Toyota, Honda, big lawsuit, because it was bad design It means crypto graphic identity Imagine you have a chip on your machines or laptops. It needs to be distinguishable from one another, so when a chip signs something, you know it’s coming from that laptop versus the other PC over there The fourth one is a difficult one We kind of call it TCB dynamism Why am I talking about all of this stuff? Could these features also be inherent in robots? Could you have robots where this is just built in? You can use it or not use it? Imagine if your robots are bored, and it comes with one of these TPM chips, about two dollars each now, and it has all this capability, how would you use it to secure that particular robot and how can you build up that layers of trust? And how can you use the feature of all the robots as a swarm to a particular goal or to do a particular function? One of the things we’re looking at is extending these four properties and adding two more properties. Group membership When a robot, when we think of a robot as belonging to a group or a swarm, how does it prove that? I belong to Group A and not Group B. Right? We don’t know how to do that today. But with this model, you could If a robot wants to join a swarm and leave a swarm, it needs do

get permission to do that. It doesn’t get to get permission from all the members of the group Truthful at the station Imagine you guys know what SGX? You guys are in hardware have probably heard of it This is the trusted computation, the enclave, secure enclave What if each of your robots actually contains a secure enclave? Meaning you can do secure computation and do everything, all the approval work, hash computations, so on, within a trusted compartment, within the chip set So the robot could do that, you could ask it to report truthfully, not just the firmware, but also its memory status. What’s in the memory and who put it there? So diagrammatically, it looks like this. You have a bunch of nodes and robots out there forming a swarm Collectively, it makes available these two additional functions as a collective so that you can have applications and make use of it The applications need not be aware of DP1 or DP4, and in fact, doesn’t need to understand DP1 or DP2 Think about robots that have to carry a mission out in the field, like military robots So the same problem, you know, it was already discussed 20 years ago. Imagine you have troops out in the field in a war situation, carrying back packs. There’s always one guy carrying the radio. How do you update the firmware of one of those boxes in the field? The platons are out there You want to do a firmware update by is at light. This is almost the same problem This is one of these use cases Could you ask nodes to do regular pings to each other, each time reporting its status? It’s not just enough to say, hey, I’m here, I’m signing this ping to do my keys, but who put the keys there? If you don’t know the providence of the keys inside the hardware, the signature is useless When you do consensus, could you incorporate that? Meaning I will accept the proof from a node in that node or robot accompanies it with his report of the internal status, memory status, and so on and so on? And so when I want to confirm this, I need to also check the reported status of the robot Governance When you have a group, a collective of robots, owned by an organization, private organization, and there’s governance, so typically in the BC world when you buy a BC it comes with firmware, hardware, software, and so on, coming from different vendors When, let’s say, I have a BC and it has whatever bias version, 6 point something, that list of components is called a reference manifest Could each robot be given a reference manifest as defined when the robot left the factory and when it was getting deployed? How can you make sure nothing has changed in that manifest? Governance here means you can be part of your collective and join, but you have to have the same reference manifest as the rest of the nodes, the rest of the robots That’s kind of the end of my talk It is a bit, I admit, deep technology, a bit complicated If you want to read the paper, it’s on Archive A shorter version is going to be in Frontiers Just take a read If you’re interested, reach out to me. My cards are there Ned Smith is from Intel. We’ve been at this from ’99. This is an old problem. The industry takes a long time to evolve When I say blockchain is in a nascent state, believe me, it is, compared to the hardware and software that needs to be used

for the nodes and the blockchain That’s kind of it. Any questions? [Applause] who is next? Did I go too fast? You’re going to ask a question? Thomas, I have one question I think it’s very interesting that, like, you represent, like, the world of robotics and especially from your end. Do you see, for example, any applications in which you see these are two worlds combined, right? Do you see this happening more into household robotics? Like with nests, or with Alexa? Or do you see this happening in more a city level? Self-driving cars? Public infrastructure? All these things? What are your visions? Definitely, yes, particularly for high-value assets There’s a whole discussion about industrial IoT, what people are going to use for the next generation nuclear reactors,s. Those sensors need to report correctly It needs to have these features and be able to measure environment report inhinderred I think going forward there will be a lot of application for this technology for what I call static IoT’s all the way to various robots There is a group in the TCG working on this It’s called the cyber resilience group. How do you create a future infrastructure that has resilience against all these possible attacks? Definitely It’s also price It’s kind of interesting Router vendors, I mean the big router vendor in San Jose California I shall not name, they consider a two dollar chip is expensive This is national infrastructure and you think two dollars is expensive? For a five thousand dollar box? So there needs to be a change in mindset Also, in understanding the value of the infrastructure and also develop data that flows through infrastructure Thank you very much You didn’t touch at all on the work on blockchain-based identity systems, the standards work being done by the worldwide web consortium and four or five sort of related groups. It seems like there would be, from the blockchain sense, a great deal of overlap between what you’re talking about in the IoT world and that work. How far along is that? What happens next? Sure, absolutely, there is a connection IEEE has a standard called Device ID It’s an 80211AR specification which is now five or six years old. It’s for device identity Ideally, using our language, Adrian, a device needs to be able to produce an assertion about itself So the question is what keys are being used for those assertions? Is there a key hierarchy and key provenance? This whole identity problem has, again, exploded because people are interested in blockchain and crypto currency and digital assets and virtual assets and they have a key. How do I prove to you this is my public key and I haven’t stolen it from Adrian? There’s a lot of fundamental revisits of problems that people kind of ignored 20 years ago right now, but, yeah, this definitely ties into this whole, you know, blockchain-based identity work that’s happening There’s a number of groups working on this right now Thank you for going through kind of the computing base Are there attacks when you modify that computing base temporarily and then recover? It seems like, long run, it’s not really —

So the chip set I’m talking about, the TPM, actually has registers inside So you can detect if the outer firmware was modified and then put back again, you can detect from the, you know, registers inside a TPM Every time you do something, the reenlister updates. You can run through a machine the history, and what you think should be the correct history, and if there’s a mismatch between the internal registers, you know something is wrong The TPM hardware is temper resistant. It probably takes about a million dollars to scratch off and do physical attacks on that thing. It truly has shielded locations, just like your smart card, sim card, has the same technology But it’s low cost We talk about this every week, several working groups How do you detect something like that, clever manipulation of the firmware? [Applause] I want to introduce the next speaker Professor Fabio Bonsignorio He is the CEO of Heron Robotics And Professor at Scuola Superiore Sant’Anna Let’s welcome him This is just a way, this long lists, is to say I’m involved in a number of research strategies and roadmapping exercises at the European level What I want to do today is to share some points about, actually, and make a kind of rant in favor of so why blockchain matters to robotics Because to this point, it’s still something which is not controversial but maybe not a lot of communities do it First, I give you some context from the point of view of robotics, and I will talk about why peer-to-peer in general is smart. And why it matters in robotics applications I will quickly give you a couple of examples Actually, the first one was already explained or shown in a couple of talks before, but Marco and Eduardo And then I will do some final remarks So the context. We are having some impact on the planet, to a point so we are significantly already creating problems to the survival of the planet, to a point we might be quite close to the tipping point So significant change on our situation In the meantime, this is what actually is going on in robotics If you look at this chart, you see that robotics is a very popular and growing field So it’s 9% is actually — more than 10% But you see when you go to about 80 billion, whether it’s dollars of euros, but if you compare to the product of the planet economy, which is just 1,000 So very good if you’re in robotics, because it’s a growing field. Very nice idea to join the robotics field. But still it’s a niche. It’s a small niche So I think you have seen in many places, in particular, on YouTube, from where robotics can do things like opening a door, sitting on a chair and put your hands on the steering wheel of your chair, or operating a vault. We know from that, this is more or less the state of the so-called mechatronic paradigm

That means where you run a complicated machine server programs, and these programs operate by electrical models. This is what happens So the movie is not starting, but you may — you may have seen — we open the door about one time in twenty So this is more or less the state where we are We do that by employing ten people operating the robot That’s why many people thinks we should go in the direction that Marco was showing for swarm robotics We probably need to look at more deeper inspirations While our robots are all designed top-down, so you have a central brain managing information, devising a plan, implementing the plan, in nature we typically the keyword is emergence and self-organization So we think it should go a bit forward, and actually many people thinks, for example, in this flagship preparation, we are pushing, but we should A few significant technologies, like Internet of Things, machine learning, and deep learning, how long before we get it? And computer, if you don’t have too many objects, also work This is a complete reinventing of older manufacturing processes, which is not a small thing Now, we typically have still mass production So you can order your car, but most cars are built together So you have some options, but you cannot perfectly customize your car You have options, much more maybe than in the past, but you still cannot have your own with preferred side and colors colors, like, in clothes, for your jacket These, they learn a lot, but they lose a lot Actually, the key factor today for a company in service or customer or if you think of the iPhone, a product, typically, you sell a product with a service, it has to be fast and adaptive If you have a stock of blue shirts, and now people want red, all your stocks are valued zero So this capability of reinventing industry in such a way, like a “craft,” like” craft shop “is critical This is free enabling technology There’s all these new kind of robots Not some we know already Also, some interesting application Another interesting thing, which is related, is bottleneck in sustainable production. There are things people have been talking about for a lot of time One is mass customization, what I was quoting before The other is sustainable production We all know if these new billions of customers adopt our Western lifestyle, we don’t have enough resources. Some would say we need two planets. Some say eight planets. Maybe mining the asteroids. But we know that So side effects of these technologies is that we can

recycle materials Before, I give you an example, again, of cars You have every minute a new car coming out of a production line At the ends of life, you should take the same car and dismantle it Today, still today, despite these new technologies, you do it by hand. There’s no competition. So you have no way to — but, if you have vision, and if you have controls, you can disassemble cars. This means that one side effect of these new technologies is that sustainable production, which has been breached for many years, is now possible I go faster This is related to some work we did, which is typically giving two hands and putting — or giving and forcing control Now we enter in why this matters This is from the World economic forum They actually dream of what? Having a satellite cloud network managing the whole logistic networks Now, I think, most of you know, we have a very complex logistic network, supply network, which gives us our products So a car is actually manufactured by several suppliers connected into a network So it is disposed of at the end of life Until this other paradigm, at the end of life, you should have logistics networks and do the same thing And the idea is this A big centralized system, managing logistics And you see some problem here Apart from — so we can really think of it as such a network that’s global. It is not distributed. This leads to some political and financial issues So excessive concentration of humans and power and capital It’s really secure Cannot be tampered There’s a lot of discussion about the fact that now, actually, you don’t need money You need data If you have data, you can make money A friend of mine says we are transitioning to data-ism. You should not focus about money You should focus about holding as many data as you can Okay We may decline may make the same discussion for food supply, and also constructions So can happen ten or fifteen years? Okay. We should have some new technologies I mean application, not just interesting topic for research We will have more computing We should go towards a system where swarms and networks is really the backbone of the system Another thing is that we have transition from browsing, so internet would be mainly used to share information, used to do computation Maybe you don’t anymore rely on car factories, because you will have the car, like, in this case of a 3D-printed bike. Someone close to your house will assemble, downloading from the internet, the designs of your cars On the other hand, we still have someone building chips or building 3D printers You will have someone build your roof,

but maybe a big centralized fusion power facility, such as ITER So what we can do now, in short, is a structured environment, network of connected agents. I have no time to enter into the issue of self-guided cars. Maybe they’re not ready for real autonomy, but if you have a city infrastructure which connects the cars, then you can have it probability already now So we need to adjust the environment This goes to the responses So from my standpoint, a smart city is a huge collection of many robots, many AI, all interacting. You know my story, but everything will be connected A lot of things will be automated by robots and by AI There are already examples of smart cities projects around the world. These projects typically are implemented by choosing a main supplier, and to give the keys of the city, the data of the city, to that supplier, which is the equivalent And so if you think about it, you’re going to robotickize everything Whether or not we reach this level of deep buyer inspiration and solution, but this will allow cars to go around and people and whatever, this means with the current technology, assuming it is scalable enough, we need to move — we need to satellite everything If you think today, Amazon is big, tomorrow it will be ten hundred — one thousand times bigger This is because remember the chart I shown at the beginning Today, robotics is one thousandth of the economy AI is a bit bigger. But we are still there If you think this is going to permeate all the system, all the entities in this space are going to grow by, if not a thousand times, a hundred times, or ten times So we know that there is a solution. Peer-to-peer. Right? So why it matters from robotics? It matters for robotics because it provides, and we have seen an example about that, the distributed and secure transaction It allows this. You see there’s a problem I emphasize it , the aspect of centrallation of power, but do you really think that we can grow at the current level to the level that we can really manage this complexity? Honestly, I’m not so sure it is feasible So there could be a technicality before all the considerations For example, Europe, the European data are going to the U.S. And the blockchain distributed approach to data ownership is seen as a possible solution And distributed computing Here is a point I wanted to make Do we really need a blockchain in robotics? If a single robot, maybe no. I could just keep a connection to a server. It works. It’s stable If I have one to ten, okay, still, so we have seen, Marco was saying before, the amount of computation and communication, especially communication exchange, is

probably still a bit — so you need to balance it with respect to the benefits, because, again, you may have a turning The point is we have millions of interacting agents Can we really think of managing those millions or ten of millions of independent intent agents without a centralized cloud computing infrastructure? Probably not So the point is about service centrality, actually, one of the important topics is network science here So what are people doing with blockchain? There are areas in robotics where you probably don’t need to use blockchain So it will be interesting But it seems we are heading to very big swarm of things Even managing within the companies, it seems we need to start on this problem Big swarms could have different set of rules to follow than small ones So what people are doing? Security We have seen examples about that I think it is interesting to, when someone is doing some experiment about that, to use a platform like this to distribute computation If I don’t want to use a remote cloud, I need to have something, the mechanics, to do the computation It’s slow so far, but if I could make it, it could — then I probably will need a market infrastructure, because if you think about the city of tomorrow, not as — let’s assume that I am lucky and it’s a fabulous city, so there’s a growth provider managing your city, but if you have think you have a different actor, and this is advisable for many reasons, you will need a kind of market because we’re living in a market economy And those markets are very efficient So apart from the fact that there are politics, the first speaker, Adam Smith and Marx many times said similar things, but we need a managed market So the blockchain provides tools for that Then there’s some work we’re doing together with Aleksandr about environmental monitoring I started with the fact that we have seen that we are overloading, a bit stressing, our environment on many respects Typically, you have public agents, taking the same person and telling you about the environment Having transparent third party provide, for example, by distributed ledger to take samples of environment can be — can have value At this point, you can’t trust the results because in all countries citizens are being suspicious about what the public entities are really doing. So you see the final results You cannot You also have the data but you are using a skew I think from a scientific standpoint

it’s very interesting to see how you can match market mechanics with swarm mechanics Swarms are one thing and markets are another thing, but both rely on many agent interaction I was talking about a supply chain If you want an open supply chain, for example, I’ve shown an example of the local 3D-presented bike Now Of course, you need a supply network But now supply networks or supply chains are top-down If you have a big company, for example, Kawasaki, so if tomorrow you wanted do start your bike business, not selling bikes but making and selling bikes customize for your friends, you need a supply network, which is really a network And this requires to somehow abandon this kind of chain set and build them as distributed structure The agent, of course, you can move the same logic to intelligence You can have a registry You need to make — you can manage a birth registry of robot If you think about it, what I think is that we are all people in robots are preaching about the smart cities, but I see huge problems to manage a smart city with current technology If you don’t suggest to me something better, I think that the swarms and the blockchains is the only thing — is the first thing to think about, because really I think the complexity of the smart cities is being a bit underestimated Okay In the future we may have some — now, problem with something we’ve not seen before Okay So for example, one, you have seen already I’m sorry The second example is the idea of building a market We needed to implement something like a market If I want a city where I don’t have an owner of a city, I need to be able to manage the transactions among different operators. You can call it money or not, but you need it to manage data. You know? Remember, data. It’s not money You need to manage an economical tranceaction If there are two actors provided automated parking services, and automated self-driving Uber service, the self-driving Uber may have to pay a toll for parking and will have to interact with the agent managing the parking lot If this system is automated, I will need it automated in market transactions. So I buy the service, and I give you money, or tokens, or whatever But this is an example of it You will need it to implement a contract We know, for example, in a network, you can associate a piece of program that can manage a program So final remarks I focus on this particular thing Even a simple electrical car is made up of hundreds of components The iPhone has less components, but still has a lot of the

components, and still you have hundreds of suppliers involved If I wanted to allow the local supplier of car phone or product, I need to abandon what are the currently hierarchal parameters of supply chain, parallel, and one going to Kawasaki, one going to Duchati, one going to BMW, if you talk about bikes So, for example, typically, when you show this, you use this kind of picture, but this is for human variability We should have hundreds of nodes And relations are much more complex Yeah I see it doesn’t fit well So this is what I already was anticipating If you look at our supply chain today, how it’s managed, they’re managed by typically what is called electronic data interchange, and they use the internet, which is highly distributed, but they are not distributed So this that point is a bit confusing So where is the problem that I see? There’s also an associated problem. We have a confident — for example, I have tried to solve this idea of a blockchain , the supply chain to some company. Right? The problem is that they’re not really used to this mindset of having a distributed supply chain So the main problem that you see is the issue that we are use to working in a different way Apart from the technical issues, we also have this problem I mean the industry is already managing very complex networks with very huge investments and involving a lot of people What we are proposing, it’s changing what we are doing The reason why I wanted to tell something about the markets is that it’s very mature risk kind of change. We already have a system that is working below the possible effectiveness and efficiency level We have this demands which go up and down, but we still have a very rigid supply system So in very, very short, making short of it, we have this new field, if you wish, of network robotics with many, many robots, in the millions, or hundreds of thousands, and above, where you need market. You need data interchange You need distributed ledger And just to finish, yeah, these are, to my knowledge, has been organized by blockchain on the interface, blockchain and robotics. And I can tell you more or less this is the size of the individual map So someone said before blockchain is a massive feel, it’s an em brieotic field, but I think there is a convincing case for looking on this interface So I think — I could bore you by ranting, but actually the summary of topics that I did before is taken from browsing what has been shown at this workshop one and the major things coming out of it So thank you

