Leveraging AI & Layer 2 Blockchain Solutions - Humayun Sheikh (Fetch.ai) - TOKEN2049 London

hello thank you shane um well lunch generally it's quieter after that so hopefully i won't send you to sleep left it for 15 minutes so we won't do too much um dull and boring stuff so five minutes at the end will be for q a i'll do the 15 minutes so a little bit different to the usual nft um and all the the hype stuff which is happening but this is more about the uh back-end infrastructure that we we at fetch are building um so i'll take you through um multi-agent systems and you know then we can go through the questions so what what was what is fetch what's um [Music] what is it about we'll take you through that i'll explain to you a little bit uh about agents that we're building uh most of you might not even know what they are how it why do we need blockchain did we really need it to be on blockchain and how does it all work then we're going to take you through what what ultimately we want to achieve from this because it's all great blockchain it's all great automation machine learning and ai what is it in the end that it's going to deliver to us and then we're going to take you through some use cases which we have built which do apply to some defy space applies to mobility applies to other things so let's uh let's begin the show then i guess so we were set up in 2018 which was kind of the end and we did an io and our token is called fet with team of 78 people mostly researchers programmers and business dev we've survived so it's been up and down but we've survived our backgrounds mostly come from deepmind which is sold to google six seven years ago now some of you might have heard the name but it was based in uh london and we sold to google seven seven seven and a half years ago uh we headquartered in cambridge so if somebody really wanted to speak to us you can come there is an office but we mostly distributed team just want to quickly set the scene our partners um i'll take you through the crypto ones as well but names like bosch you probably have heard festo bmw mercedes daimler uh they're kind of working with us to see how we can bring this new uh technology uh actually in real world use cases this is a bit more a detailed partnership list most of you know most of the names here but so just to explain what fetch is so we we're trying to create organization in chaos and you need the infrastructure for doing that and that's what fetch is building so what does it actually mean in terms of dlt crypto what it means is we're building interoperability that's the key blockchain has multiple stakeholders they need to all come together and i'm taking through the basics here now uh most of you probably already know this now they're open source tools which then will bring ai agents onto the blockchain why do we need them you need them for automation because if you i'm sure most of you have done this gone on and tried to stake a token take it out improve the yield it's quite a complex task it will improve and i'm sure as we as time goes on it will improve but these are the tools which you will need to do all of that so how do you take your car and attach it to defy you know that that's where we are coming from how do you connect how do you connect your home to a defy system which can finance your washing machine for example so these are the kind of tools which you're going to get from fetch which enables you to build those solutions um so to do all of that we've kind of put it into three categories so what are we building so we're building a multi-agent system blockchain itself is a multi-agent system and multi-agent system is where you have multiple stakeholders trying to interact with each other now multi-agent system can be automated based on machine learning and ai or it could be just humans um so we're building a multi-agent system we're building the tools for that multi-agent system but now if you think about it these agents which are autonomous and are trying to interact with things what is the best thing they could use is machine learning and ai so how do you bring machine learning and ai to these agents who need to operate on your behalf and on other stakeholders behalf so you have to build the decentralized machine learning core engine for that then the next bit is you don't want especially in d5 and especially in decentralized space um most people know if you lose your wallet keys you lost them so how do you and and if somebody's hacked them they're hacked so how do you secure not just your keys but also how do you secure data which is going to be moving around and we talk about data privacy a lot how do you enable that in a system which is autonomous and it's going to run in a decentralized fashion so we've been building tools to bring that to life so what does it actually mean you probably most of you will be able to relate with this it's a we have our own chain which is cosmos based we don't just work with evm which is the ethereum virtual machine we have cosmosm which enables you to do a lot more complex machine learning tasks um and we if you think about the stakeholders and at the moment how do you find each other so if i said to you you have a wallet and somebody else has a wallet how do you actually connect with each other without you actually putting somebody's wallet address in so you have to find a way to do search and discovery and and that to be fair is the core um problem that we need to solve so how do you so if you look at google google is the search and discovery for websites how do you bring wallets and how do you bring these agents and multiple stakeholders to find each other so we're building that search and discovery mechanism and that's the agent based framework which can actually be used for layer 2 scaling because you don't want to do everything on chain you want to do stuff off chain and i'll take you a little bit more through that detail in the next few slides so how do you build um how do you build these connections so we build the tools we give you the tools to build these solutions so you can when you come to fetch and if you're a developer or if you're a business we're making it easy for you to deploy a real world solution on a decentralized system using machine learning and ai so that's what you should be able to do is we're making it as easy as deploying a website so a little bit more towards the l2 scaling solution so what does it actually mean so we the multi-agent system and the agents which we