Ben Sheppard  00:43

Hello, everyone, and welcome to another episode of the TX – Tomorrow Explored podcast show. I’m Ben Sheppard, Managing Director at TX. On this week’s show, we have Humayun Sheikh, who’s the CEO of Fetch AI. Welcome Humayun. Great to have you on the pod. How are things?

Humayun Sheikh  01:08

Thank you, Ben, thanks for inviting me. It’s a pleasure to be here. Things are interesting these days, with COVID and everything else that is going on and it’s been busy. It’s been good, busy, and very interestingly busy, especially with all that is happening due to COVID and all the new opportunities, which are kind of arriving in this new space, which is evolving. So, yeah, we’ve been lucky that, you know, our development has been going well.

Ben Sheppard  01:42

Correct me, if I’m wrong, you’re located in the Cambridge area of the UK. So, on a map that I’ve seen coming out from various memes on the internet at the moment, that would be in the safer part of the UK in relation to COVID. I think anything above the middle and starts to get you know, there’s a lot more cases up in the northern part of the UK. Is that right?

Humayun Sheikh  02:03

I think, yes. At the moment, it’s been pretty good here where we are, Cambridge side has been pretty safe. I think it’s people just being sensible. Really. I think the controls which are in place are doing well. I think we’re not seeing that many people going out and partying. So, I think it’s a good region. Who knows? Let’s see how it goes. We’re looking for another two-week, possible lockdown. It’s not been announced yet, but let’s see.

Ben Sheppard  02:37

Yeah, things are tightening up a little bit here, but yeah, we’ve got quite a small population in Finland and people socially distance quite naturally, when COVID first came out, there was a bit of a joke going around that we’ve been told we have to social distance by only two meters now when our preference is already three. This isn’t comfortable for us.

Humayun Sheikh  03:06

It’s pretty sparse.

Ben Sheppard  03:08

Yeah, it is and I’m lucky I live out in a nice part of Espoo, it’s everywhere in the countryside in Finland. So, I look at my windows and I’m surrounded by trees, but you’re never too far from a mall, either. If you want to get some groceries and stuff.

Humayun Sheikh  03:27

Cambridge is quite similar. It’s not that densely populated. It’s surrounded by, you know, more agricultural sides of things. There’s quite a bit of countryside around.

Ben Sheppard  03:39

Yeah. So, tell me about or rather tell our listeners about Fetch because we’ve obviously worked together in the past on pilots through to alliances and things, but I’d love to, if you could just explain a bit about what is Fetch what brought the idea around for this company, when was it established?

Humayun Sheikh  04:00

Right, so Fetch idea in its own self has been around in the co-founder’s head for the last probably 10 years. When I was involved in Deep Mind, it was quite interesting times because Deep Mind was creating the artificial general intelligence story. The artificial general intelligence and commercialization of AI has been on our agenda for quite a while. It’s not an easy one to commercialize unless you’re a big, centralized entity, which has got a lot of data. When we talk about AI, and generally people find statistics as unsexy and AI sexy so everything’s statistical is now AI. Anything you learn from history via statistics kind of become part of it, which it truly isn’t. So, we been thinking because Toby, my co-founder was from the gaming sector, he worked with massively multiplayer online games, we worked with luxury of working with Dennis, who was also part of this whole gaming environment. I come from a very commercial background, where, you know, the objective is really how to deploy these technologies to make some, you know, money and build a business around it. So, the idea we’ve been putting together has kind of its diverged a little bit when we did Deep Mind, because Deep Mind was very research oriented, very focused on research. However, ultimately, you need to start commercializing the technology. So, what came out of the Deep Mind, venture, and us after the sale to Google exiting, which was probably the natural exit, because you need a big entity to take benefit of such technology. Google was one with a huge amount of computer power, and you know, plenty of places where they can save money, plenty of products where they could deploy it. Very good, excellent, but what also came out of that was that if this whole space AI space has to really take off, you need to have a way of deploying these entities, these machine learning scripts or AI, in its own true fashion, you need the ability to deploy them.

The problem is that the data is so everywhere, you need to really pick up the data, if it’s sitting with just one entity, then the competitive advantage stops them from giving it to other entities and using it. So, the idea came about where we felt that we need to create a system where different people can connect easily and utilize machine learning AI, even statistics in a more collective fashion. Where you have multiple stakeholders, not just one stakeholder. So, that’s the start of the whole Fetch AI. Now, two years ago, we kind of actually officially launched it but a couple of years before that, we looked at blockchain. What was quite interesting was, Blockchain is a multi-stakeholder system. You have entities which do not align in their incentives, but you have to kind of come to a decision, you know, to do something. That’s quite interesting, because if you look at what is happening is people want to keep the competitive advantage when providing data or doing something in a collective fashion, but they also want to benefit from each other, which is exactly what blockchain is and does. So, we started looking at that and it was very interesting, and it was obvious for us really, that if we want to do something like that, we have to do it on a blockchain based system. It doesn’t have to be just a true blockchain system, it could be a kind of a variation of it, but it has to be a system where multiple stakeholders can interact.