[Applause] auto I think you mentioned about market being a very efficient way to allocate resources. But there’s no guarantee that the allocation is going to be very just or fair or socially equitable So if you have, let’s say, a blockchain algorithm that is used to manage a market, how can you make sure you don’t have something that is very similar to 51% attack? Let’s use your example of the Uber market. Let’s say you have a market, let’s replace Uber with ambulances. So you have a market for ambulances that is managed by a blockchain algorithm, how can you make sure that, you know, people from the rich district of the city cannot, you know, abuse the system, you know, like use a lot of competition power to create 51% attack to always get the ambulances for themselves? Very good question Of course, I have half an hour This was another interesting aspect Markets are social constructs When we talk about natural markets, typically they are not natural. But they come from a kind of stratification of the habits over time And we know that market created inequalities I would say it’s one of the typical effects of markets Yeah, when we talk about a blockchain for doing something, for example, management of a city, the difference with natural market is that it is an efficient market, so you write the rules While the rules of a real market are customary — so we used to do it, by the way, this is also true of stock exchange, algorithmic rules in an artificial market are algorithmic So a nice topic of research, when I was talking about research on the interface, between, for example, a swarm meaning as big multiagencies, is that you start the rules which lead them to a more effective approach Just to go into a bit politic, not politic, but compare the Soviet Union to the people’s republic of China Typically, the Soviet Union was top-down planning The This was a mistake In China, now they mix planning with markets There’s the problem of managing market in such a way that they benefit all or the most — the biggest number See, I didn’t touch, intentionally, the political implications. But it’s true I think that the advantage of having an artificial market is that you can design it, and you can simulate it But when you have a nice topic of the confidence of existence of a natural market, from the neighboring markets or to the stock exchange — by the way, the stock exchange is, I think, 70% of the transactions are already automated. So it’s not — it’s already something which is similar to an algorithm that you run to see what you have to do I see your point We need to be vigilant about privacy issues, these kind of things, and we also need to be careful about how we design these new markets because it’s true it might end up being a nightmare with a few rich people. This is true with robotics and AI Robotics and AI, I multiply by, say, one hundred times the productivity of one hour of human work This can be used in many ways

I hope I answered your question Yeah? Hi You mentioned earlier about citizen transparent environmental model different from environmental science. Can you speak about that? Yeah, Aleksandr will speak more about it, but if you have a peer reviewed way to check samples, it’s an application of what has been shown from before You can avoid that fake data and enter data into the data chain So actually, I see this big potential in having citizen doing science on data, peer-reviewed by citizens Because you can use these technologies, like any technology, you can use in many ways and you can develop different blockchains with different purposes For example, this is a kind of blockchain to give more power to the people. Power of the data Because as before, if data are the real worth, giving ownership of the data to the people, to the distributed data, it’s a positive thing I don’t want to go too far with this, but you may also think about, for example, electronic vote I’m a big fan of electronic vote because I think it’s easier to move physical data But you know it’s not so difficult to tamper electronic vote This, again, could be an application. I didn’t go into that because I wanted to stay on the topic of robotics And then you have all these strange things But if you have these robots as a service, how you can buy this service? Who can buy the service? And you have the potential for inequality, again, but you also actually — I’m a dreamer of the possibility possibilities So it is not only the citizen collaborating with the data but also the data with the citizen overseeing that the data is not tampered [Applause] the last speaker of the session. Professor Alex Kapitonov He’s a pioneer of this field and from the ITMO University Without further delay that’s start the presentation I’ll bring my little friend with me and just put it here Today, my talk will be about more futuristic scenarios of application for de-centralized technologies Right now, there is several really good stable solutions for information exchange, data storaging, communication, and so on And right now, it can be implemented to the real tasks and it’s really scalable for purpose of smart city and citizens inside smart city Okay. Let’s go further Just some short information about myself I’m an associate professor in St Petersburg and working with robotics things already eight years and mainly focusing on robotics and last four years I’m working with the de-centralized technologies, starting with Etherium blockchain when it was launched, and we mined one of the first thousand block inside that was really amazing for us And after that, we found that smart contracts and other peer

technologies that makes possible execution of the source code in distributed ledger, it’s really useful and it can be used and applicable for robotics stuff There is the main idea, what we found, and we started to develop it Okay And what we found that’s really distributed things, peering communication, is needed not in all spheres Because there is a lot of disadvantages, but advantages, of course, will solve the specific problems Or it can be really useful Of course, the first thing, it’s direction related, with sharing economy Sharing economy right now, there’s a really good way how to maximize the utility of some cars stuff, equipment, and you can see that many companies already going in this direction, for example, in aviation They already selling their engines for their planes not like the solution, but like the service The companies who are buying this plane, they’re paying for the time of the engine, not for the e engine And the same things, I think, it will be with the cars, with the industrial equipment. It’s well on the way. It will come soon everywhere The next thing, it’s crowd sourcing, crowd funding stuff, that’s shown doing powerful things, to solution Different crazy ideas and implemented to the real world And the last thing, but not — it’s really an important thing Public assets management That’s what, right now, is executed by governments, mainly, all the public assets management And sometimes that’s not so easy to make some really fast changes in this sphere It’s a big problem that we should solve and the computing technologies will come there to make it more transparent and immutable and speed up this sphere Okay All those things are really needed As we talked previously, we found such sphere It’s environmental monitoring Environmental monitoring is weak point from the side of transparency and immutability of the data, and participation of the citizens inside this process as related with environmental monitoring is really needed It’s really important that every person should be involved in the process to collect the data, to providing the data, sharing it with participants, but other task is analyzing of this data, interpretation of this data It can be, of course, made with cooperation of the big laboratories or research centers, so on Right now, there’s a project, that’s mainly popular in Europe, when the governance tried to share the liability of data sourcing, crowd sourcing, all the data involved in the environment, like CO2 Dust, the solar energy value, water quality, and so on There’s all those tasks

going forward, and citizens are involved in the process to measure the data about the quality of the water and so on There is, like, a presentation of the main project, what they found, especially for the soil, for the air, for the water, and the common platform for aggregation of this data and showing the interpretation of that information Here is in this process, it’s really needed in peering technologies Because how can you trust the information when you’re, for example, sending — you’re measuring your data. You’re sending it to the server And after that, you are getting the average level of the dust in the city. That’s usually how it works But the average level, it’s kind of a tricky thing For example, during the day, the level of the dust in the air is really high But the night, nobody on the street, of course, it’s not so important In the night, of course, it’s much lower Average level is okay, but we are getting most, the biggest part of the day, where the high area and there’s all the dust in the air This process should be fully transparent for citizens How the data is collected, where it’s stored, and how it’s processed, it should be all those steps should be transparent and clear for every participant of the city, for every citizen And of course, we’re starting from the stationery sensors that can be placed on the house, or the roof, can be mounted on some wheels, buses, but there is sometimes it’s not enough to get the data only from the single point where the sensor is located Sometimes you need to reconstruct some dynamic field of the population of the dust or some gas, and estimate where it’s going, how the flow is working in that area This task , especially for robots, in my opinion, they can be, like, defender of the nature They can connect us with the nature and explain to us what’s happened, how it’s going in the realtime, and this is really an interesting thing, when you can get more data about what’s happened in the environment, how it’s working, and what will happen , if you’re deciding which decision to make I think autonomous systems and robots can connect us with nature and give us the common big picture of this area Here, just a couple demonstrations, how we implement the mobile robotics with peering technologies to collect the data involved in the environment Here, you can see the drone It’s flying out of the city And you can see here the landfill in the top corner

This is the really big gray area , the value of the land fill, how many tons are inside it, it is quite difficult to estimate it But using the mobile robotics and sensors and efforts, we can get the real information about such things For example, on this video, you can see there are lakes around the landfill and it looks like not well It smells like really bad The monitoring of these things can be solved with autonomous systems and robots The next task is water monitoring We made the water draw to collect information involving water quality inside one of the rivers, one of the biggest in Russia We found really interesting scenarios When you’re showing the quality of the water, not in average letter, but you’re searching and estimating the flows of the pollution, you can find the initial point, where it’s polluted, inside the water, and so on The discussion inside society started about, okay, who did it? The companies, of course, are trying to say, “No, that’s not mine That’s not mine.” But the next step somebody will ask, “Okay, please provide me some monitoring system. I want to be clear I want to show that I’m not polluting environment.” This is the next step And how — it’s how it can be in the future when the companies will connect to this system One of the really — the next big thing, what I want to just show you and try to explain, try to imagine that, for example, we have a forest equipped by such system, with a sensor It can measure the CO2 or O2 and other constituents of the air In this case, such system, forest plus a total system, can be like a separate agent, and okay, if it’s separate agent, we can make a deal Okay, forest, for example, lending the land, the landphil, and saying, okay, I’m renting this land and giving you O2, getting the CO2, providing the place for animals, and so on And we can estimate the economic value for such actions And in this scenario, the real nature or part of the nature can be, like, an engine in the economical sphere It’s really, really weird It’s amazing, if only for example, definitely, but I like this idea when I can communicate with the nature and the economical weight Just a short two topics that I want to discuss just a little bit more This is the picture from the publication of the Cornell University. They start a discussion, okay, in the future we will see the problem, the economy of robots, because we were already told they’re a big part of transactions already made automatically Can you imagine the big value of this another mat tick economy and real human economy when we’re exchanging the money with each other, peer-to-peer And there is a problem. The human economy will be really small. Really small

The robotics part will be much more bigger And if we will see some oscillations in the robotics economy, it can be dangerous for human economy, because it’s much more smaller So what the solution, I found, for example, for that scenario, it’s described in we should maximize the communication between each other, and tokenize it. I mean if I’m discussing something with you, I should pay to you If you’re asking something to me, I should pay to you, or you should pay to me, to maximize the value of the human economy. This is some weird things, but it’s one of the ways how to make humans’ economy much bigger than the robotics one Of course, right now, there is not only such aspects, economical aspects discussed with our colleagues and scientific society, but there’s a lot of things, and that’s all described in ethical AI design for robotics researchers for robotics sphere There is a lot of things, starting from the rules of the communication with AI, and finishing some intimate relations between the robots and robots and humans There’s really interesting topic for discussion, and all those things described here One of the ideas in this ethical AI design book, the distributed networks for different AI and robotic solution. Right now, it’s really needed We need to ensure that any actions that can be from the side of the artificial intelligence, and it shouldn’t be in the single point. Single point for decision is dangerous for us Okay I think about the futuristic things, that’s all But just one more moment, I want to show you one demo I hope it will be good — yeah, please, Vitali, can you assist me? This is a demo I just want to show you how it’s possible to apply the simple car, simple robot, can be connected with the peering technologies And here is inside this technologilogical background of this demo, there is a lot of peering things, starting from the peer-to-peer communication, transactions, IPFS, information storaging, and finishing the robot operating system, robot software, that’s also needed to control some aspects of this work Okay. Just let me put it I hope we will get the transaction from this robot But there is a process, what I want to show you At first, here is existing — do you know the interplays or file systems technology? Who knows? Just raise your hand. One-third of the audience. It’s really good Interplay file system makes possible to provide free communication channels, peering channels, for different purposes

And the next thing, interplay interfile system can storage the information inside In the peering distributed network After that, if you have a free layer for communication between the different parts, different robots, you can send — you can broadcast the information about the deals, what should be done, inside this process If you’re finding in the first layer, in the free layer, somebody asks for the work and someone else rated to make it — oh, perfect. It works And somebody rated to make it, that’s — oh Yeah, yeah, yeah, just wheel it out. Not a problem It’s rotated around Somebody, I think — yes The first rows saw how it works The interplay interfile system, you can put information through the blockchain, because that transaction is inside the blockchain. It’s costly And you can just broadcast in blockchain some information It’s really costly But when matched, it can be put in the blockchain The thing about the third part, where a smart contract has been created, and in our case, this smart contract collect the information about rows back for this robot. There is special instruction about what it should do In our case, it should rotate on the time in ten seconds, because the price, what I paid, previously, it was ten tokens And look at at this. Right now, peering technologies is really simple. It’s really simple, applicable It’s really simple to use it Just several months ago, for example, there was a release for the application for the smartphone that you can control the peering communication for your smartphone. It’s amazing That’s why we should use it. We should apply it We should promote it and collect much more information about all aspects of the environment and try to improve our world Thank you [Applause] Any questions? If somebody wants, we can discuss after the demo and I will show the details, if you want If nobody have a question, you — yeah, please So I have two questions The first one is is really love the concept you have about nature as an agent Or the symbiosis with robots, right? Especially because we don’t tend to think of this, but we live in a tractional society Because I’m an agent in society, and I pay my taxes, and they give me things, et cetera, et cetera, mainly because nature is not an agent in this contractual society we tend to not respect that agent If we were to have this interaction, it would be way better. I really love that example The second thing, could you explain a little bit more, what just happened with this robot? How was the flow of information? What happened in the back end? It takes several minutes, but can I

change again for the browser, yes, here is de-centralized applications for this scenario, what I showed you, to send the robot liability for to make them work Using de-centralized application, you’re communicating with IP first layer, interplay interfile system layer, where you are broadcasting the proposal that I’m ready to pay for this work And you’re providing the description of this work, description of this work, according to robot operating systems software in the Rosebeck file When it matches, somebody tells you that I’m ready to make it It matches And sending to the blockchain, if you’re in blockchain, to mine the smart contracts inside the blockchain You can use a contract And I definitely ask you to save this contract, because I hope it will be historical [Laughter] And when the smart contract is put inside the smart chain, it starts executing The Etherium virtual machine matches the addresses of the robots and address of the payors who pay the tokens, and start executing the smart contract, sending the hash of the Rosebeck file, which was put in first to robot Robot uploaded the Rosebeck file. After that, it executed After execution, it’s uploading the final result The Rosebeck, because when it’s rotating, it’s recording the Rosebeck file and putting it back to the IPFS, providing it to the next transaction, to finalize the smart contract This is the flow, what I showed right now Thank you a lot [Applause] So this was the final speaker for this session. Now, we have lunch. Yeah, the other side of this wall. Please enjoy it Yeah. We’ll be back here at 1:00 p.m. with the paper presentations. Don’t miss them Now, enjoy lunch. Thank you [Lunch break]


[Captioner standing by for audio


>> We’re going to be testing this right now in a second

Just getting the audio going

If it’s easier, just do a voice test, so she can hear. Sorry

about that

>> Hey, hey, check, one, two

Check, check, Skype check

Hey, hey, one, two, check Check, check

Great view here

It’s hard to speak in front of this view It’s harder to speak in front of this view. Yeah, it’s great Amazing Okay Thank you so much for your attention My name is Fabio Petrillo We start to work with my two new Ph.D students, especially Marcella did his master in robotics and moved to Canada to work with me in software engineering and robotics, and you say, oh, there is a great symposium about blockchain and robotics. Maybe you can share something. And you prepare this work today It’s preliminary work in progress but you have and insights we would like to share Just to present, Quebec is a huge province in Canada, Society of in the north. It’s a good place to visit. If you have an opportunity to visit. Take a flight. There’s an airport It’s far, but it’s — there’s not just bears there. There’s flies ab also We are trying to organize a team around software, software opening nearing. Soft is my art Software is more art than technology sometimes because it’s hard to understand this phenomena you try to do Together you organize this stuff Finally, we start a discussion, with Marcella, it’s hard to use robots yet. Robotics is everywhere in the industry, but it is cost

efficient. Why is it not everywhere yet? You can say in the discussions, because in general it’s not as safe to share So there is the point that safety is something to improve to put robotics everywhere One discussion, one option, to improve the reliability and the safety, maybe, is blockchain That’s our discussion that we are trying to motivate this work So what’s our goal to try to identify the state of art on blockchain and robotics. Okay? There’s a bit of point of view of the distributed systems and software So our goals we try to identify and classify and evaluate the work on this topic So how you did that, you do a systematic review. Yes, you read a lot And try to find the papers, follow guidelines, traditional guidelines, and systematic review, in software engineering. Okay? You organize research questions and search strategies, blah, blah, blah, that you know probably, and you try to show you some cool research questions, as what are the main challenges that you can discuss, the main approach in blockchain, benefits, and limitations Okay? So what you did, you propose a research stream, search stream, in traditional libraries. Okay? And you compose 89 papers and automatic search on blockchain and robotics That’s the keywords. There is 89 papers And also, we put together the papers manually from the first symposium, to try to collect more papers, and after a filtering the criterias, that is able to answer our research questions, you select 14 papers to analyze So I can show you the catalog of papers on this topic, so you can continue and use this work And one important for me, it’s really, really extremely new topic, is 80, 26, 27, and 90 It’s really, really recent topic. So you are in the trench, you can say There’s not a lot of, but in progress, starting to come And different people are probably here working on it So what’s the opportunity is you start to make mapping and you analyze these challenges So it’s not a surprise, the majority of papers discuss about issues, the challenges, on communication. Okay? How robots work together and how they communicate and several of 14 papers, working on collective decisions and other important challenges, discussed in this topic So communication, but also communication every time and point of security. This is the main point that you can realize And the distributed decision, making algorithms, is something really important, I can say, in this work. Okay? If we try to show you some quotations, really new paper, they usage of blockchain paradigm on embedded systems

for distributed moot agent robotics is too uncommon This is important when I see the announcement for this event, say, wow, it’s great; blockchain and robotics. First time, there’s no meaning, because blockchain is something, for no realtime system, if you can think, but, in fact, people starting to use this technology to improve the systems as robotics However, the limitations of embedded hardware is yet an issue to support this topic, because as you probably know, blockchain is an intense proof of work, so this is an issue that you can work together to use blockchain in the robotics systems. Okay? So also, the discussion of using blockchains to improve on the tech. This is one topic to discuss What’s the main approach or that people are trying to use blockchain technology in robotics? Not surprised is the smart contracts, that you had an example in the first — last presentation here Used smart contracts to organize tasks in blockchain context Okay? But the majority of approaches are smart contracts. That’s the point that people were focused on in blockchain So blockchain approach has a potential enabling to build security and scalable distributed cultural systems device such as a robot in IoT environments. So some people try to focus and say it’s a path There’s a potential to use So the user of blockchains provide a mechanism for entering , durability to storage, and There’s new papers that can share this kind of discussion Here talking about, also, there’s a couple of word discussion about legal and safety regulations, also, in blockchain. Okay? However, you can also discuss some traditional issues in blockchain, and especially in robotics And latence is the most important one, is a traditional one, that becomes probably the most significant — one of the most important issues in literature identify it to use this technology in robotics systems So as a consequence of the communication systems in large swarm of robotics to deploy this network is not a simple task So this is another quotation about communication So just to finish from a couple of discussions that you prepare and to highlight, so regarding analyzing the mobile robotics systems, is a predominant robotics use of blockchain technology. Okay? And the integration of blockchain in robotics systems could be, in fact, a key serious progress in the field of robotics It’s in part trying to observe that, and you are here just to put together this idea, probably Okay? It is probably a huge impact in the economy that you discuss It’s clear in the literature, as confirmed that So just to share some maybe points or recommendations that you can discuss, maybe, to create some metrics parameters to evaluate this practice, especially,

from the point of view of security, okay, methods to compare blockchain and other different methods in distributed systems, okay, and, also, requirements, specific requirements, to put together robotics in blockchain There’s work to do It’s something that you imagine that is in need of progress Okay So that’s it Some kinds of attributes in distributed systems for robotics Also, maybe the kind of, as my research on software engineering, how to put some software engineer aspects around blockchain and robotics. Okay? I really appreciate your attention I This, as I said, is a work in progress. We will work to continually improve, and work how to publish a really good paper, and thank you so much for your attention [Applause] Hello. Hi name is Renita Murimi. I’m an associate professor at the University of Dallas. Today I’m going to talk about my research on a blockchain framework. That is for social robots and the means for analyzing this framework will be a mathematical concept called shift theory I apologize I don’t have anything 3D or that kind of graphics It’s mostly mathematical work and due to time constraints I skipped the mathematical equations as much as I can and just simple visuals to how shift theory really works Before I go into the framework for social robots, I want to speak a little bit about the motivation for this problem So the way I’ll start is a traditional conundrum in decision-making. We all know of decisions we have to make in our lives, personal and professional, trivial and nontrivial One of the biggest hindrances is imperfect information or not having all the information we need. It could relate to games or where to apply to college or a job application, where we simply do not have all the information we need. In the absence of perfect information, or rather in the presence of imperfect information, we adopt an action That action might have, may or may not have, suboptimal outcomes for us. These effects are compounded when we’re working now not as an individual player but as a group. There’s a swarming or collective behavior when a bunch of people like us all make suboptimal decisions based on the imperfect information we all have The second impediment to good decision-making is irrationality There was a study how people did not always do the right thing, especially when talking about outcomes of a game. So if you have the choice between getting ten dollars and fifty dollars, there’s been studies in various domains they found people are consistently choosing the ten dollar outcome. That’s because they simply did not understand the mechanics of the game or rather their perceptions of the utility of the game seemed to be that choosing the ten dollar would be more profitable, when, in fact, it was not So what these researchers did, they turned the traditional model of utility arising from economic games over and they said humans are not always rational. We have here an irrational actor who does not have good information, who is behaving in such a way that the outcomes are suboptimal, and often what plays into our decisions is biases When we do not have good information about a circumstance, we rely on patterns that have worked well for us in the past or have worked well for others. We rely on these biases which are really shortcuts to making good decisions Now, I’ll move on to the current conundrum in decision-making