are building which are software agents they combine state channels so you have a peer-to-peer connection with each other all the agents can speak to each other peer-to-peer and then you can actually deploy multiple chains apart from that so you you know there could be 10 agents which get together to do a task they deploy a chain they run the chain for as long as they need to do and they then remove the chain so you're kind of combining state channels and the techniques from plasma into one kind of system which is the layer two solution which the agents bring then you have the aggregated signature verification because if each and every single agent and there could be millions of those or stakeholders have to do that on chain you've seen what happens when you do a d5 solution like a uniswab and the chain is busy it costs you a lot it takes a long time so you do all these things off-chain so you need the signature verification off-chain techniques then we have the open economic framework so what is happening is all these agents are doing these tasks but what are they doing it for they are doing to exchange economic value and what you need to do is give them a ability or protocol to actually connect with each other and exchange the information speak to each other the same language so we're giving we're giving them this protocol to to do carry out all those tasks and that's called the open economic framework the next bit is there is multiple chains getting deployed how do you tap into all of them there is a lot of issues and there's a lot of bridges which are being built most of them are centralized at the moment how do you resolve that issue how do you create this interchange bridging bridges for the interchange communication so we're building the tools to do that um i'm just going to skip through some of this but just to give you an idea you could look at some of the applications if you can see there is charging points there is mobility there's micro mobility you can do parking now if you bring all these agents which are helping you to do all of that and you try and put that on chain that's not an easy task you need tools and mechanism and you need the underlying framework for them to be able to do that and that's when the true decentralization starts to come in because just trying to exchange tokens from one chain to the other is only the start it's only the beginning of you know this whole process but the true adoption will come when you can use uh decentralization for multiple things for data sharing for mobility for travel and that's that's the second layer which we're building on top and and and as i said in the layer one we're building our own chain where you can have all the tools search and discovery and that just takes you through that um so just to give you an overview where where do we think we're going i quite like this uh this uh architecture diagram from bosch it's very general and it's very you know top top level but if you think about where we started from we started from telephones we started connecting yellow pages and we started becoming a web-based economy we started to have websites and then came e-commerce and e-commerce brought new business opportunities it's all about connectivity as soon as we take a shift in how we connect uh there is always new business models which come around and and and if you look at where we are we're going from roughly economy of things and we kind of moving upwards so so where there will be a super modular economy where everything can speak to everything and you don't need centralized entities to carry out your work for you uh you have your own agent your agent looks after you uh nobody can uh try and take advantage of you because you're running your own uh finances or you're running your own machines or you're running your own space and that's overall when you have that much data and that might be transactions what you then do is that's when you start seeing the true intelligence coming and evolving out of those decentralized space so our objective ultimately is to build this uh artificial intelligence which sits on top of everything and we don't like to call it general intelligence because it isn't it's more collective intelligence and we'll take you through what we mean by that um in a quick uh succession so what we just talked about um this is um kind of a concentrated effect of that so you have you have delivery networks we have the we have actually created a proof of concept for delivery networks so you can actually decentralize uh uber for example or a distribution network you can do parking so you don't need these big parking companies in the middle you can connect directly to the parking spot we've we've delivered that recently in iaa and that was received very uh very well we have a ai autonomous travel which is quite interesting because at the moment you have these aggregators which sit in the middle but you don't need them so you can start seeing the potential of crypto or effectively dlt when you start thinking about what it actually can do and you on top of all of that you can pay with crypto or you can pay with any token or with any currency um but to bring that to life you need that framework you need that whole architecture in the middle um for d5 especially an example of uh if any of you have participated in uh which i'm sure at least probably half of you have um into liquidity pools and if you think about six months ago to pull out liquidity from a pool when uh there was a rug pull happening some some project was scamming you how do you automat how do you automate that process how do you monitor if somebody is going to pull the liquidity on you uh you don't because then a lot of people lost a lot of money and you have to be vigilant your agent can do that for you and it's a it's a four-click system where you can actually provide liquidity and pull out liquidity so you can build solutions like that uh very easily on fetch just an example you know if you went on and you can see this is an agent-based dex that we're building so it's d5 agents doing multiple tasks we did as i said we did something in mobility which is you know you can deliver people packages and parcels uh this is all a decentralized system it's up and running uh we haven't launched it it's commercially not launched but it's in its mvp stage with trialing at the moment again this is all built on a decentralized space you have agents you can book you can book things you can book people you can book