For the past few minutes I spoke about the traditional conundrum Currently, we have enabler capabilities, we have machine learning and massive data sets, and really good algorithms that parse through those data sets, and these have tilted the scale of rationality in favor of the machine You’re all aware of recent advances where the robotic version of Jeopardy or chess defeated the world champions in those games, when those games were played, and they continue to improve And what has happened is at the heart of this we have a game between two unevenly skilled players On one hand we have a robot, a robotic version of the software, that can perform one thing really well So the robot just plays that one game really well You would not expect it to pick up a block. On the other hand, you have a human who has imperfect information in a range of applications, not perfect information on that particular game, but a range of applications, and does things fairly well on a range of domains. It really, because the scale has tilted in favor of the machine, that is because the human has usually limited recall of the past, is far-sighted in the future, and even the current we are plagued with imperfect information So what works for these robots? Why do they usually have the upper hand in these complex environments? It’s because they do two things well. They reduce the imperfect information, because they can store everything in the cloud, and they have access to these large data sets, which we humans cannot. It’s beyond the range of normal human cognition And they’re also usually not irrational Given appropriate code, it will always choose the best outcome, which cannot be said for human beings all the time. These algorithms are not just gaming versions. We see them coming through data sets of demographic information, voting polls, SAT results, and they’re working well It helps when these algorithms help detect cancer faster, they help us correlate income to majors, and in general they are able to predict the future, let’s say, stock prices, medical tragediry of certain illnesses But the problem comes when we think of these devices or robots as not just our helpers but coequals with us This morning we heard various talks about robots working with humans, becoming coagents, interacting with nature, so we’re moving in that direction This is not anymore in the realm of sci-fi future We already have smartphones that are talking assistants, and they keep the company of our elders They entertain my children, for sure When I was here last year I learned about a drone that has its own social media page It orchestrates pick ups and drop offs. So we’re looking at a future where we have a society that is inhabited coequally with robots and humans in both virtue and physical spaces In such a scenario, it is a problem when the robot always has the upper hand If we’re thinking of a robot has a coequal member of society, it’s hardly comforting to think it knows everything, in all situations. Who would want to play a game with an entity that always wins? Right? Or to socialize with a robot that knows the answer to every question? At least speaking for myself, we all have these memories of conversations or quarrels with a friend, a spouse, a significant others, where ten years down the lane something was said and there’s imperfect information about who said what and we just let it slide. That’s not the case with the robot The robot knows what was said and who said it, along with a time stamp So the work we’re proposing for social robots is to make them more social. It’s not just in the form of aesthetics. There’s a lot of fantastic work done around robot skins and appearances,

texture, but the work around human empathy I mean human society is made up of these glues of empathy We’re like-minded, likely able individuals, with common goals, concerns, challenges. So to make the robot more human in that sense Now, I’m not advocating for a robot who behaves with imperfect information and is irrational all the time, and that’s where we bring in the concept of smart contracts So our proposed framework does three things The first is to have these tuneable parameters, both for imperfect information and irrationality For example, a robot that’s talking to a child or playing with a child has a certain level of imperfect information and irrationality, and then when the same robot is used in a different application, it has better information and exhibits a more serious demeanor, if you will And these classes are then embedded in smart contracts Now, smart contracts work well in the blockchain, and that’s not the only reason we pulled the blockchain in here It’s also because of the inner properties. It’s distributed, it’s trusted, and it’s immutable Now, for analysis of all this, a common concern that was raised in this morning’s talks was that the blockchain is limited in how much it can do We have all this data that’s coming up and analysis of this data sets is hard with attributes that maybe have a hundred different features It’s easy to represent an interaction or transaction between entities up to three-dimensional space. That’s good for us humans With algorithms we can go up a few more dimensions But imagine a transaction — I use the word transaction to record any interaction So if there is a transaction between a robot and a child where we’re trying to decipher sentiment, emotion, facial expressions, the content of the text being spoken, all these features quickly add up. If I were to represent this as individual points, it looks like a very dense point cloud. What happens in dense point clouds is you miss important features You miss components such as tunnels and voids in the graph You miss features about clusters and anomalies, because it’s just a very dense point cloud and analysis is difficult That’s where a fairly new mathematical concept developed around the turn of the 19th Century, called shift theory, comes in Shift theory actually has its root in algebraic topology. So this is the contribution of this work So social robots analyzed using shift theory, made more or less social, based on certain tuneable parameters stored in smart contracts This is the premise On the left, you have a line, an edge, which is one simplex. A simplex is a building block Then he have a two simplex, a triangle A three simplex is the tetrahedron This is the extent of all the math that I will put up on the slides. There’s more in the paper and I’m free to talk with anyone interested about the details of this But algebraic topology really looks at the shape of spaces and it does well is it understands patterns and the structure of spaces, which is excellent when it comes to things like graphs By visualizing the blockchain as a structure like this it becomes easier for us to find our patterns faster More and more applications that envision to blockchain, there are some really nice visuals in the talks this morning, but one example that I put up is that of fraud detection Fraud detection usually depends on a range of characteristics Patterns in location, buying behavior, the amount of the transaction, and envisioning this as a blockchain of sheeves makes it easier A sheev is a gluing of features that has certain local and

global properties. And that’s where the faster computation comes in So more motivation for using algebraic topology, last year I proposed think framework for storing social network data on the blockchain That also is massive amount of information Lots of lines about how many tweets, posts, shares, that are generated everyday. Storing this information and studying this with traditional graph theoretic tools becomes cumbersome The default now is to just collect the data and store it and analysis comes at a later time Why use shift theory? There are some tools like principal component analysis that can help us study the structure of these interactions But those are limited They do not work well in manifolds or curved spaces, but shift theory does really well on end dimensions, which is harder for us humans to visualize, but it’s very forth coming and very powerful using sheaths Blockchain transactions, there’s one other paper I’m aware of that studies blockchain using shift theory. There they do work concerning distributed consensus protocols. Regardless, there are other work, purely from the field of mathematics, that looks at sheaths They study things like eccentricity. They’re studying how well a node is integrated into the graph or how isolated it is. How does that translate for us? It helps us find anomalies and outliers and clusters and all sorts of interesting things about transactions Transactions I use the term loosely for interactions Smart contracts, usually smart contracts have underlying clauses, just like contracts in a non-blockchain world Where upon the fulfillment of certain clauses, an action takes place So in terms of a graph, again, depicted this way, that when those conditions are executed, another edge is formed These smart contracts are the potential to balloon very fast because all these smart contracts lead to more actions which probably have more smart contracts embedded in them Quickly, we’re looking at very dense point clouds which are difficult to analyze as in traditional tools. This scenario is further complicated by the use IoT devices This devices, late test statistics say 20 billion, 22 billion, by the end of next year. So if these devices and various virtual agents are on the blockchain, that results in computationally intractable techniques for analyzing the data produced by those devices This is an analogy of sheaths A sheath, think of it as a sheath of, you know, wheat. A sheath is made up of stocks What I have is a stock The power of sheath theory is because it’s so versatile. It’s applied to categories of objects. These objects can be heterogenous These objects are also module and the categories of modular You can have objects of various elements Right now, I have a stock of blocks and the common spine can be depicted as the spine inside that stock Each of those blocks are analogous to the seeds or the germ inside every stock If you put together a bunch of these stocks, and bind them with the smart contract, then we have a sheath This is an example On the left, I have a single block, depicted as a shock. On the right, I have an entire blockchain depicted as a stock So immediately, you see the versatility It’s not limited to a single kind of object to be linked within a stock And the individual blue circles, they represent nodes Each of these nodes can be transactions. Those green circles are features. The features record about a transaction. These could be sensors that are monitoring

various data points And then whatever is the common linking point, usually, is the hash, that comprises the stock in these diagrams Here are some applications The one on the left is a blockchain stock for logistics and tracking You see now the structure of the sheath where you have multiple stocks bound together These are bound by smart contracts, which are a reflection of imperfect information and irrationality This can be tuned On the right, we have a blockchain for the pursuit of an evasion game where we have maybe a dominant robot and maybe a slightly less dominant robot Then we have swarm robotics behavior, where robots are measuring different things, and they’re all bound by the smart contract The applications as we mentioned in swarm robotics and smart social robotics, also consensus applications, which could be related to building mechanics, and crypto exchanges can be analyzed very well using sheath theory. We have central nodes behaving in different ways. All of this can be captured and analyzed very fast using sheaths and then stored on the blockchain And certain challenges and limitations that are not just unique to our framework They’re the result of bringing together disparate disciplines One is quantum computing. It helps and it hinders Once the technology can be embedded in regular robots, it, of course, leads to a very fast robot and therefore a very slow human. But on the other hand, al of this, if it’s on the blockchain, key management and a big issue And quantum topography can lead to issues in the provenance of the key and how strong it is, how fast it can be broken So that is one challenge that affects such frameworks There is currently work done in post quantum computing. That work is still emerging, that produces ledgers that are resistant to quantum algorithms So that will help this challenge The second is added regulation There’s a concept of distributing computing. So no one really owns who goes first, how the consensus is formed It’s a natural, organic way, based on mathematical constructs But in social robots, and smart contracts, it might be necessary to have some kind of oversight For example, if a robot is deployed to understand the prognosis of a patient, it’s helpful to decipher intent and emotion And these, when not done well, might lead to catastrophic consequences, especially when things are not very black and white for medical conditions Another common challenge is that of diverse blockchain environments There are so many environments used across so many applications It becomes a challenge when we want a seamless way to interact with all our devices. That is another limitation. And information overload Blockchain in itself does not store a lot. Therefore, sheath theory is critical in reducing the amount of information that we gather and storing that on the blockchain So in conclusion, this work brings together smart contracts that are tuneable, that behave differently in different environments, and these smart contracts are created with the soul purpose of making these social robots more social or human, and imperfect information and irrationality will be the tuneable parameters here And the analysis of this is done using sheath theory, which is a computationally efficient tool to analyze large volumes of data Finally, storing all of these on the blockchain, not just because of smart contracts, because of the trusted and distributed nature of the blockchain For future work we’re looking at elements of how AI technologies, specifically, the algorithm, can

help in looking at sheath theoretic implications for analysis and broader work is social analysis through algorithms which we haven’t hashtagged yet but we’re still in the initial status That is it for my talk, and if you have questions [Applause] Good afternoon My name is Jorge Pena. I’m here with my colleague and we’re going to be talking about utilizing blockchain technology for managing collaboration in heterogenous swarms. Before that, just a brief introduction to what our research is about in this area We are working in the intersection between robotics and cloud computing, so IoT domain In this direction, our goal is to design methods that enable truly autonomous reconfigurable robotics swarms. We think about this from the point of view of how you can think of deploying an application in a distributed class set of cloud servers You can think about it as with cloud computing you can take this, without knowing exactly how it is implemented, you can deploy it in a set of servers, to some extent, in a distributed way What if we could do the same with large robotics swarms and heterogenous? What if you could take a set of robotics, and take an application that somehow is coded in a way that is obstructing the capabilities, so doesn’t necessarily know about the specific hardware or specific sensing capabilities of these robotics, and just deploy it? There could be a system that takes into account energy constraints and take into account how to distribute the computation and how to distribute the tasks so that this is not given to the developer of the application But this is a long-term goal Before explaining how or what can be deferred off the blockchain in this idea, I want to say or let me say the specific questions we want to ask in this paper, and the specific problem settings So I have said that we are talking about heterogenous robotics swarms. What does heterogenous mean? For us, heterogenous in this setting first means about variable operational parameters So it means robots that interact with environment in different ways Robots that can operate in the air, drones. Robots on the ground, cars. Or why not robots in the water? Autonomous bots And by robots we mean in general any sort of autonomous agent in this context But it’s also a variable in difference in sensing capabilities Robots that understand their environment or agents that understand their environment in different ways. They can have geometric understanding or isometric understanding or any kind of sensor, unless they have variable processing powers How can we take this into account? We can have an autonomous car with probably almost a super computer need, but drones have heavy payload limitations We cannot put a lot of processers in there. There’s that limit From our point of view, where would this be? This image only shows — We have autonomous delivery of robots, which are smaller We also have delivery drones Or we have automated tracks tracks or logistics industry What we want to answer or what we’re trying to discuss about in this paper is how can we achieve consensus in a collaborative heterogenous more difficult system, but potentially ad hoc We don’t need a priori what are these robots. We mean achieve

consensus in the collaborative effort. We have multiple robots and they want to share data They want to share data to improve their situation And they want to raise their degree of intelligence or level of autonomy. To do this, we have to promote high-quality data. So we have constraints in terms of, for example, bandwidth in the peer-to-peer network What data needs to be shared? Which robot is going to send which data to whom? And when we think about data, it’s not only about the quality of data or a specific characteristics of data, but the amount So what size of data we’re sharing? This is not only in terms of the network limitations, but also in terms of processing limitations of who is receiving the data So how can we take this into account in this collaborative or distributed concepts or system? And we believe maybe not the whole answer but part of the answer might be or can be in blockchain technology So in this paper, we are talking about the idea of modeling to some extent sensing capabilities and processing power of different robots with blockchain or with technology that is part of the blockchain stack And specifically, we are proposing the utilization of proof of work or pictographic proof or algorithms in order to gain power. This may be new in robotics but it has already been used in mining, for example, in the Bitcoin network It can be used also with robots with limited computational power because we don’t full proof of work. We don’t require robots to solve the whole — to find the proper path, we can use proofs of work We can have an estimation of the processing power of the different robots without being aware of what power they have, but just by knowing how much of this proof of work they can solve within a certain time And then in terms of data, we can somehow utilize the blockchain to characteristic data, to try to promote this higher quality data So in this sense, we can leverage the immutability and the data integrity properties of a blockchain in order to have a history of data samples provided by different robots And one thing that I have to say at this point is that as it had been mentioned blockchain in the current state of the art is not ready for lower latency realtime systems and data exchanges. So we are not proposing that all data is shared through the blockchain, but we are proposing that robots have requested to share a sample of that data and store it in the blockchain These samples have to be significant enough so that though robots, either at the same time or later, operating in the same environment, are able to find the same features It could be features, for example, can be a code name of a building or something another robot can find, and this data, which we call data stamps, that are stored in the blockchain, can be utilized and be compared with other robots. This takes time It’s not trivial. It’s initial research. So it’s still ongoing, but it has a lot of potential And this can be used not only to rank the data and choose what is the best source of data, but also characterize it It can be images or semantic data, so forth It’s just a sample from any kind of sensors It can be geometric data, or just points in space From this, all robots have access to the blockchain They can get an idea of who is able to produce what kind of data, what is the quality, for example, you can think about number of pixels in the windows, or in the building, or if it’s geometric how many points are in a certain area, if I’m in a corridor or in a window, as well But this all sounds very nice How to deploy it? Not trivially, it’s not clear

Here, we are putting two main questions, which are from the point of view of deployment are we talking about an open unanimous blockchain, permissionless where everyone can join? We are talking about potentially safety critical situations These are autonomous cars or robots that might be operating in the same place as people So having anonymous agents in this setting might not be ideal However, having permission to blockchain managed by a trusted authority, for example, public infrastructure in a smart city, or mobile network infrastructure, so this kind of blockchain not only allows to manage identities and we have a set of partners that are maintaining, even in the case of network loss, something that’s been recommended this morning, also, what happens if if very few robots are cooperating, and there’s a higher chance of attack and security risks, but we might have a solution if a vast enough of participating in the blockchain The other question is we’re talking about a single blockchain or are they ad hoc blockchains? This is also not clear how to define this In the case of trusted infrastructure, yes, it can be started by these public authorities and maintained by them and we can have a single blockchain, but what to do in the case of ad hoc blockchains? One option that we propose in the paper is this might be used by private parties For example, think of Tesla or any kind of car manufacturing making autonomous cars They could deploy this in their own fleet. It’s generated when there’s a high enough number of cars within a certain area, some distance, or within the same environment One of them could automatically start this blockchain and start collaborating without the drivers even being aware of that. So it could be from a private point of view Just to finish some discussion on conclusion There’s a lot of challenges in this sense We believe there’s also a lot of opportunities, but there’s been a lot of talk today about scalability, about realtime computing, about security, and what happens in these cases of safety critical situations if we have 51% attack or if we have similar attacks Then we also have the problem of identity management and accountability, specifically because we’re talking about autoton mouse robots that are operating with people How do we make them accountable or how do we make sure that we know who did what or what was done wrong? And in general, data integrity, the blockchain in this setting is not used as a crypto currency So the value of the crypto currency is mainly used for transactions, but what would happen, for example, if a set of malicious robots are the only ones operating and providing fake data and validating between themselves? There are strategies or should be strategies to find these outliers, based on, for example, some work presented today, but it’s not clear and we need more work in that direction Just to conclude, we’re proposing blockchain could be a good tool to work consensus into robotics systems. This has been said a lot today, as well. We can use immutable and the integrity of the distributed ledgers to manage and classify the data provided by different robots, but we can also utilize part of this technology to obstruct the different — the different resources in robots we don’t usually — usually a robot developer knows what hardware or software it’s using But if we want to open this robotics swarms to a wider application, we need to be able to extract these resources and generate new ways of interacting and controlling swarms of robots So it’s been said very well this morning, we have to go beyond human robot interaction and

start thinking how to do human swarm interaction and control That’s all. Thank you [Applause] Any question? So good afternoon, colleagues. My name is Vitaly I wanted to see in the morning, in the beginning of our symposium, it was mentioned the need for new business models on the market. That’s what I’m going to talk about today First of all, let me ask you a question How many of you have used an Uber in the past month, by show of hands? Most of you did. The sharing economy became really a common concept in our everyday life in our economy And I think that the same way that digital technologies has led to different relations between the customer and the service provider, the advent of cyberphysical systems paves the way for completely new business model. This business model, it’s not just about fully automated production lines and enterprise I’m talking about universecal access to robotics capabilities for small and medium businesses and even individual use There’s the famous idea of robot as a service model. It’s a service oriented architecture that integrates different types of robotic devices. The whole concept is that people refused to buy a software or hardware directly and instead they use it as a service. This allows to reduce significantly the cost of adoption of robotics Let’s go deeper into robot as a service research This consists of basic services, some user services that can be added The publications that we researched, most scientists actually discussed that in the context of cloud computing So publications about robotics as a service are widely represented by Yeong Chang and his colleagues. They published on the topic and it was one of the first to use the term “robot as a service.” In the first publication they say in robot as a service there should be the same functions as in service oriented architecture, which are robot, robot as a service unit, before which this functionality is described, the service which is the cloud provider, and then the broker is an interactive shell that gives access to those functions in the cloud And then in the next research, those same authors expanded their research and identified a few narrow points for robots as a service platforms They proposed to use standardization and redundancy They state some de-centralized elements are required to improve the reliability of centralized systems And then in their last publication, they presented an architectural scheme for their solution with multiple levels, and we thought the most interesting was the presence of business layer, which normally manages the fees for the services, but also it looks for new business opportunities to improve the whole system in the long run Some other research on the topic were published, which is RoboWeb It was one of the first to use robot operation system. It gives us a lot of opportunities to integrate different types of robot. In this publication, they really focused on robot as a service application for emergency response use cases And the interaction interface they present here is really similar to publisher subscriber system In most academic environments, most research on robots as a service is focused on cloud computing After reviewing these publications our team decided

the feature of robot as a service needs to be revised That’s why we propose new architecture and we did, actually, practical implementation of that Here it is We offered broad de-centrallation and we offered giving them economic autonomy De-centralize allows to reduce the computational burden on the agent management system and a more robust system overall Secondly, even though some components of the economic components were discussed, we think that we need to propose a new way So every device needs to have — needs to be able to create transactions on its own, in order to achieve fully automatic analytic system In our case, we organize the work of robots using market mechanic and robot sends the supply message through the special channel and same does the client, sends the demand message We use interplanetary file system to do that. After there’s a match between supply and demand, an economic transaction is created in a form of smart contract in one of the distributed registers We started working with Etherium a while back, but now also support other ways Basically, we can improve — have the high level of production against malfunction and hacking And in previous publications, some methods of identifying But specific security issues for robot as a service were not raised, and neither were transparency issues The use of distributed registries allows transparency of transactions and allows to take the cloud server to those independent software nodes We have independent software nodes and robots And then finally, there’s a need for standardized protocol on this The standardized communication protocol is robot upgrade system We pass the command to the robot using a Rosebeck file which describes the functionality of the robot This is our vision of how a robot as a service should be organized I’m happy to discuss business models or robotics market further and I’m happy to answer your questions. Thank you for your attention [Applause] Why do you think this is something, compared to the other people that are working on something like that, can you tell us more about that? Exactly. We think of liability as basically the robot that needs to deliver something to the client So basically, whenever a client has requested service, and pays for it, the robot is liable to deliver that. By using those peering technologies we just described, we can give this economic autonomy to the robot so it has its own identity, its own wallet, and it’s able to create contracts. Now we can make sure when the contract is executed, the liability is transferred automatically and we can keep track of that You might have kind of answered it through this previous answer, but if you didn’t use the smart contract, there would be other ways to do it, and why smart contract more advantageous to the other ways? Yeah, well, in software manufacture there are a lot of different ways to do things But I think it’s important to think of robots as independent economic agents and to give them identity of wallet and ability to break contracts. By having the smart contract — a smart contract is essentially a native contract for robots

That’s how I see it. It’s basically an instrument for us to create a contract with a robotic entity Thank you [Applause] So we have a lot of applications of UAV drone systems nowadays From surveillance drones to search and rescue exploration, inspection missions and more But how can we improve these applications? Hello, everyone My name is Mario Santos and I’m going to talk about implementation of blockchain in surveillance problem What’s a surveillance problem? It basically consists in a set of UAV’s, completely autonomous, and must patrol certain points of interest or PoI’s The patrols must be completely unpredictable, and it must be efficient as well Efficient for us means each PoI must be visited as many times as he can. In a perfect system, a PoI would always be visited by a drone, but this is not possible because we have a limited number of drones, but we want to maximize the number of visits But how do we relate to blockchain? The algorithms, they’re developed and the papers are published to solve these kind of problems, but they admit this data is always available The communication is 100% secure instantaneous. They don’t take in account the communication process between the drones They usually are relying on servers and relying on non-distributed products So this is a problem Because they’re not resistant to single-point failure It would be very nice as well to have anonymity and transparency For example, I have a drone and I want to rent my drone to a company who runs a patrolling service I want every transaction to be transparent But how can we connect these elements together? We think blockchain is the best answer So regarding the decision-making algorithm, how to decide how a drone decides which PoI to visit, we have two options We go with a classical way, but for implementing on a smart contract this is too complex and too resource intensive Were talking about drones and guard computers, very small, very limited on power The other solution is game theory It’s great because it’s high efficiency and low complexity It can be easily implemented on smart contracts. The way we thought about this problem was using an agility function Basically, this function takes the input, the position of all the points of interest in the system, the position of every drone in that particular moment, and also the idleness of every PoI, which means the time it took since the last run visit to every PoI By minimizing the function, the drone can decide which PoI to visit next Then there’s a problem How does a system ensure the drone visits a certain PoI? We brought two solutions. A proof of visit or a proof of work The proof of visit is intended for smart devices or electronic devices or devices with computational power, for example, a beacon You can communicate with the drone and see that the drone is there and they both sign the action and publish it to the blockchain. So you can be sure a drone visits a certain PoI But for example imagine I want to navigate my drone around a forest or around a tree This is simple, as well, by using a proof of location. The proof of location can be implemented by proof of location service-based blockchains that are available right now. I’m not going into very deep into this topic, because it’s out of the scope of my presentation, but it’s a topic that is very well known and there are many services that provide this kind of proof of location already

So talking about the smart contracts. Which smart contracts do we need to apply in order to make the system work? First, we need the system manager. It’s not a smart contract. It’s just the entity running the system. It can be a company or an individual or it can be a foundation. It can be open source project They need to write through smart contracts. The first one is a subscription If we have a house, and I want to basically rent a service to patrol my house, I need to pay some tokens, and these tokens are included in the subscription, subscription smart contract Then there’s the decision smart contract. This smart contract is basically the implementation of the algorithm I told you before It’s to make sure the drone computes the correct PoI to visit after But the drones can be, for example, owned by a company or can be collected from individuals We need to make sure that the drones go to the optimal interest. Why? Because in the end, you have a real-world smart contract, basically saying to the drones every time they visit the PoI, and because we are paying tokens to the drones, a malicious drone could just fly around the PoI’s just to collect as much tokens as it could get That’s why it’s important In order to embed everything in the drones, as I told you, we’re embedding these in the small card computers, in the drones itself, the blockchain is running on the drones We want all the communication to the handled by the blockchain and you want every UAV node to run a blockchain node This is how the systems looks like. We have a control box, basically just to handle the flight control log, and we have the navigation The navigation is just to decide, for example, going from this point of interest to that one, the navigation, basically, he handles the path And the decision box actually decides which PoI to visit This is the implementation of the smart contract And then you have all the communication handled by the blockchain In order to select a blockchain that works with this multi-UAV-heterogenous devices and so on, like, for example, in Bitcoin, a regular blockchain, it’s very hard, because it’s very intensive and we’re dealing with very smart cards These cards cannot compute crypto graphic puzzles. They cannot mine blocks as a very large cluster Our approach wassing to in iota It’s a very different approach This graph has many interesting properties One of them is that it doesn’t require miners The way it works, this is attached to two transactions Indirectly, all the other transactions are connected to these two And this is already known crypto currency We can buy Iota tokens right now We can implement both the smart contracts and the payments on the same network This is very interesting and applies very well to this kind of project Another advantage of Iota is the way it handles partitions On a regular blockchain, if we split the blockchain, we cannot reattach it without deleting one of the chains In bit hub, for example, the chain that lasts is the longest But in Iota we can actually do a part anywhere one. For example, when you launch a campaign, we launch in one of these transactions. And we settle every parameter of the campaign Then the nodes which are built into drones, you will only process the transactions related with that campaign, with that mission In the end of the mission, we reattach this tangle to the main end so indirectly every other node will approve the campaign and we don’t actually waste resources of the drones approving transactions In order to see if it’s possible to

run an Iota node in a smaller computer, we ran some tests in our lab. We had an O Droid Looks like kind of Rosebeck It’s very similar. A bit more powerful, but similar in terms of characteristics In the X axis, transactions per second, these were using a software that the Iota foundation provides. It won’t be the final software for running the nodes because it’s still in development. They plan to launch it in April next year, but currently they have a test software that we use in order to have these graphs Basically, in the case of 30 transactions per second, running for a long time, and then we took the average In the end, we obtained this linear behavior, and this we can actually run both control, by the control, and the blockchain in the same card We also ran tests of our algorithm. The game theory algorithm that I showed you before This is an image in our lab We run with three drones On the left side we can see the PoI’s We basically integrate 5 by 5 with 25 points of interest. You can see in blue there are the areas that were recently visited In red the areas are visited — like, they were not visited in a long time And the drones discover, this will validate the drones will cover very efficiently Finally, I would like to thank you very much for your attention. Feel free to ask me any questions you may have [Applause] I’m wondering on the Iota network, given it can split like that and come back together, what actually defines the main net? If you can have lots of these orgs, but you’re saying they can come back together and then be recombined with the main net, how is the mine, main net, filed into these others? Basically, the main net is where the majority of the devices are connected. And it’s supported by the foundation And where the Iota tokens can follow actual functional similar dollars. In this image, it’s just a small partition, but in reality it’s a very, very small partition, huge network exactly. Only after reattaching, we can pay tokens to the drones, and it will value some money, because in the partition itself, it’s called a soup tangle, they won’t value any money because they aren’t connected to the main net I was just curious about your experience in terms of trying to implement the UAV with Iota compared to with Etherium I think that one of the benefits of Iota is it’s supposed to be more light weight and you’ll be able to put the node on to more like a single port computer, because of the light weight of it I also wondered if you’ve tried this with Etherium, as well? Yes, currently, in our lab, we’re doing work with Etherium, and with Iota as well What we notice, it’s very, very — the Iota is very different It can run, the card we’re using is 32 bit card If you tried to run Etherium, the majority of libraries are not compliant Iota is way easier to implement and, for example, in order to run a blockchain like Etherium they would need to somehow decrease the complexity of the proof of work to be able to run in these small card computes, but due to Iota design, and it

doesn’t need any miners, we don’t need to actually trade security for being able to run the network only on the small card computers One more follow-up If you tried running a client node on the UAV instead of using it as a full node? Yeah, we tried. The position we took in the beginning was to try completely independent network of nodes Basically, they run a full node embedded in their software They don’t need any external computer or external servers to connect to the network For example, if you want to drone a forest and have no internet connection, the drones can communicate with each other and in the end we attach the information to the main net so we don’t actually need to have internet connection all the time which makes the system more flexible and operates in basically every scenario You can use a run station In our case, it was just a design decision we decided to go just with basically drones We wanted to be sure that it was possible to implement and we didn’t need to rely on run stations or any other equipment Maybe one of the best ways, like, to conclude, like, these papers, for example, is to give this analysis That, for example, he was just asking about, how Iota could be compared to Etherium, especially because the previous works are based on Etherium, and what are the pros and cons I know Iota has been criticized in the past, right? So it’s good to have, like, this comparison, and if the advantages are way over, these advantages, maybe you can claim that, like, there’s a new way of doing experiments in this field. That would be really nice Thank you for your question, actually. We are currently working on it. When we wrote the paper, we weren’t really sure if we were actually doing that, but we are right now implementing Etherium and we’re waiting for, in order to implement the final version of Iota, but yes we plan to publish in the final paper the final metrics I will not claim Iota is better than Etherium, but in this particular case, we believe Iota is the best option. Thank you [Applause] I wanted to ask just a last question Can you give a brief explanation about the costs of Iota? How much cost the transaction? Actually, there’s no fee associated with every transaction. Though it works — sorry I see these light gray blocks When you want to append any transaction, you connect your transaction to two blocks, at least initially it’s always two, and you’re directly connecting and approving these two transactions. There’s no mining process Basically, you do a small part of the work. There’s no need for external miners When you approve two transactions, then another transaction comes and approves yours, and when the network has a high number of transactions, the number of transactions that are connected to yours will increase, and will increase the trust in your transaction That’s the way it works. That’s why it doesn’t have miners and it’s very scalable. Actually, the performance increases with the number of new transactions and there’s no fee associated with a transaction Thank you [Applause] >> Dear colleagues, my presentation is the last one There are already a talk about the robot painter, and I will explain more details in this presentation what we did, and why the new approach, what we’re promoting today, like entrepreneurs that separate

individual economical agents, it’s really interesting topic, not only from the technical sides, but also for researchers, and the future development and ideas how our world will look like Okay. Just a short outline I will explain all the steps, how we make the work for robot painter, and show the de-centralized applications and the details about the planning, painting, drawing, and so on, and show the several experiments, what we did, and some ideas about how we develop this project in the future Actually, the common work flow right now looks like this We are starting from the auction We are opening the auction on the de-centralized application About who wants to buy this picture And there is a picture, the first picture, is defined It’s hieroglyphs, translated from the hashtags from some social networks, like Twitter, Instagram, and so on After that, we are collecting such several of these auctions’ results, and checking what the price was for every picture, and use it for the future steps Okay During the auction, the robot starts to paint the picture And people can vote by the coins what the price is fair for that picture After finishing the auction, as I showed previously, the smart contacts are also created, and a Rosebeck file put inside, and the owner of this picture put it also in the smart contract When the smart contract is finalized, that means the owner should get his product, his picture How we synthesize this initial source for the picture, for the hieroglyph, would be drawn on the paper The first implementation we did Twitter search with the hashtag Sotoshi. After that, we create the least of the words that’s related with that hashtag in Twitter And choose randomly the K word for that set of the words And after that, translated The next step that’s created in the picture, based on the image of the hieroglyph We convert hieroglyph to image, and the next step we skeletonize it, and find the bounds of this picture, trying to clusterrize it, maximize, with an open CV applications and several additional libraries And after that, we are getting the path, what paths should be for robot, which drawing this picture And when we have that trajectory, when we have the path, we can follow it

using a usual approach, doing manipulator task or so on Here is a search of the hardware and software, that we use for drawing and communicating with a blockchain, what’s connected with that auction And we are sensing what is happening on the picture, that we look at the high and getting the image from the usual camera and trying to get the information, is it okay? Is it correct? Following the trajectory or not? We used the real sense here The next thing, the information from the sensor going to the main software to computer or control for the robot control, and after hieroglyph selection and image processing, it’s starting to use the trajectory execution There’s the usual local desk planner and task execution and there’s — we used Dakuka robot for that There’s a special software they used to connect with their robot operating system. It’s a special interface This interface can give the control to the drivers of this robot Here you can see how it looks like, the interpretation of the hieroglyphs for robot This is the scheme of the auction We are starting from the initial point and painter send the transaction that he started the process of the painting, and you can participate in auction to make a bid for this picture And we are collecting the bids during the one date, 24 hours, and when it’s finished we are choosing the maximum price for this image for this picture, and finalizing the smart contract for that deal, just covering all the information with the smart contract, and the Unit liabilities smart contract, and finalizing This is an interface, web interface, how it looks like Soon we will have more experiments and I hope we will ask you to join to participate in this process to collect more data, because right now it looks like — oh. Sorry Does the video work? No? Okay. Yeah. That’s what Eduardo will start to show And here I want to give just a more explanation about. This is the robot This is Kuka It’s created for soldering, mainly, but we requipped it for painting I think it’s not a bad thing for this robot to be a painter, not a solder Okay And here is the demonstrated how the initial information about the keywords from Twitter collected And after that, we just translate this word to the hieroglyph

and put in the searching here for the bounds of this picture, and after that, when we finish this search we just sending the trajectory for robot and here is the process of the painting, how it looks like During the painting process, participants are making the bids, and this improvized applications, and who will give the maximum bid will get this picture Okay But we, right now, we’re thinking about not just taking the hashtag and draw it We want to add some — right now, we’re on the way to adding some AI elements to this robot and our topic is robotics and AI And here, we start to work with information from Instagram and the process will look like this We are posting to Instagram some picture, and collecting information about who likes this picture. And when we collect the information about who likes it, we’re checking the areas of interest of these persons, and based on the information from the interests area, we are synthesizing the new word, the new data And after we know the area interests for participants, we can — this visualization of such things, there is, like, purple it’s not relevant, not so relevant topics, but the yellow is one of the hottest ones, or the closest topic, what is really — what is important for users who like this picture, who like this, follow this Instagram Using this data and combine it with the data about the price from the previous auctions, we can synthesize different proportions, some new words. Right now, it’s words, but in the future, I hope it will be, like, a full picture Okay. That’s all. Thank you a lot [Applause] Any q? If no, coffee break What is the price? The highest price was — just let me — 0 7 efforts, yeah, for the picture It covers all the expenses for the process, I mean for the drawing process, not for the robot, of course. [laughs] Anymore questions? What I find most interesting about this research is the fact, of course, that you have a new loop You put the robot not only in the labor part but also the capital, but also what I think is very, very interesting is the fact that now you could have, like, some autonomous entities, for example, with only one robot as the employee Right? So as you have, like, entities for example trading in the stock exchange with an algorithm and an entity, now you can have corporations in which the only real employee is a robot. Right? We have seen many cases, in which, especially artists, are trying to think about what happens if you as an investor can put some capital up front in order to set up the system, but then it’s the robot, through doing good and

understanding what are the trends, are getting more benefits out of the action, to get some benefits that could pay back this initial investing effort. Right? Once this initial effort is paid, at some point in time, the robot can buy itself out [Laughter] Right? And it’s interesting You know? What comes from there. People here at MIT are also thinking about this One of them is called Dazza It’s a guy that will come at 4:30 in order to give some ideas about computational law on how this thing could go. But yeah Interesting footnote You have a question, sir? About three weeks ago the U.S patent and trademark office issued in the federal register a list of 13 questions looking for information about how to apply trademark law to intellectual property, like, for instance, your drawings, that are generated by autonomous systems You should all contribute Yeah, yeah, yeah, maybe we should be talking to these people. Yeah. Yeah I think the paper section is over. Now we have some coffee I think the industry session will start at 3:00 p.m So we have 15 minutes. So yeah Let’s get back here in 15 minutes. Thank you [Applause] [Break] [Testing,