taxi you can deliver packages pay with crypto as i said we we did a trial with the iaa mobility and that was received really well we our partners really love it what is the key point the key point is you need agents to have this identity because if you don't have the identity the how do you prove who you are so you need tools to take that identity you also need payment tools to sit with agents so they can make autonomous payments and you need the search and discovery so i think i'm running out of time but i'm gonna quickly go through it and leave this last bit for some questions um we we built some co-learning uh platform where you can actually all of you can participate and we did that uh within our project to just show machine learning is not some scary thing we all participate in it all the time and we're giving you tools to connect your tokens you connect your agents to actually deliver machine learning and these are the arts which pieces which were delivered using a colon and a lot of people participated in it i don't know why that one looks like elon but it does maybe people are talking about elon somewhere so um you know this is and the other one is pretty twisted anyway so um so you can see just to wrap it up you can actually use fetch tools to build new solutions it's not just about defy it's about bringing it's about bringing new processes and new ideas to life so if you if you're thinking of deploying a business um in the real world but using crypto and decentralization you can actually build it with fetch tools you you you don't have to worry about which chain you just build a solution and you choose the chain or you can say you know i want to use it for all the chains you can you can use that fetch for that purpose uh you can actually do scaling quite well with the agent-based l2 solutions because you don't need to worry about transaction speeds of a particular chain if it's if you want to do it on ethereum but you want to do millions of transactions you know we have an l2 solution built in with the agents then the last bit regulators kyc aml id i mean where do you start all of that taken care of you deploy your agent you can just connect kyc ml and it's done so you can actually deploy a real world solution in a new business space which can actually work with your crypto token you can deploy the token you can buy and sell in the token all using agent-based system so that's what we're doing we're building this bridge between the traditional e-commerce and crypto itself and that's that's it from me and thank you very much awesome round of applause thank you i'm always fascinated just how quickly you guys at fetch ai are moving that was an awesome presentation thanks very much first became aware of you guys i think it was from outlier ventures and jamie really pioneering what you guys are doing but it's great to see you coming out of cambridge and really flying the uk flag would anyone have any questions for fetch ai gentleman here i um hi there think thanks a lot for the presentation so i had a question around what's the main difference between collective machine learning and decentralized machine learning i mean i mean collective machine learning and cloud computing machine learning and like what would you say um problem what problems can decentralize machine learning solve that cloud computing can't that's that's a really good question because what you've done is really gone to the core of the problem so we have federated learning what federated learning does is you have this model which sits in somebody's control and you say all these people have this data i'm gonna train the model using all your data but i own the model that's that's the main thing so let's say google deploys it you connect your android's phones with it you teach the system and then you pay to use it because all your predictions come with a payment right somehow you you know so what we're proposing in a collective learning is it's truly collective because you train it and then you own it because what would happen is the the model itself belongs to the people who train it so you train the model you connect your phone with it and you say okay i'm gonna train this model and and all these people do the same when the income comes out of the model it also gets distributed to everybody so if somebody says i need a prediction from this model they have to pay for it and the payment goes to all of you so that's the difference between these collective and running it in a federated model so the one business doesn't own it the people who train it own it great question one more from the gentleman here hi could you could you please go um more into depth about how your uh like liquidity provider agent manages against the security risks of operating in those liquidity pools and its anti-rug pool uh solution sure so so you can do multiple things um so one your agent is in your control and if you think about a liquidity pool itself uh it's it's we call it decentralized but it's not truly decentralized because it's one contract where millions of people put liquidity so if that pool gets hacked everybody's at risk right so when you have an agent and this is this is the transition which fetch is making right now so if you have an agent and you have five tokens whatever the tokens are you're in control of your agent and your control you're controlling your own key password and we have tools to make sure we can do some social recovery on your uh on your key as well so so that's the that's one thing because now all of you have your own tokens with your own agent you don't put it in a liquidity pool you trade it on the fetch system and it automatically treats it as a order book or a liquidity pool you don't actually go into a pool now but but let's let's rewind back to the current system where you have this liquidity pool what happens is few things which is somebody comes and crashes the price the sentiment is going wrong somewhere and you can see it coming but you didn't take your action at 3am in the morning when you were asleep so the agent is monitoring all that and as soon as he sees the sign either via sentiment on twitter or telegram or you know just in the news it pulls out the liquidity and when it pulls out the liquidity the actual key control is in your hands so it's not so you hand it over to the agent only to do one thing which is pull out my liquidity that's it you can't move it the tokens can't go anywhere so that's how we take care of that security risk