Hi, everybody. Welcome back from the coffee break. I like

the coffee break because after

lunch you usually have a glucose down. I know because I teach a

lot of people At 2:00 or 3:00 they start to nod off after lunch. I’m going to be talking about a specific topic called digital assets or digital investment assets and discussing where this whole piece is going and how it’s converging with artificial intelligence and blockchain There’s no robot talk here Maybe at the end I’ll tie it into some robotics, we’ll see So a quick little bit about myself I spent over 20 years in technology. I teach graduate level blockchain, artificial intelligence, at Michigan learning machining learning, at a bunch of different schools in New York, and I run Chain Haus. We’re building some products in space I’m a coder. I spend time at night coding I’ve been doing that for over 20 years. We do in terms of our services, we do education, about blockchain, we do events. I run one of the largest blockchain meetup groups in New York City And we have a de-centralized finance area, and we’re building some products in that space One of them is something I’ll be talking about today which is digital assets So — and we’re looking for cofounders if anybody is interested. Please hit me up These are some of our customers we deal with, people we’ve educated, and companies we’ve taught blockchain and AI stuff to So — and we just wrapped up a project with the World Bank We took Haiti and put mango farmers on the blockchain, and helped design and architect that system, and also involved with a company that came out of MIT that is building — a blockchain accompany building a mortgage platform, and will move to mortgage trading, on a DoT, which is effectively a digital asset I’m also writing a book for O’Riley, which may be outline in January or February. I have about 80% of the book down This is the meetup group, if you’re ever in New York City It’s over 5,000 people, fairly active, and AI and data science group and blockchain The benefit for me, I get a ton of market intel. We learn a lot about what other people are doing. We have a bunch of events coming up We have an event on AI and art in about two weeks. If you happen to be in New York City — I , if you want to present, please let me know I’ll set some up front baseline and then talk about digital investment assets, where it’s heading, and where there may be commercial opportunities for people looking for commercial opportunities First, in the blockchain space, there’s significant amount of noise You have people posting things like this This is the real Rubini, a reading economist in the U.S He This is pinned to his Twitter account. Blockchain is a

failure To some degree, there’s some truth to what he’s saying Maybe there’s a little bit of attention grabbing here. Right? On the flip side, you have people who are advocating a different picture or thinking there’s going to be a different picture Somebody called the Pomp, who believed at one point in time that Bit coin would hit a hundred thousand dollars by the end of this year. Obviously, that has not happened. The secret, he says, though, is that the prices don’t matter. I don’t know what that means. I think that’s all that matters. That’s my view So yeah. So all of this is happening. I as a business and I as a teacher and educator, as a professor and all these kinds of things I do, I try to avoid this stuff and try to find the diamonds in the rough, the business opportunities. And we work with a bunch of different companies and we start to see patterns and trends around where there’s money to be made. So there is a quiet storm where people are adopting blockchain, and where blockchain is moving forward. It’s not very sexy, but it’s occurring. Right? And that happens to be in the enterprises right now So if you look at some of the news articles that have come out, like HSBC just came out with this article they’re going to be tracking $20 billion worth of digital assets and going to be using certain blockchain technologies I want to cover that a little bit It’s not a small amount of money It’s not a large amount of money when it comes to GDP or market size, but it’s a good indicator of where things are headed You see here another effect here, FX transactions, a drop in the bucket of the total FX transactions The Bitcoin market cap it’s about $130 million and they trade $25 billion a day. So this is not a big deal, but this is just the beginning. This is the starting point. The banks are starting to look at this What are they using? What kinds of technology are they using? They’re using blockchain, but it’s not typically the blockchain that you might automatically think So these are some of the chains they’re using. They evoke emotions in people, certain types of reactions Some of it is justify and some may not be justified, but these blockchains are getting out in the market and getting adoption People are paying licensing fees for, or deploying out into the production. They’re systems going into production based on some of these Now, whatever your opinion on Lib bra and things like that, I try not to take an opinion and just look at what the situation is and find what’s factual and not factual around it What is blockchain? I don’t need to do a blockchain class here, but I think it’s important to understand that some of the technical terms around blockchain are not translatable or understandable by businesspeople So you say immutable and consensus, and these kinds of terms, their eyes will start to glaze. What I try to do, because we talk to a lot of executives, they say blockchain is a platform for digital exchange, and you saw double spin and all that kind of stuff, and it’s a place where economic agents have incentives and disincentives are to get involved with it. What those are is a different question And enterprise blockchain is more about mediation rather than dismediation How do you get people to work together rather than use the middleman? Some people get religious on me and get angry, but, you know, it is what it is So there’s a battle between — I was here at the — I spoke at the Bitcoin conference here at MIT six months ago and it was a bit of a ruckus because I head the permission world and permissionless world is almost indifferent. They’re converging If you look at a blockchain or DoT called Quarterra, it’s a “permission chain” but they have

a permissionless version of it open to the public You can get Certain things are baked in This is happening everywhere If you look at Hypoledgeer, it’s donated by the Etherium alliance, it’s an Etherium client designed to be enterprise friendly and they can be both permissioned and permissionless, in both use cases You see this as a trend. We’re going to start off as permission and then expand and become permissionless. Eventually, what does that mean? What does that mean as things start to expand? If you have a large number of participants, in a permission system, it effectively is a public system. Right? You just have to get permissioned into it. Like getting a library card. Right? So then the other question is of centrallation Even in the Etherium world and Bitcoin world, these worlds are still grappling with de-centrallation. A ton of power is concentrated among a small number of blockchain participants. Miners and things like that So decentralization overall, from a purest’s point of view, is still somewhat elusive It’s still somewhat of a pipedream It’s kind of there’s shades of centrallation and decentralization. I personally try to avoid that debate, which condition of happens quite a bit. Especially in academia People take really strong positions And focus on some of the things that really, really matter Here is another paper that refers to the trilemenna of blockchain You can have only two of the three Self-efficiency, rent-free, meaning you don’t pay for transactions, or resource efficient, so there’s no maining costs of things like that. You can only have two of the three, not all three. Which two of those you pick depends on your use case, your business case So permission and permissionless world, the two worlds, are converging, to the point of little indifference. There’s very little that is different between the two You’ll see in five years from now, the public blockchain world will become more permissioned because of regulation and things like that, and the permissioned world will become more open. It’s converging What matters, especially in things we do, does this technology move civilization forward? Does it move a use case forward? That’s the only question we ask. We don’t ask whether it’s permission or permissionless or things like that. Does it move it forward and solve a problem? And we adopt it That’s kind of our view Now, these are great — when you talk about transactions, rates for blockchains, I think most people here are familiar with the transaction rates for Bitcoin and Etherium, relatively slow, and then these other chains, like XRP, somebody talked by Iota today, are significantly on the high side If you look at Visa’s annual report, 124 billion transactions for the year. This is from 2018 They’re working at a different level compared to Bitcoin doing seven transactions per second. You do the math, it’s a drop in the bucket, and not even And their capacity is to be able to do over 600 transactions per second They hit those peaks, like over Christmas and so forth So these public chains have these issues. I call them trade-offs. There’s some things you’re willing to take, some issues you’re willing to take, and certain benefits you’re willing to get. Sometimes you’re trying to trade off for other things. Whether you go for a permission chain or permissionless chain there’s certain trade-offs you make If you decide to make certain trade-offs, will you arrive at the conclusion that you want to build a platform around, let’s say, using a permission chain. You get certain benefits One of the advantages you get, is you can start to build rich and deep types of toke.S or digital assets. That’s what a lot of the enterprises are doing

That’s occurring now. We’re engaged with some of these organizations This is what I really wanted to kind of cover in the next 10-15 minutes So number one, if we assume — and there’s economic theory and research that supports this, if we assume supply can create a market, if you have supply, not demand, but if you have supply, that can induce the existence of a market. If that’s true, right? And if transactions per second correlate with capacity, and capacity is correlated with supply, positively correlated, right? And if the size of a market indicates how much data is emitted by that market, so the larger the market the more data that market would emit, if that’s true, and if we agree that where there’s sufficient data AI comes, if there’s no data, there’s no AI, right? If we agree to these things, right? Then — sorry, my clicker is not working Then we can say the rate of AI adoption is correlated to capacity. Right? Of a market Right? So the more I can do in a market, the more transactions I can do, the deeper the transaction, the faster the transactions, the less fees on the transaction, the more data I will produce, the more likely I can have AI come along and have — do something with that kind of data. Right? Interact with that kind of data This is where the financial world is heading. Right? They see exactly this Say, I can build certain types of assets. I’m going to talk about a little bit about that And then I can start to employ a certain type of AI There’s a breakdown it’s going to go down to like four different steps of how the industry is moving forward Some of those steps already occurred is and some of those steps we’re still in the middle of and some of those steps we can clearly see are coming First, I’ll call it eras or steps, whatever you want to call it, is a proof of concept era We’re kind of past that, especially in the permission DLT space. People say, hey, we’ve done the proof of concept. We see this makes sense. I now want to build something and I want to take it into production Era two is people start toying around with smart contracts and tokens Any blockchain event, it’s pretty much about smart contracts, and tokens Then the era three, which we move into now, in the next two years, is native digital assets and machine learning applied to that And finally, AI I don’t mean AI as some of the stuff we saw today, but AI directly involved in the blockchain, not something external So kind of at quarter one 2019 a lot of enterprises said we did our proof of concepts in 2018 or we’re wrapping up a lot of proof of concepts and now we like what we’re seeing with these DLTs or using or blockchains we’re using and we want to build a team into production At that stage, there was no AI or machine learning needed For you to even say you wanted to bring machine learning into that project would require extreme audit audacity. Like, there was no real data for you to apply We’re now in the phase of adoption of low-hanging use cases I had to put acronyms there These sectors are heavily investing in blockchains now Enterprise blockchain, and use cases and applications We see that happening because we’re getting the calls and emails. Again, I got an email today from a major bank saying, hey, can you come over and talk about DLTs? Important point: At this stage, tokens represent something A token represents an asset or it represents a right or it represents a utility. That’s what we think of who we think of tokens, but that’s not where tokens are going to say A lot of the DLT projects right now are around cost reduction How do I eliminate or reduce reconciliation costs? Which is hundreds of billions of dollars in costs that are hidden in most

businesses because businesses think that’s how you do business. Those reconciliation costs can come out. People are using spreadsheets People are rekeying stuff in, all that kind of stuff At this stage, AI and ML become applicable It’s, hey, this is something we can start to combine. How do I figure out certain things using, and I’ll give examples, using, let’s say, machine learning and token economics The third era we’ll move into, especially the beginning of next year, is native digital assets They’re digital assets that used to represent things but now they are exactly those things. For example, I’m going to give you an example, like a credit default swap, you may have a token that represents a financial instrument, like a credit default swap, but that token is going to go away. The credit default swap itself will be born on the blockchain Right? In fact, I will start to be able to create my own types of financial instruments on a blockchain We call it smart contracts, but it’s beyond smart contracts It’s smart contracts that are tradable themselves. I’m trading those smart contracts Tokens in the Etherium world are just a number in an allocation table. You have the Etherium address and a number associated with that address That’s what a token is But tokens will shift from being from represents an underlying asset or underlying thing, to becoming that underlying thing Right? I will now create a digital asset I can design its economic behavior. This is the economic behavior, and push it out on the blockchain At this stage, AI and ML become a strategic competitive advantage Companies say I’m going to combine the two and anybody can create digital tokens and digital assets but now I need a sustainable advantage, and this is where AI and ML will converge with blockchain, and there will be tons of data at this point all sitting on the blockchain. As much of data is sitting on Etherium, a lot of it is garbage There’s a lot of garbage data on Bitcoin. The fact you can extract that data and mine it is of little use. What are you going to do, predict the price of Etherium? There’s not a whole lot you can do. If you have a very rich trading data, I’m creating all these assets and trading all these types of assets, you have an enormous wealth of data you can mine and do predictive stuff with For example, there’s an organization that creates standards for complex derivatives. If you wanted to issue a credit default swap, you would use ISTA contract, let’s say it’s a template, and you fill out the fields in the template, and, boom, you have a credit default swap. They just announced they’ve created smart contract templates, and then you can start to use this to create your own credit default swap that is totally and natively digital It doesn’t refer to a contract or it’s not a token that refers to a contract, but it’s itself a digital asset So it’s digital first, and maybe not even paper at all And then you see semantic analysis of the legal documentation which is ultimately training AI machine learning models to eventually produce their own legal documents So the ability to do semantic analysis is the first step for these models to be trained, then to reverse out and produce documents themselves This is from Gartner By 2025 we’ll see 176 billion-dollar blockchain market By 2030, that’s 3.1 trillion That’s not from necessarily public blockchains This is from enterprise blockchains and enterprise adoption and the movement of money and movement of value through enterprises, which is significant. Right? And this is going to be the ramp towards, which was mentioned earlier, the digital assets. I can design these assets It could be visual or programmatic I design the asset’s economic behavior, or I let AI do it They run the risk and the model and the Monte Carlo, and then I

click a button called publish and I publish the asset on the chain and it gets traded What happens then, exotics, exotics are these esoteric financial instruments. They’re not the norm. They become the norm Meaning an enormous amount of financial creativity comes into the digital asset creation, says, hey, I want to create an asset that is pegged against the treasury but I want to cap it here and reference that and do this, and I create this really interesting digital investment asset that’s never existed. I can do it rapidly. I can prototype it rapidly. If I like it, I run it through my models If I like it, I publish it, and it’s out on the market fairly quickly And it’s traded Then I can apply machine learning So I can put a letter of credit as a digital asset, a digital investment asset, a letter of credit. I publish the letter of credit I can do predictive analytics on the letter of credit. When do I expect it to be cashed out by the issuer or the beneficiary and all kinds of things I can start to apply machine learning to that That’s probably the next couple years Then we move into this other probably a little scarier world Where AI become economic agents The AI start trading and the AI are doing the risk analysis and even the AI is doing the designing of the assets, based on the data of the counterparty they may have on the blockchain The idea of smart contracts go away, because the term smart contracts is a bit of a misnomer. It actually means nothing. Right? These smart contracts are basically AI agents. They’re plugged into the blockchain. Right? Because blockchains will extend themselves and be able to be invoked by external things, like oracles and things like that, and smart contracts basically became AI agents The blockchain get pushed down into the stack, just basically part of life, not something we think of consciously, like we don’t think about the internet consciously anymore Maybe ten years ago we did. And the need for AI shifts Instead of being a competitive advantage, those who don’t have AI and ML cannot even enter the market Because the incumbents would be significantly stronger than the new entrants And we’ll start to see that And what are the agents trading? Digital investment assets AI creates an asset and it feels like there is a market. It detects there’s a market. It structures the asset. It runs the models on the asset. It publishes the asset Another agent, AI agent, purchases the asset, and it’s traded. How do we know that will happen? Because it happens now There’s Algo and high frequency trading. You have algorithms plugged in. They just don’t happen to be using blockchain because it doesn’t make sense to use blockchain and there’s no blockchain that can support the types of transactions per second, the high-frequency trading and algorithmic trading required. With algorithmic trading the way you do price discovery is flash bids You keep doing that, until somebody nibbles That’s how you find your price That’s not really possible with any blockchain today. Once we get there, maybe in a couple years, blockchains will be mostly amenable to doing that are these enterprise blockchains that can sustain high TPS rates will be the places you can do this high frequency trading and basically the trading is automated It’s an easy step for these AI to start trading and collecting data they can collect directly from the B blockchain So where is this all going? We are going towards a world where AI will create assets and trade assets and these assets would be traded on a high-speed blockchain These digital investment assets, DIA’s, traded on a high-speed blockchain When that happens, maybe five or ten years or fifteen years from now, but that’s the direction we’re headed And the blockchains that support that type of capacity are going to get their faster, are going to get the AI adoption faster, are going to

need or require the AI adoption faster An example of AI trading is this project on GitHub called Genotech. Basically, it creates random algorithms, completely random Takes yesterday’s closing, let’s say, adds two, subtracts five, and each algorithm is associated with a single bot. It bonds a million bots. These bots trade And the bots that do well or trade well create a benchmark and spawn additional bots. The others are killed off And then they’re evaluated and improve on their performance This work has been ongoing for some time, at least for five years. You’ll start to see projects like this get on high-speed blockchains Say I want to be able to trade native digital assets and I want these bots to find my alpha or the profits above the market return rate And then also getting involved in the digital investment assets base or the digital assets base are central banks They use systems called RTGS and LVTS, which are these core systems to reconcile and to net out If they’re also using blockchains, there may be a possibility for AIs to plug into that, as well This is the statement on how the central banks are starting to position. The U.S Fed said they’re not really involved, but I think they are, but other central banks around the globe are starting to look at and were involved in advising a company that’s involved with the central bank, as well In conclusion, DI’s represent a real commercial opportunity for blockchain and AI to converge with real impact. There’s real money to be made My robot tie-in is that they’ll show up at your door. You can’t make the margin call That’s all I have. Thank you I’ll take any questions if anybody has any questions I have the slides up there Yes? [Applause] Thank you. Actually, I have two questions The first one is more of a design question I’m just trying to think about how you piece together blockchain and AI together. Are you thinking more about using, like, a blockchain data to, like, train AI models? Although if you have a lot of, like, data in a blockchain, it may not be very sustainable from the blockchain’s perspective. Or are you using blockchain as some kind of a secure storage for AI, for example, storing model programs or check points of AI onto the blockchain? I’m thinking about how you’re thinking about putting those two together It depends what blockchain you use A pedibot for pedibytes for certain blockchains is not a problem For certain other blockchains, it might be a problem Your point, yeah, this AI would mine that kind of data. It would be an enormous amount of data to mine. And they would build models off of that Doesn’t mean that’s the only source of data they would use, but that’s a big part of that In the academic world, it’s very easy to do — and I know because I’m in the academic world. It’s easy to create models off of clean data In the real world to get clean data is very, very difficult But I think there’s a point I put up there You have crypto graphically ensured consistent data model Right? All of the nodes must agree that’s the data model Therefore, that helps with the data quality Thanks. And my second question is more about the fourth era Where AI agents becomes the norm. I think one of the risks a lot of people have about AI is really deep nerdy models that ends up becoming a blackbox that makes it difficult to debug. I was wondering, given the scenario,

let’s say that you have an AI agent that has developed this very complex structured exotic of derivatives, and then decides to then sell it into another AI agent buyer, if there’s some type of complex negotiation, do you think that this could be a risk that this exotic derivative structure could become something that may not be explained or interpretable? Yet, it’s been conducted by two very deep AI agents? Yeah, I think that’s a very good point. That’s a very fair point I think that is definitely a possibility So today, when you design — like, you have financial engineers that design an instrument. That takes a long time. It can take six or eight months or even longer than that And then you have to go through compliance and actually before that you go through risk models, do Monte Carlos and all kind of stuff And then things can still fall through the cracks, even though relatively speaking those models are simple let’s say compared to an instrument or model the AI produces. There’s all these correlations and I’m going to produce this here now, the digital asset, based on what I see in the market. This produced an alpha for me. Could these slip through the cracks and not catch something and potentially become a financial contagion? Yes. We see that in the market, is one bad algorithm can rip apart a trading floor I have a question You are tokenizing the assets, on smart contracts So smart contracts do create it or use — and if so, if smart contracts are using, like, open, how are you approving for — Era 3 — or era 2 I was referring to tokens, I meant in a very generic way. It could be any kinds of things. The point is you have a digital — if you look at a token, what a token is is a hash map. It’s a hash table. The key is an Etherium address and some arbitrary number. That’s basically what a token is Right? Yeah, my question is because you were saying you were creating, it seems to me, probably I misunderstood, that you’re creating those contracts on the go. In my experience, creating automatically the contracts, nonstandard contracts, you have to have it right, so you have no security issues and stuff Yeah, because you’re coming from the Etherium world. I do a lot of Etherium. That is a problem in the Etherium world Then you have to go to Zeppelin and get audited and all that stuff and they bust you up But in another world you have a lot of tools because you’re using the full strength of a well established programming language. Right? So the ability to create that in the risks on that are lower. There may not be a crypto currency involved, let’s say, Etherium, are not native on the chain, even though that digital value is something that’s traded, but there may not be a crypto currency involved The complexity around creating a smart contract in the Etherium world, it’s not from one-to-one translatable into the DLT world, even though there are complexities there. Your point is correct there are complexities, but it’s not the same as in the Etherium world The Etherium world, one tiny mistake, like some of the best guys can’t catch, and, boom, you’ve locked up ten million dollars worth of ether, right? And boom, you’re done In the DLT world, if somebody does a bad trade, you pick up the phone and call the guy. You say, dude, I know who you are, and that trade was bad, and we’re going to court. Right? So there are other types of circumstances around it Okay. Thank you [Applause] Thank you, Eduardo. By the way, this is my first time in

Boston. I came to New York back in August and I must say it’s a lot colder now, but there’s definitely a traditional Christmas feeling because in Australia it’s summer, it’s hot, so it’s great to be here My name is Emma-Jane I’m part of AEROID technologies, it’s an Australian-based space companies Over the next few years, machines are going to start playing a much bigger and much more complex role in industries and each one of our lives I also want to look into the future and convey an idea of how self-verifying communication protocols modulized swarm robotics and de-centralized autonomous organizations are going to change where we are now Let’s start with what industries have achieved and what we know so far Software is crucial for communication, verification, and connectivity Lately, lots of the research and development has been around applying secure immutable technology layers, like blockchain Most usage of DoT’s in robotics focuses on maintaining a secure log, serving as a ledger, storing event, and validating and publishing information to the network Blockchain is still in its early development And yes, although improvements have been made on individual blockchains, integration into the real world is still in its early prototype stage So talking about applying distributed ledger technologies into robotics, the main problem that is still yet to be solved is having a truly distributed and de-centralized connectivity from the lowest layers so that all the robotic components with verify and interact with each other, removing the probability of single point of failure within individual systems Imagine the main components of a robotic system We have vision, control, power distribution, communication, locomotion, all checked and verified by a secure peer-to-peer communication protocol One of the most challenging aspects of robotics and system automation, as you all know, usually comes down to the senses, because they are never 100% reliable If you tell a robot to move 100 centimeters, most likely it is going to move 90 centimeters or 105 centimeters So we need a better way of validating that the output information is accurate and reliable So why is this so important? Well, often, it involves people’s lives, and hundreds of millions of dollars Let me give you an example Reliance on sense information, without proper verification, caused the crashes of the Boeing 737 Max planes earlier in the year This cost the lives of around 600 people Another example where the risks are higher and the chance of failure is much greater: Outer space In 1999, the Mars Polar Lander went on a mission that took over 5 years of planning and work and cost over $300 million. It crashed into the surface of Mars after a sensor received the wrong information, causing the engines to switch off some 100-200 meters above the planet’s surface

An entire mission failed in a matter of seconds because of one wrong data feed from the sensor I can stand here and give you dozens of examples, but these incidents, these failures, they slow down the progress In fact, often, they slow down the entire industry There’s still a long way to go There has not really been any memorable or inspiring missions since the 1960s when the first astronauts landed on the moon I have always had a passion for space I remember begging my parents for a telescope for my 15th birthday I remember the day I got accepted into the space engineering program at the University of Sidney And I remember writing letters to NASA since I was ten about my ambitions to become an astronaut Since they never replied, I promised myself by the time I turned 21 I would be part of the space industry I am fascinated with the idea of applying emerging technologies into space to make a change and bring about new innovation Technologies like de-centralized verification, or machine-to-machine communications and operations can give a greater level of autonomy This autonomy and its capabilities are vital for isolated outer space operations and missions If we are really thinking about setting up colonies, operations, in the next couple of decades, then we need to start seriously looking at these technologies, especially when thinking about the communication relay issues that come with outer space So to give you a little bit of context here, every one hundred thousand kilometers you go from earth, there’s a latency of 0 74 seconds between the ground and the spacecraft I know this sounds insignificant, just but latency creates bottlenecks For the moon there’s a two-way time delay of just over 2 seconds and for Mars this latency goes to over 4 minutes In the Mars mission, because of this latency, sometimes the rovers could literally only move a couple of meters in an entire day So we can’t scale these missions with the current way that we’re doing things Imagine even having just a few rovers or some small pieces of machinery and equipment It is going to take weeks to months to actually achieve anything useful Providing accountable autonomy, self-verification, and cluster-based communication capabilities is the only way that large-scale operations like natural resource mining and human colonization can take place in outer space At AEROID we’re aiming to develop and apply a communication layer to the machines at the most fundamental layer We are developing software for robotics and autonomous systems Our goal is to create a scalable and compliance-ready software platform for machine-to-machine communications and operations Many space innovations over the past have been focused around hardware For example, Space X with reusable rockets I believe space hardware innovation should go hand-in-hand with software So we are addressing one of the key needs of creating a stable and scalable software communication protocol for outer space missions and operations We have been working for the past six months

We have been looking at light weight and low power consumption blockchain protocols We have been understanding accountability and governance in autonomous operations And also, verification for machine-to-machine communication methodologies According to our findings, some of the software architecture used in space applications today is around 10 years old These systems are highly concentrated They’re highly centralized As you know, this brings in the idea of single point of failures, can limit the data transfer bandwidth, and also leave the system open to lots of vulnerabilities If we talk about sending payloads into space, this costs tens of millions of dollars Right now, the costs of sending around 1 kilogram into space is around $20,000 I know that these costs are reducing with the use of reusable rockets, but they’re still very high So similar to reusable rockets, why wouldn’t we send payloads that are adaptable and reusable? Imagine self-adapting, reusable robotic modules that can connect like LEGO blocks to work together to perform different tasks, achieve different goals, far from earth, without having to rely on any communication from earth At NASA, for each space mission, there is a dedicated operations team By this, I mean they have to allocate specific resources and people, purely for the task of watching over the machines, watching over the spacecraft, the rovers, performing checks and balances, and monitoring the machine health This is hugely expensive It takes up a significant cost of the overall mission budget And also, it’s not feasible, because sometimes these operations teams end up costing more than sending the spacecraft into outer space itself So we really can’t scale We can’t do a lot of things efficiently and properly if we are always relying on the ground Over the next few years, we are going to see a huge transformation in the commercialization of the space industry This will not be from NASA for private companies doing exploratory or discovery missions A huge focus will be on industrial engineering, resource extraction, defense, and tourism So imagine if we could set up mining operations on the moon or an asteroid Imagine multiple machineries, robotics, working together, at a far distance, without having to rely on communication or verification on earth or systems on earth Having secure self-governing machine-to-machine interactions and checks and balances will give a greater level of autonomy for the operations, and also take off a huge load of pressure from these operations teams IOT, robotics, blockchain, machine-to-machine economies, and the space industry are the potential to generate over 8-$12 trillion in economic value by 2025 So something back to AEROID and our machine-to-machine communication technology, we are also looking at incorporating the concepts and principles of DALs, de-centralized autonomous and principles of DAO’s, de-centralized autonomous

organizations They work through smart contracts So this brings in the idea of machines performing checks and balances on each other In terms of validation and verification Think of it a bit like machines policing other machines, but having the communication and the verification operate on a more cluster-basis mechanism. Now you have a hierarchal system in the network Unfortunately, I would love to sort of go into more detail with you of exactly the technology that we’re working on, and the specifics behind it, but we are currently in the process of acquiring two patents, and as you know it’s sort of very touch and go with these things, so I’ve had to restrict a lot of what I really wanted to say, and I guess that is one of the reasons why my team and I are so excited, because our technology doesn’t just apply to the space industry, but theoretically it applies into any machine-to-machine operation, including IoT devices and industries like smart cities, autonomous vehicles, and supply chain logistics Profession going to Boston, I was actually in Dubai. Over there, they’re very future-forward thinking in terms of making their city a lot smarter and more connected They even have a government blockchain initiative and just last week, actually, they announced that blockchain policy strategy approach for 2020 To give you one example, the road transport authority in Dubai are looking at implementing autonomous vehicle infrastructure to that a driverless car in the city, they can pick up, drop off passengers, collect payments, pay for tolls, pay for parking, act as a monetary agent So I know that there are a lot of use cases like this emerging around the world And my team and I do some work in Du bai. So we’re excited to really see these technologies being properly implemented Going back to AEROID and our vision, we really thought about how we can effect change with what we’re working on, how we can show it is something that is really important moving forward, and also inspire and catalyze the future of innovation Particularly space innovation So we made a huge commitment We made a very bold move We started the lunar industrial initiative 2021 The lunar industrial initiative is our plan to send four adaptable robotic modules to the moon in 2021 to demonstrate our proof of concepts of our software communication technology So I’m sure that most of you are wondering how are we going to execute this? Well, there are three major stages Stage one we are applying our software protocol into swarm robotics, swarm technology We currently are working with the University of New South Wales robotics lab in Australia to refine and improve our protocol Stage two we will be testing these adaptable hardware modules, these custom-built rovers, in moon-like environment test chambers So in Australia there is a space testing facility, and this is really good for understanding what changes, what adjustments, need to be made to the hardware and, also, our software protocol, because I can tell you that building robots for earth is very different than building robots for outer space

As part of our initiative, we also want these robots to demonstrate some proof of commercial viability So at stage three, we are looking at incorporating aspects of project wilds into our initiative Project wild, it is headed by Professor Andrew Dempster in Australia, and also part of the Australian center for space engineering and research The main goal of Project Wild is to eventually mine water on the moon Right now, they’re looking at the mining at the moon’s polar craters, particularly the south pole Why water? Well, again, in terms of commercial viability, water is the most versatile natural resource to be found on the moon It won’t only be used to resupply astronauts with oxygen and water, but also power generation and refueling rockets One of the unique things about Project Wild is that they are taking learnings and methodologies from Australian mining sites So instead of taking a space engineering approach, they’re looking at current earth mining The Australian mining industry is very well established, well advanced, particularly in terms of the tools and the extraction techniques they use for extracting and mining natural resources in remote, isolated environments, in the middle of Australia So as part of stage three, we will also be mapping and modeling the specific course and details of the operation We know that mining water is a huge undertaking, so we’re also looking more closely at navigational operations for the swarm robots to perform, because our software technology will give these robots better communication and better verification So send these robots to the room we’ll also partner with a commercial space carrier We know this is a huge feat, so we’re not alone. We’re working with the University of New South Wales. We’re working with members of the University of Sidney, the National Space Society and also aerospace consultants is and a capital venture firm If we pull this off, it will be the first autonomous effort of this kind on the moon This will be amazing. This will be memorable. This will go down in the history books I’m sure that most of you are aware of the huge benefits of using swarm robots, swarm technology, instead of having one specific machine set for one task These benefits become even greater in rough terrain environments like the moon We’re sending one moon where sending one robot for one reason isn’t viable The Apollo missions cost around $600 billion in today’s standards And project Artemis, that is NASA ‘s later lunar mission announced this year, is set to cost at least $30 billion Major if we could perform mining operations on the moon and also solve logistical problems that will inevitably come from sending one piece of hardware or machine to the moon to perform one specific task These are the kind of technologies that are going to contribute to human colonization in outer space So if the problem is high costs, time delays, and single point of failure, then adaptable,

reusable, robotic modules and self-verifying communication protocols is the way to go Providing a de-centralized self-governing infrastructure for machines will give greater accountability, will improve autonomy, and will also enhance the communication capabilities We know how crazy, how risky, our dreams are So whenever anyone says we’re working on something that is too ambitious, too challenging, too hard, I always like to think that someone has to do it Because deep down, I believe that we are all explorers, and we are all curious creatures It has been an amazing journey for me I am learning everyday from the challenges that I have to face, but I love the energy and I love the inspiration My dream is to say that one day that I contributed to the space industry, even if that is just industry It has been a pleasure speaking to you all. Thank you for having me. I know I had to keep it high level. I’d love to speak to some of you more technically afterwards. Please, if you have any questions or feedback, I’d love to hear from you. Thank you [Applause] It’s too crazy of a question It’s more a statement than a question. When you do it, yeah? Maybe you should make an announcement, like say you made the announcement here, that we remember you were here when you made the announcement I just have to say, yeah, you should push like for that direction, definitely The goals are very high, but definitely somebody has to do it. Yeah Let’s have applause for Emma [Applause] Once again, just so that everybody remembers, blockchain is immutable, de-centralized, distributed, and synchronized database. I won’t spend too much time on it There has been in 2019 itself, 2.7 billion spent on blockchain solutions, which is set to grow over the next four years by about 48%. I don’t know why this is not coming here. And a total of 44% blockchain companies which reside outside of the U.S The major investments and the major experimentations happen in the financial services industry, with retail just lags behind by about 5% A few of the statistics which are important in the retail industries is that they are 18% of retail companies who already started working, all in the POC and experimental stage, nothing to scale at the moment And about 9,000 projects every year over the next few years which is set to grow yearly by about 8,000 projects Supply chain in the retail and CBG world has the most number of applications and most number of POC’s to date

A lot of new blockchain applications and use cases are rising. Nowadays, everybody wants to know whether their food is organic or where it’s come from or whether the products they’re buying is creating deforestation or not. These things have a huge application with regards to blockchain and companies are more responsible and corporately focused on sustainability to be able to now do this The next thing which is very interesting is online marketplaces People want to know whether their products are being sold by third-party sellers, who they are, which marketplace changes, which reacts fastest, and who was the first to move These are things which have huge applications, which blockchain can be of use There’s also others which is, you know, payments which I think everybody is aware of. These are the things which are upcoming and a lot of companies are focusing on them Out of all the projects, only 8% make it to finally production and are actually maintained The reason is I will talk about the top three challenges that we currently face in the retail sector with regards to implementing blockchain Yes, even though nobody in this room would confuse it, there are a lot of executives who still confuse it with Bitcoin. There’s a lack of understanding as to what blockchain really is, to be able to actually convince the top management And of course, the fear of missing out. Everybody wants to implement it. Everybody wants to know what it is Everybody wants to do innovations in it, but to be able to scale it, it’s not yet happening Every company has an innovation champion. They would say, yes, they have a pilot running in blockchain. We are going to set aside budget for it in the coming years, but that year has still not come There’s also a lack of standardization because there’s no actual, you know, regulatory body which is in place, and that’s — there’s no network effects to be able to actually scale There are two many different consortiums There’s food trusts, taken into consideration with about ten of the largest food companies There are many others wherein they all form different consortiums To be able to get an actual scalable solution is difficult The main problem for implementing would be scalability In retail, three things matter: Performance, privacy, and ease of use Performance with regard to processing of payments, and I think Jamiel had given you around the transactions so I won’t go into that, where Visa, , where blockchain and Etherium is not up to the mark like with Visa The different nanosecond decisions that go into loading a page with Google with advertisers is really important. To be able to get that all working takes a lot of computing power and energy, which, at this point, we’re not there yet Privacy with respect to the new regulations of the European GDPR, wherein we cannot store consumer data. We have to be very careful about how we store it, how we share it That is something which takes into consideration wherein blockchain is not there Certain companies like Microsoft have started the confidential consortium wherein they actually make Anubis model about privacy and be able to track anonymous users and Bitcoin and other chains. This is something that is between Microsoft, Intel, and a few other companies, wherein they use the trusted environment to actually be able to keep your data private One thing in use of use, Switch, which is a company that came out with an application like a virtual wallet to be able to use gift cards and select different gifts online It basically helps you between transacting between traditional and crypto currency This is not up to scale, but it’s something wherein you sell

direct to consumers And of course, efficiency related to the energy consumption of what we actually use A few internal obstacles that companies face, whether evaluating the cost/benefit analysis or whether convincing decision-makers are some of the things that companies really consider today to be able to see it as an impediment to actually implement blockchain There are some use cases and POC’s which are in production For example, chasability, McDonald’s use it to be able to track what the IOD sends, the transportation of food, so when the temperature decreases when you’re actually transporting chicken, it would be send through IOD sensors to your database and then to be able to actually track that so you know and reduce spoilage. Wal-mart also did something similar with regard to traceability from origin to scale with regard to, you know, deforestation in that aspect Loyalty programs is something that companies are also experimenting with Once again, these are all in the infancy stage Digital identity, this is something that a lot of both financial as well as retailers are looking at with regard to having your own digital identity and to be able to use it to create your facial recognition This is also used in the retail sector where companies like beverage in the retail industry They need to validate with respect to age, et cetera. This comes in handy a lot with regard to that Alibaba in Chinese who is using — I would say China is the leader in analytics They are using — they have actually piloted their online luxury pavilion called T-mall where they track fake goods, which is a huge industry in China, being one of the largest producers of fake goods Alibaba is actually using a lot of their own cloud-based blockchain services to be able to combat that China is also one of the largest countries with respect to filing patents in blockchain. As you can see about 49% come from China This is earlier, a use case which actually happened, in which a large toy manufacturer, they wished to have a blockchain-based digital identity What we did was put their identity on the website, on the phone, and they have a private and public key, and they were able to scan a QR code, once they log into a URL, and then you can actually get through to whatever URL you want to. This is within the enterprise. They didn’t want to go public with it. So it is — they just wanted to test it within the enterprise and then kind of move to their suppliers and customers, et cetera You would think this would be the solution they asked for The project architecture is pretty simple. We used the blockchain service and we also were able to exchange with the tokens to be able to actually access the URL and provide, through identification, your access to the log-in page However, once we presented it, then the challenges came into place where they had more questions, wherein, do I have to maintain two separate systems? Is it cost effective? Can I access all applications? If I do that, do I need blockchain? This is really secure? This is what I was talking about earlier with respect to having a top-down approach and understanding with prioritizing security We had to integrate this with an open source and actually redesign it to be improving or including only the blockchain, which as you see, the small part, but you actually have to do the tunneling, et cetera, and then kind of basically replace your Cisco pane, which everybody who works with it knows what I’m talking about, with regard to that So this would be integration with other applications to actually be able to pilot blockchain There are a lot of technical innovations coming up. There

are a few things which people have talked about Something which is the blockchain has a service wherein companies can call an API. There is hybrid blockchains where you can use public as well as private You can have the transparency of a public blockchain but still have the security of a private blockchain. There’s different blockchains A federated blockchain is where a few people are identified from different companies or different people to be able to actually edit or, you know, validate or give consensus to the data Interoperability. As many people talked about the different types of blockchains available in the market, to be able to transact from one to the other and seamless for the public is something that’s happening and people are working on it to be able to come back and have it happen in the future Recardian contracts, which is the new version of smart contracts For example, it is actually a hand-written contract which are readable and then converted into machine language through blockchain tags. This is something that if, for example, the person is insolvent or the deal does not go through for some reason, this can then be tracked and it would not be executable which definitely is something better than currently what smart contracts have with regards to its limitations And finally, stable coins Stable coins would be the new Bitcoin wherein people are trying to make it a little bit more stable so that it is not — I would say it is not susceptible to price fluctuations There is a lot of work going into this to be able to have one-to-one kind of transparency with regard to the normal fiat money and stable coins currently So the enterprise-wide blockchain shift and outlook for the future, I would say, would be that the consortiums would gain speed There would be a lot more but they would be a lot larger, to be able to come up with a scale application Companies would then be in the POC stage and use case stage and then move on to cross-industry applications based on what we’ve just seen and there would still be a focus on payments, trade, and supply chain finance The key take-aways, I would say, integration with other devices, interoperability, and finally, IOT devices and analytics to scale applications Thanks [Applause] Any questions? I think this session is very interesting because it gives the mindset of the industry, right? In comparison to what’s going on with academia You know very well the systems like the blockchain. Right? And also how people in large companies think. You know? Do you agree with previous speakers that it’s basically it’s not understanding between, like, what these systems can give and how middle managers talk or their benchmarks, like, what is basically hindering the progress of proof of concepts, of new ways to understand and incorporate this into the corporate world? Or do you think it’s because definitely the technology is not mature yet. You know? And there’s too many gaps that we need to do more research on, and we need to be more sure that it’s not going to cause any problem? What do you think? Right. I think specifically within the retail sector, I think it’s a combination of both. So why people want to use blockchain and why we can use it to power many systems, I think it has to be integrated as a holistic solution Say, for example, the traceability would be the main part wherein you can’t really track it from

can really track it, like from palm oil, from the time is tree is cut to where it goes, but that’s one part where it’s very difficult to track, and it’s just one example To have an holistic solution powered by blockchain would make people interested and give them what they want with respect to understanding so they don’t have to understand how it works They just have to know this part of the traceability aspect of blockchain can actually give you a holistic solution and have your origin table to be able to track easily I think it’s a combination which once a lot more applications and a lot of the usability becomes available to consumers and other businesses, this is going to take flight. And really take off That’s how I see it Just a final comment, one of the things you’re saying, my understanding of the whole day, we also have a lack of interfaces We didn’t talk today about this, but all these systems have nice capabilities, like with robots and AI and data analysis and data science, but we don’t know the interface to that same. We don’t know what’s the Google of blockchain, right? And somehow, what you’re answering me, is a little bit like that Yeah, about like the farmer, needs a very, very engineered interface that allows him to do whatever he wants to do. Right? But at the end, he doesn’t need to understand what’s going through the whole thing Right. I mentioned there was one company called Switch which are now creating a user experience wherein like any other online website you can go and transact and use a gift card. So it’s not complicated to use but in the back end you’re using the technology Like any other technology, everybody wants a simple user interface What happens behind, the programming, nobody really cares or wants to understand it That’s how you have to actually look at it Maybe there’s a new field coming, like human-to-blockchain interface or interaction. Thank you. Thank you [Applause] So finally, we are going to close this event with a work shop tutorial about computational load. I’m going to introduce two guys. One is Brian The other is Dazza They belong to the Connection Science group at MIT They are going to tell us about technology and law, which is a big thing we don’t tend to care, but I think it really matters Without further delay, I’m going to let them start Of the two, I’m Dazza This is Brian I run something like called law MIT edu, which has been the wrapper for computational law research here at MIT and we’re housed in the Human Dynamics Lab, Sandy Pentland’s lab We’re about to launch the MIT Commutational law report, but we did some research into AI robotics and law actually goes back a little ways. You’ll hear in our presentation roots back in 2011 when we started modeling autonomous and law, and we’re finally at the launch of this publication where things are coming together and it’s a perfect time to speak with you about it and wanting to hear back from you with your questions as well. With that, Brian? Let’s take it away We have this idea of DAOs plus robots And we have this idea of autonomous legal entities. We have to embed the legal thinking from the get-go so we can optimize the protections of the legal system with these new forms of entities that people are coming up with Because there are going to be all these different questions that start to arise Who owns the entity? Who owns the IP? How are things produced? How are any proceeds divided?

Who is capable of entering into contacts with the DAO? What happens if they do something illegal or goes bankrupt? These are things you don’t really think about when you’re setting something really creative up It’s mostly on the back burner We wanted to start showing why this type of thinking is important to, like, get in front of It’s important to get in front of this type of thinking ahead of time, instead of when everything hits the fan Right. One of the kind of questions embedded here, I think the way we phrased it, is who is capable of entering into contracts for the DAO? On behalf of or behest of the DAO? There’s a deeper question, when, if ever, is a DAO capable of forming contracts for itself? When it could or should be treated as a legal entity itself and therefore capable of forming and enforcing contracts and having them enforced against it So the answer to this kind of, we break it down into a few different pieces, but one of the key pieces that you were touching on there was the idea of legal personalities That you could have a limitation of liability that creates a separate legal entity apart from yourself and gives you something that you can kind of hide behind in the form of liability Then we get to the notion of investment securities and contracts and intellectual property and contracts and agency law and then tort And we have some different use cases that we walk through and kind of talk about how they apply This is the road map for the next 15 minutes, not for 2020 Generally, the container of legal personality rights requires a few of those common ingredients. Registration with the state, identification of certain governance mechanisms. Like with the DAO, it could be the voting mechanisms or things like that. The identification of the individuals who are in charge of administratorring the governance mechanisms. It’s sometimes called costotome And then the legal purpose. You have to operate within some specific legal purpose and you only get these protections if you fit within that container Just to break this down a little, registration with the state, like how many people have ever formed a corporation or LLC? A smattering. There’s a step when somebody goes to the secretary of state’s office, in the United States, and fills out a form. There’s always a check involved. They need their money And they will create the corporation for you. Your name will show up, along with who the directors are, on the registry I think it’s noteworthy we had a law and technology conference here not long ago, one of the presenters was interested in a legal personality that doesn’t require registration with the state It’s called a Massachusetts business trust. It’s just a trust agreement where the parties to the trust sign something and it’s a legal entity. Although, when we explored it, it turned out in order to maintain its existence and pay the taxes and disdissolve it, you ended up having to do the registration with the scope So it seems the first bullet is something that’s a capability we need to have in order to incorporate itself Yeah, another thing to point out, some of these protections are at the state level and some are at the federal level. In the U.S., you have different state protections that are related to business entity, but you’ll have federal protections related to securities law. It’s important to know which domain you’re operating in in order to optimize for that domain We should probably disclose we’re operating within those different levels, we’re very much U.S. centric for this conversation. But we’ve been dabbling and collaborating with people in Europe and other countries, as you’ll see So one of the most progressive states in this record is Vermont. You register with the state, and the state reviews the operating agreement to ensure the safety and access of your permission protocols You have a summary of mission and purpose, like I talked about. Then there’s some indication as to whether this is fully or partially automated Then you specify the voting protocols The way this has played out so far with this organization, they set up their organization, they paid

the fee, and figured out their governance operation De-centralized ledger, is how they described it I wanted to highlight that, but it’s not showing up very well But yeah. And then other states follow approximately the same recipe that I laid out So with Delaware, the way they got to the end result is a little bit different. Instead of having a stand-alone business entity, they allow the use of electronic networks or database s One thing to note, the Delaware corporate law permits the registration of series of entities If you have an entity that’s nested like a thousand times, as these shell entities, you can set that up theoretically in the state of Delaware One way this is playing out is through the LOA, which is a legally compliant DOA for investments And specifically, within that, it would be another instance where you have to go through the IRS’s requirements for securities registration, and I think all the members of this DAO have to be accredited investors It’s an additional protection you get because you’re an investor in the company so you have to meet even this more advanced threshold Should probably say that this is a project from a civic hacking group in Brooklyn We have a bunch of sites, if you want to follow up on any of this And these slides will all be available Everybody can use them and play around with them Wyoming did something similar to Delaware They set up so certificate tokens could be used instead of stock. Wyoming also set up a special purpose depository bank for crypto transactions. Now you have a specific place you can go where you can hold some of these things, using some smart contract framework as a way to, you know, keep assets and keep track of all your crypto assets, essentially One of the ways this is playing out is through LASSO DAO. I believe somehow they’re involved in co-working, as well, along with DORG, but this kind of gets us into the next bit I was talking about earlier, where investments and securities, governed by the IRS. The big question here, is something a utility token or — SEC Oh, SEC, sorry. That was a mistake Does it meet the Howie test? Are you putting money into the entity with the expectation people will do work on your behalf and get you some return? If so, you need to make sure you’re compliant Otherwise, the SEC can come after you There are only one of only two companies in the U.S that have successfully beat a no action letter Who has heard of a no action letter? So you could just go and do something, and believe or hope it’s not a security, and discover later the securities and exchange commission says it is, and then they can launch an enforcement action against you. That’s a bad day On the other hand, you could sort of get proactive with legally structuring things and that’s very much the spirit of how we treat computational law, is designing legal processes, and engineering law and legal processes And one of the mechanisms you could use in this case is called a “no action letter ” You basically going to the regulator, SEC if it’s federal, and there’s also state securities regulators, and explain in detail what you’re planning to do, and ask you if they’ll basically issue a no actions letter, and sometimes they will, and that gives you a safe harbor that they agree in advance there’s nothing wrong with what you’re doing During a bunch of legal hackothons we’d done a couple of years ago,

including revolving loan funds grading and other investment vehicles, through DAOs, it was a legal hack-a-thon on the teams, they structured a no action letter with their code So we’re looking at ways to embed that within the process Here is an example of a company where you’ve got a no action letter from the SEC This is a good practice when you’re creating your robot AI investment funds Exactly The next kind of issue that we can run into, and I think as I touched on something that’s kind of like a theme, when you’re more proactive you have more flexibility. With regard to the intellectual property rights, you can set a lot of this stuff up by contract and a lot of it is set up by contract, among the members, or between the individual and the state, some sort of letter that functions as a proxy of a product between the people and the SEC, for example And intellectual property, it’s especially interesting, because you can start dividing fractional ownership rights using tokens, and then different people can programmatically verify they own part of an entity or part of the intellectual property of an entity, or whatever, however you want to slice it And that really gives people a more granular control over all these things, and it provides new opportunities that people had not had before And so in another context, if you look at contracts and agency, this is a typical example of, you know, what an agency relationship is You have a principal The principal has the agent do some task The agent goes to a third party to effectuate that task And they can either be acting within express authority, inherent authority, or implied authority Basically what that means is I can say, hey, you know, you’re authorized to go to buy, you know, a bunch of watermelons. And you have the authority to do that I can say you have the authority to go buy produce, and so by buying watermelons you would have the inherent ability to buy that because I said you could buy produce But if you buy furniture — So the third party wouldn’t have to, like — the third party wouldn’t have the ability to go after the agent in that situation Just another example is like a house. You have a broker It’s not uncommon to get a broker to sell your house The broker is a kind of — or an agent, real estate agent, you’ll usually have an agreement with them and they can usually do things like solicit offers or maybe get pre approvals on loans some cases. You can structure it more deeply to get a letter of — I’m blanking on the word, but what’s it called? Essentially authorized Power of attorney. POA. I was trying to do “letter.” It’s not letter of attorney. It’s power of attorney. They can actually do the closing documents for you. These are all different degrees of authority that a principal gives an agent to act on their behalf with a third party, and one of the reasons this matters in this context is when you’ve got, let’s say, people that are operating a robot with AI that’s maybe a DAO to do transactions, whether it’s an investment fund or whether it’s, you know, we’ll show you some other interesting use cases like a publishing company or things like that, that entity is going to be interacting with other third parties There’s a whole framework of agency law which looks like this It comes down to what are the — what are each of these parties, what are their rights and responsibilities with respect to the other parties? A big question is what would the third party have known this agent was authorized to show me the house but not sell me the house? To get furniture but not watermelons or both or one and

not the other? So there’s questions there So we’ll talk about how this can be designed into a legal process and have this run smoothly We can do this such that all these things are narrowly scoped and clearly understood so you don’t have the edge cases where things go terribly, terribly wrong. Hopefully Tort is another one We’re seeing more with autonomous vehicles. There are questions about, you know, what happens if the autonomous vehicle decides to hit my car instead of run over the baby? Who is liable in this situation? And what it gets back to is this idea of an accountability gap So basically, taking the analogous situation for if a person was there and then figuring out where the liability would have been apportioned if the same thing happened So if instead of, you know, this was being driven by a person, instead of an automated vehicle, the liability wouldn’t necessarily go to the person, except if these five things happen. Then you can point to those five things and have a little bit more of a protection there Good enough for now. We have so many slides. Let’s keep moving We’re going to get to the fun part. That was all background so we could set up the things that are the most juicy So the use cases and the research history I’ll let you talk about this Oh, yeah, I promised you there was history. Here is some history Has anyone here heard of the firm Robot and Robot and Wang? It’s kind of a joke but it’s also real Wang used to hang around here and he went to get his law degree at UC Berkeley He has this concept of how much of a law firm can you automate? In 2011 we collaborated on a project called Corp Bot where we wanted to create some code that would code to a secretary of state’s office and form a corporation and conduct a single function like sell a book on Amazon, upload a book on Amazon, sell it, get some money, and then dissolve the corporation We made some progress, but then we all went and did some other projects. We never completed that one Also it turns out it’s harder to do that kind of robotic negotiation with more nuanced legal requirements, much harder than we thought I’ve been working on this since 2010, I would say. If we fast forward a little bit, in 2016, thanks to blockchain, we were able to make more progress So one of our collaborators, she wanted to do what she was calling a blockchain border bank, to make it easier to get microloans. We did what we could to model that Turns out that was really hard, especially with the banking and everything. We ended up revolving a loan fund operating under Massachusetts law Or at least I was licensed to practice and I could understand what the forms were and we could model it more and test it all the way through without going to a Dominican republic jail or something This is basically the UML that we came up with in order to figure out how to do the loan application in an automated way, who would issue the loans to, and make sure we have the balance in the fund, and receive payments, and finally provide the acknowledgment the loan was paid off which is a big legal document under Massachusetts law that you want to be able to show We modeled that pretty well Again, it was a little hard to do the test all the way through The best way we could figure out how to do it was use Pay Pal to send and receive money. We thought, give me a break We needed to look further, but we made

a lot more progress in modeling the entire entity Let’s go forward You want to talk to this point? Yeah, this is the fun stuff We’ve been working together on this project related to automated and autonomous legal entities for a while now We cohosted a workshop as it was remotely here and I was in Berlin with a bunch of people We wanted to walk through it’s pretty basic schematic for what all of the actions that would be required in order to create a publishing DAO. If you have a network of people together, and you wanted to produce a book or something, or produce different books, or start hiring people to write for you, what would that look like? And so what we came up with was, okay, so there’s a publishing DAO, the publishing DAO invests, in order to buy this es presso book machine, which is basically a printing press for different books, and you create this smart contract that allows you to pay logistics partners to pick up the books and deliver them to people in the public, and the public can deposit money to receive books and the money goes to the publishing proposal which is either confirmed or denied, and then the book itself is printed So we wanted to see what legal rights and obligations existed in all these different steps One of the things that we kind of came up with and one of the insights that we’ve been really trying to drive home here is that, you know, there’s a really strong need to narrowly scope exactly what a DAO is doing from a legal standpoint so that you don’t run into any of the contractual issues, the agency issues, those issues of uncertainty where people might be out of money One of the things I glossed over before and should have mentioned, if you don’t have one of these legal containers, the United States and all these state governments will assume you’re a general partnership, which means they can go directly after you for whatever liabilities that the organization occurs Just to identify the “you” it’s all you all Every member is jointly and severally liable, with a general partnership That means if, you know, like member one has, like, an extra Toyota that can be impounded and member two as a vacation house and $20,000 in their savings account, and member three has whatever, like a painting, they can go after everybody until they paid off the debt So general partnership is, like, ultimate liability exposure. If you’re going to have a legal entity, you don’t want later somebody to say your DAO is a general partnership. It’s better to get ahead of it to use some of our open source code so you can select the entity and then engineer the legal relationships and roles according to the business model that you have in mind Yeah, that’s especially the case if you have member four who lives in a trash can and screws everything up and does something wrong. They get in trouble, then, you know, you lose a vacation home and all the good amenities This is an interesting hybrid on the last one, where the DAO was a legal entity itself And where the individuals, where the DAO is more like a tool or a platform, and the individuals maybe had a different corporation I call this an automated and significantly — like some level of autonomousness, but there’s humans doing the book. Do we like this book? Do we not like this book? Do we want to put more marketing behind this or not? So they’re choosing the distribution of their resources against selecting and then pushing the new materials and who they want to work with This is hybrid approach on the distribution. At one end, there’s a completely automated entity, and you could create partly automated entities to go and form a new entity, and then dissolve the entities. You can imagine this different ways That’s the far end of the scale Most of our work here is more practical where there’s hybrid between existing businesses and existing business models and human beings very much in the loop or driver seat but disappearing a lot of the complexity and making things more responsive to the strategic and tactical decisions you make because it’s all encapsulated within a single

integrated legal entity So you imagine the bookkeeping and the financials and the inventory and strategy and operations and HR, when you encapsulate that, we believe you can make decisions and adapt closer to the speed of thought, and that you can manage and be much more flexible. It will be a much better form of business Yeah, and to that end, one of the things we’re also doing right now is we’re launching a new publication which is going to come out Friday, the MIT Commutational Law Report of which I’m the editor in chief and Dazza’s the executive producer. The whole goal, it’s a little bit different than other publications, one, it’s focused on law, which is a new thing for MIT, where, you know, looking at ways you can reimagine and reengineer the law so it functions more like a computational system. So we have a lot of interest in learning, you know, what is bridging that gap look like? We also want to, because it’s not too — because these aren’t two disciplines that have traditionally been connected with one another, we want to do some field building. We want to have conversations about how these processes take place. We want to get people together and see what the good ideas are And then we also want to produce content. That content comes in a few different forms. This is where we’re really excited about what we can do, because the content is going to be traditional written articles, but it’s also going to be rich media, so podcasts, video lectures about how to code something, so it can produce some of these things It’s also even going to like a data playground where you can upload a prototype of an app, people can evaluate it, comment on it, deploy it themselves, iterate it, and the goal is to come up with better solutions that are accomplishing some of these goals we’ve been talking about Indeed. This is MIT pre-competitive research. As Brian said, it’s field-building It doesn’t exist yet, this field. And others that we’re working with, maybe some of you, perhaps, if you’re interested to find solutions and design patterns that work and evaluate them, and then the next step, you could choose if you want to invest in a startup or put something out in the market One of the things you didn’t mention and I’ll highlight in the data playground is reproducibility It’s hard to — what we really want with engineering the law as a computational system is predictable legal results. So you don’t want always to be talking to lawyers and have them say, “Well, it depends.” Well upon what exactly does it depend? Can we know that up front? Can we engineer a system to achieve more at least predictable legal results? The answer is, yes, we can. We think using the scientific method and the tried and true almost culturally DNA at MIT, if we test this, we think that’s important and that’s how we structure the data playground Yeah One of the other things that would be especially pertinent to this group here is we’ll have a podcast that will come out on Friday What are some legal primitives that we can come up with and fine tune and allow people to containerize and take away with them so they know what they’re getting in all these circumstances You might be thinking like cavemen or so forth, but think more about building blocks Crypto graphics primitives are the signature, dual key cryptography These well-worn primitives that are reusable We’re looking to apply this to legal premise. The previous speaker talked about identity, to a legal contract There can be some overlap, but there are some others that are unique to law That’s actually the Berlin working group at the top. But yeah. So to kind of like accomplish some of these things, we’ve been hosting these workshops. Here is one of the guys who came up with the statute in Vermont This is a drill-down on the DORG. This is the drill-down

It went almost an hour and a half on agency law and cross-mapping it to DAO’s and all the roles and rights and responsibilities of the parties and play it against scenarios This one, I think, that one is probably contracts This was — no, this was the publishing DAO. There’s a few more We kind of ran out of slide space But we do a lot of convening as an input to the design and prototype of systems One of the things we’re really excited about with this first release is we actually have a challenge. If you want to contribute this challenge, we welcome it, but we want to build up this repository of people who are working to produce code that automates certain of these functions. If you’re working on some small piece of it, maybe you want to understand how to integrate, like, a voting mechanism with one of these BBLC’s or if you want to go the other direction and figure out how you can, like, automate something in a way that produces certificate tokens, this would be a place where we would very much welcome, like, that sort of stuff And if there’s any interest in staying kind of up with these things, we have a computational law telegram channel where you can get involved and kind of spit around feedback and ideas and start populating this space together with us I believe that might be it Probably the best thing to do is go to and click on forward slash “contact.” Join the email list and we can have even more curated list of when we’re doing things and communication. The telegram is great. I live there. But it can be chaotic if you’re not used to dense chatter on Telegram I feel like we should say something else about this This — this challenge also is part of the release, the first release, of the publication, which is — our soft launch is Friday of this week. The theme of the first release is automated and autonomous legal entities Several of the articles are on that. Several of the projects are on that. Several podcasts are on that. As well as other law themes There’s the anchor article where the big vision is set on what is computational law and that’s amazing And one of the things I know we want to work on, with the conference organizer, is actually modeling the legal entity aspect of, like, a robot arm that creates art So we think this is a very — it’s sort of adjacent to a robotic and DAO publishing company It’s more art. It’s not different in kind. In some ways, it’s a lot easier because the housing of the robot arm actually has a place where we can understand that we can work on code You know what the robot will do We think we do And we can start to engineer against certain scenarios and hypotheses What if the artist owns the art, but the robot is doing something, or the consortium that purchased the arm is considered the owner, or if the robot itself is considered the owner, on and on There’s always permutations on scenarios. You can’t understand law or legal outcomes in the abstract Law can only be understood when applied to facts That’s why lawyers say “it depends,” because maybe they don’t know all the facts yet We think this will be a great platform to engineer all the relevant facts and play it against different scenarios to see whether they’re getting the expected results for the legal roles, relationships, rights, and responsibilities, and fundamentally the legal outcomes we’re seeking to engineer So that’s, we hope, that will be one of the challenging results that we can hack together on If you have interesting ideas on this, well, intersection of robots, AI, and law, with respect to legal entities or, more generally, we’d love to hear about them right now. We have some time for open discussion Thank you [Applause] Any questions? Hi I just want to ask you around checks and balances when it comes to information coming — lately, I’ve listen curious and just interested, AI, what’s your thoughts around sort of filtering that information

and also the checks and balances of information feeding into the system? In other words, if you put garbage in, you might get garbage out How do you do the checks and balances? So let me see if I’ve got this right The basic question is, like, if you’re set up so there’s some stream of data that you’re trying to — that you’re ingesting as a decision-making function in the internal governance of the DAO, what happens if that data gets corrupted or something like that? And it starts producing all these terrible outcomes? And to answer that, I would say, you can start to — there are certain things you can do that would be modeled kind of after, like, high frequency trading algorithms So if there are a certain amount of calls one direction or the other, that signals something super volatile, you could have it set up to it meters off, or it requires somebody to look at it or the group to come together to reach some sort of consensus before it can proceed first proceed forward You have to create a legal pause button to ensure what you’re doing is not going in the wrong direction. That’s one thing that comes to mind there Yeah, going further, some of it comes down to good old fashioned information security You’ve been on the high frequency trading platform, if somebody hijacks it and gives it wrong market information, so it starts buying other things, of course, that’s probably crime and fraud, but that is one way you can get corrupted information of garbage in to manipulate activity Information security doesn’t go away. It’s even more important with automated, especially autonomous systems, to make sure it’s getting the inputs expected, and the oracles are the other sources You also have to be thinking beyond a direct attack. Whether the sources you’ve chosen are really appropriate I think you mentioned the word “bias.” So that’s a big question that I think, for example, if you have — like this system, for revolving loan fund, you can’t really see the swim lanes on this too well, but imagine there are swim lanes It didn’t come through our JPEG, but there are decision points where all the information for a loan application is presented to a board, so they make the decision. The way this system is created people log in and authentic themselves and they have authorization to approve a loan If there’s more high risk loans or microloans. So it depends on two things. Making sure the authorized people are logging in to set the parameters and approval chains and work flow points But two, are you gathering the right information on your loan application or other information we assume we’ll be getting from Bloomberg and other places This is basic business judgment One of the things Sandy Pentland says, in his What About Computational Law article that we’re releasing on Friday, the critical importance of the legal aspects of these systems, of modeling them, and not forgetting about them, assuming all your decisions initially were absolutely correct, but he said monitoring and then adapting them. If some of the information is biased and you need other information and need to change things, in order to hone the model to make better decisions so there’s less bias you don’t want, this needs to be built into the design of the system So Sandy very much advocates the computational law systems, everything from creating a statute to managing your contracts or other business types of instruments, that most of the action shouldn’t be on the initial design phase but the design of continuous adaptation And information that might be perfectly good in 2020 may end up being biased and not particularly reflective of the key inputs in 2021 or 2022 You have to continuously hone and identify where the bias or other inefficiencies are as you go with computational law systems I suppose with any system, but we think this needs to be part of the DNA of computational law

Maybe we can’t conceive of all of this today Hey, Adrian One of the primitives, maybe not in the sense you meant it, is reputation Now, the reputation issues sort of cross over all the de-centralized AI stuff we heard about earlier today, in all the domains Where does computeddational law impact reputation or vice versa? In other words, is there a narrow subset of projects that are already underway? Or aspects of the discipline that can be applied to the reputation components? I think there are examples out there now Es stona has the e-birth certificate There’s the land registry that’s on Etherium. I think these different groups are starting to plot some points down on what the factors are of identity you need to have in order to properly authenticate what you’re doing As more and more governments and different players start to do this, that will become a little bit more clear You’ll start to identify more of the general trends and be able to say, okay, we’ve seen of all these places, here are the five most common features you should look for, and go from there Yeah. I think that’s all good practice If we go deeper into is there something in there that might be a legal primitive? First is the identity itself If there’s a creature or human being that has a legal personality. We’ll release on Friday this hour podcast with Drew Henkis to start to identify what we think legal primitives would or would not be If we considered it a legal primitives, there’s consensus around that You could imagine constructing that primitive, that concept of a primitive, such that the identity has attributes that may be part of the identity of the primmive. So maybe other identifiers Some, in fact, maybe things that adhere to it, like identification If there were a basket of identity attributes, you could have like an agnostic sort of generic thing we call “replication attributes ” At that level, I guess, there could be reputation that was like a legal primmive. But honestly, there’s not even consensus among the few people talking about legal primitive how this would play out with identity at all at this point, or whether identity is a legal primitive. We’re not sure at this point I don’t want to be that speaker that puts the question that says a question with a question, but I would love you to think about that, Adrian, and talk to people about it, and then talk to us about whether you think identity is a legal primitive. And what the association of the third party adhering to the identity might be The link between identity and reputation is context And what’s missing, because we don’t have science around reputation worth anything much these days, what’s missing is introducing — not worrying about digital identity and identity as a legal construct, but rather introducing the principles that I think law can bring into defining the context In other words, adjudication of reputation, or the gaming of reputation, or how do you control the gaming of reputation? Don’t bother about the stuff at the low levels That’s would too low level. But rather this issue of defining context in the legal sense, in adjudication or appeal, et cetera Thank you. That’s helpful Sir? Did you switch seats to be in mic position? Just two questions Right now we’re facing with a

situation with the supply chain and some process is really long, and if we are talking about autonomous system and fully automated processes, sometimes it’s hard to define in some cases, something happens, something goes wrong, it’s hard to define who is who will be in charge to pay for that And in some of juridic things, some situations really exist, when the many participants in the process, for example, of death of the person, and nobody can be blamed, because there’s a really long chain. There’s movies about that But will we see something like insurance funds for autonomous systems and robots and AI’s to insure that in any case of damages from the side of such systems that damages will be compensated? I can start us off The way you posed it, which I thought you posed the question really well, but there’s one word I must suggest we amend. You said it’s hard to know what happened and who is in charge. But let’s get right to the real point. Who is accountable or responsibility? Who is going to be left holding the bag if something goes wrong So look at that dimension of it, what we want to avoid is an accountability gap. Some people, it seems, in the early days of the DAO, especially, were specifically attempting to achieve an accountability gap where the idea if something is going to happen you can’t touch us; we’re not part of any jurisdiction. It’s very questionable whether that’s a beneficial or sustainable or desirable system at all As you look from a MIT Law perspective, we’re looking alt systems that operate well based on our social expectations, which includes accountability To me, when there are human beings and other corporations utilizes automated or autonomous systems as tools, it’s not a big change in terms of what is accountable You need attribution at that point. To whom do you attribute the act? Now closer to your question and assumption is what happens when the system is kind of taking actions and causing consequences without human review or approval or even knowledge? Okay So now we’re in the fun zone In my mind, or I believe that it is not just possible, but essential, that when or likely these systems start coming online that a major part of the equation, like required, is that there be financial and other mechanics to ensure there’s no accountability gap If one must — if all one has is the automated or autonomous system to hold accountable, we must look at things like insurance, bonds, reserve funds, and things that are proportional to the harm or accountability that may be required for the type of thing it’s doing If it’s selling books, that may be relatively low If it’s doing munitions and nuclear weapons distribution, it might be quite high And everything in between So looking at the potential exposure of different business activities is a bit of a magical art more than a science, but there’s risk managers that can begin to size up what would be an appropriate, you know, risk management kind of — a premium but an appropriate risk management capabilities to have for certain situations, like is insurance appropriate, and if so what kind of product and what would the premium be, do I need a bond, or a liquid reserve fund, or other things like that. Or is there a common defense fund? There’s different ways to start to build in accountability, but I’d say it becomes essential. And it ought to be built into the process of having fully autonomous systems that are capable of causing harm

So I guess my answer to your question, like, is this something that could be thought of? I’d say, like, hell yeah In fact, I think it must be It must be thought of and it should really be part of the core design And the second question, I think it’s related to the first one Will we see something like open source license that’s for AI and robotics? Not the source code, but to open itself ? If the robot can by itself, if it’s possible, in the future? Will we see something, some license like this? What would the license do? Like, for example, in the open source license, you are putting that I’m not — I’m not the owner of this code no more I’m opening it to society And I’m not taking the burden of the damages or anythings. Is it possible the same thing for robots, AI, for example? So I think there’s a couple of concepts there. Approximate one of them is the concept of emancipation. The open source is sort of close, but let’s get really point blank on the target. One could imagine, one could structure, like, legal documents and business models and social arrangements where we deliberately intend for some code to be emancipated So it was owned at some point, and at a certain point it sort of owns itself or it is independent It becomes autonomous or I would say emancipated A young person is — can’t form a contract when they’re 12 years old, but by the time they’re 30 years old, they can. One of the things that happens there, technically, legally, it’s emancipation. A slave, similarly, cannot own property In fact, they were considered property When the slave is freed, they’re emancipated So an emancipation-type event is one way we could see this happening Another thing is sometimes what I refer to as the broken leash You have this dog or something, and it’s on a leash, and it seems to be going pretty well and the dog bit through or otherwise just ran off and the leash is out of your hands and now it’s broken We have this rogue AI going around with all intents and purposes — or maybe the leash is relinquished because the only person who created and owned it and operated it is now dead or they went to jail or they don’t feel like doing it anymore or what have you. You can imagine conditions that result in that I think the interesting question, in fact, I can go one step further, in 2019, I would say I envision this is inevitable we will see these things develop in the next handful of years I’m not going to put a number, but in the future, in your lifetime. But the question becomes, okay, how could an emancipated AI or robotic or software be kind of a wholesome healthy desirable legitimate creature on the terrain with us? So this starts to create questions about what types of requirements or constraints might be appropriate for that This is a question that it’s just about time to have realistic conversations about it It’s still premature, but it’s not too premature to start thinking and talking about it I’m glad you asked Yeah, a few years ago, there was who owns the intellectual property rights for the monkey selfie where the animal took the selfie of itself. Who owns that? One of the things you can look to in order to determine that ownership is, like, does a legal personality container exist for that entity? So whether it’s an animal in certain places Animals do have legal personality rights in some places I can imagine if there was some sort of registration process, some sort of, you know, indicating of what those mechanisms were, the decision-making that went into it, you could have a legal personality for robots and what that actually looks like remains to be seen. But I think we’re getting closer and closer to understanding what the contours of it look like

One component and inner working that would showing up on the contour that you could connect with it, would be something like a license plate, even if it’s virtual You could say, ah, this AI or autonomous entity belongs to, you know, acme corporation or Sandy Pentland. It’s a personal shopping bot or what have you That AI, when I look at the license plate, the license plate is visible to others, is emancipated. Good to know Well, then maybe before I conduct a deal with it, I should check, does it have the standard insurance and bonding or not? Is it fully paid up? Am I doing something within the scope of its capabilities? Is there a file I can query about who owns it or if it owns itself and if it’s emancipated and what my recourse and remedies would be if it all goes terribly wrong? But we need to keep our eyes on, is it all going beautifully, wonderfully right? Some of these things can be extraordinary for the innovation and economic prosperity and social issues that they can help us to resolve and achieve some of these deeper goals. What we really need to be doing now is fundamental engineering and sort of pre-competitive research and development on designing the types of containers that we can make — we can get the best out of these capabilities while also maintaining reasonable risk kind of management, and also maintaining our values intact And I think that pretty much brings us to the end of the session Yes Sorry. Do you have a final question? My question is does already exist a framework which allows — could allow automatic litigation? So this is a beautiful idea, but this happens not to be legal, so another would detect a wrong and open litigation. Say I don’t want to pay you, because in certain context, for example, a car, autonomous car, not wishing to pay a parking lot, or in trading, many times litigation could be automated litigation might be an option But the question is, is this le we have a framework for that, so legally, it could be mitigated? On the trading floor or on the parking lot? Love it. Thank you. Last question. Quick So you could set up a framework for that. A lot of times when you enter into agreements with banks or other parties, there are arbitration clauses. You can imagine an online disbeauty resolution process, like what Amazon or e bay has. Those are a lot more efficient than courts are If you had something like that set up where when you’re setting up one of these, where you’re setting up a DAO, you kind of have a check box for, you know, this DAO prefers this sort of online dispute resolution but it will do any of the following online dispute resolutions. You could have a situation where something happens and there’s a goal to quickly expedite all of these legal processes and it can automatically run through them Indeed To play it out quickly, let’s take the parking lot. That’s a good one An at ton mouse vehicle doing things like Grub Hub. It shows up at a parking garage. It has a chip or something so it can be identified and it knows where payment could go. One of the things you could structure, on those components and building blocks, if I were a parking garage owner, I may be part of a consortium that developed a standard that would ask can I pre approve your credit card for the amount of time you’ll be staying here in advance? If I did that but it didn’t clear by the time it was time for the car to go, I could have an agreement when it entered into the garage that I can maintain possession of your vehicle until I get payment. Or something else That’s where we get into questions of recourse So the credit card was pre approved but when they did the

sale it didn’t go through because there’s a reached limit or a charge back If you can check up front as part of a data exchange that it has a certain insurance or fund, then at least you know you have recourse overall so you can let the car go So Does that make sense? It’s largely built upon and just uses the existing systems and frameworks, but we now need to sculpt more API’s and add a little bit more to the transaction codes and the business models in order to build out full use of the capabilities So with that, I think the full use of our capability is now expired First of all, I think we should thank Brian and Dazza for this amazing last session. Thank you, guys [Applause] Listen, it was great. I learned a lot Also, I think this is cool that, like, we can cool This is a green space Definitely, definitely. With this, I’d like to close the event. It’s been a long day. A lot of information. Right? But I think we all learned something new, like today So I hope to see you next year, maybe here, maybe in Europe, maybe in Petersburg, who knows? But yeah before we leave, I have to say a couple of things First, thank you for coming here Thank you for showing up and doing the networking Second thing, we’re going to start up a Boston meetup group There’s not much time left, because we have to leave at six, because, and the important message is this one, we have drinks planned in the mid hall, which is like a bar ten minutes away from here I hope to see you there and you can ask all the questions you couldn’t ask now, or just chat with a beer. With this, we close the event, and thank you very much [Applause]