Ben Sheppard  08:30

Yeah, for me, it’s fascinating, because which you are building a system where you can have a group of competitors share information in a way that doesn’t give away their competitive advantage, but ultimately helps them all benefit by building new models and other things off of that data, then it creates amazing opportunities. It really accelerates the speed of innovation as well. You know, all the time that they’re holding on to data and they’re not working collectively, you know, they can only improve as good as the data they’ve got within their own organization, but you know, they can work collaboratively with their competitors and ultimately, they all can benefit and gain from it, then you can really see fast acceleration in technology developments, lots of innovation, I think. You’ve sort of touched on blockchain there more specifically, how is Fetch using blockchain within its technology, architecture and is interoperability with other blockchains important for you because this is one of the things that we’re seeing more and more of as there’s the rise of multiple blockchains and some of them be private blockchain, some of them be public blockchains. You know, and obviously with the Defi craze kicking off we’re seeing guess prices, rocket on a theory and blockchain and it’s, you know, sort of brings about the need for other blockchain solutions as well out there if those guess prices can’t be handled. So, I’m curious about how do companies like Fetch use a blockchain and how is interoperability with blockchains important for you and if so why?

Humayun Sheikh  10:16

I think it’s just evolution really off the whole space. I mean, Ethereum did a pretty decent job in creating what they created and the movement which started, when we came into this space with Fetch, what the idea was that there is no way Ethereum can take what we want to do on it, it was pretty clear that the scalability issues are going to come into play very quickly. So, although we started trying to build a variation of a chain, which could actually work for our agents, what Ethereum has is really the momentum. So, people kind of joining that whole space on Ethereum, which obviously, as we said, you know, it was going to happen, people were aware of it, there will be scalability issues, there’s plenty of chains, which are out there, which could be way better, way faster, and would work for a multitude of different things, not just what Ethereum is doing and there are plenty of projects which have great change, I think the issue is how do you get that momentum behind each chain, and how many chains will have that momentum. So, we’re not actually betting on one or the other, we’re just saying, okay, there are certain things we want to do. If no chain is offering it, we will do it on our own chain, but if other chains are joining this whole ecosystem, we want to be sure that you can deploy on whatever your favorite chain is, you can still deploy the solution on multiple chains. So, interoperability becomes the key here. That’s what we are focusing on to make sure that we have, and I think there’s a lot of other people also doing the same, there’s bridges being built between different chains. We’re not averse to using any of those, we want to use all of them, we want to be able to say to people, you deploy your solution to whatever or wherever the favorite chain is, and you can still interact with all the other chains.

So, effectively, the multi agent system, the agent framework that we have, actually enables you to do a kind of a second layer solution of interoperability, you can actually pass data, you can actually settle a using our agent system on multiple chains. It’s an easy process, because it doesn’t try and integrate too much with the chain itself. It’s a second layer solution a lot faster and can be made a lot more intelligent. So, for us, interoperability is the key, and we want to provide tools to most of the chains for people to deploy solutions on different chains and create interoperability, of course, you know, it takes time. So, you know, every chain has its own items behind it. Every chain has its own philosophy, and you can’t do everything for everybody. So, you have to choose certain specific ones. I think Ethereum is here to stay. Yes, the scalability issues, but the adoption, the first mover advantage, that’s all-in favor of the three and if they can deliver a version two, or version three, which is long time coming, but one day, it’s not an easy solution. So, you know, if that comes, then I think Ethereum is here to stay, but that doesn’t mean other chains won’t come up and you know, kind of start competing with Ethereum. I don’t see that happening all the time. Hopefully, we can build a solution which can span across multiple chains.

Ben Sheppard  14:11

Yeah, I don’t think it’s necessarily the mission of a theory to be the one chain for anything for everything anyway, as it there very much. It’s not like your traditional sort of corporate companies out there that absolutely wants to try and realize the majority of the market in this world of decentralization and Web3 is actually almost the other way around, isn’t it? They want to have their part in the jigsaw puzzle, but they don’t want to be the whole jigsaw puzzle. They want to be interoperable with different technologies and other chains and I think you know the speed at which our industry moves is incredibly fast. I came from hard infrastructure building toll roads, and railways. It was incredibly slow. Yeah, chalk and cheese incredibly slow. Now being in this industry, you know, if you’re not future proofing, the way you build your technology to be able to, you know, work with different blockchains I think you’re chopping the legs off on the product already, aren’t you? You’ve got to, be thinking ahead for this stuff, because it’s just rapid the way it goes and that’s what makes the space exciting as well. When I first came into this world, you did a bit of research, you think, okay, I’ve got my head around a few things. Now I’ve got my head around smart contracts, got my head around cryptocurrency, got my head around this particular blockchain fight, I’m good to go and then a week later, a whole group of new things comes up and you’re like, oh, my goodness, right. Okay, back to the drawing. Then you realize, actually, that is just the pace of this industry. So, for us one of the things that we like to do in this part is we try to educate our listeners that aren’t working in the industry on a few of these different terms that come up. I think this is a great opportunity. While we’re talking about blockchain, if you could just talk to us a little bit about smart contracts, how Fetch is using them how perhaps, you’re using the fat token as well in those contracts, that would be great for our listeners to understand not just how your project is using it, but actually what these different terms and technologies are.

Humayun Sheikh  16:27

Yeah, so I think in a very simplistic way, and I don’t want to go into too much technical detail, but the simplistic way of saying smart contract is a piece of code, which executes based on some inputs, provide some outputs and it’s run in a decentralized fashion, which means that, you know, once it’s running, the nodes are deployed in various places, so no one switch can turn it off. Once it’s deployed, it’s public, and people can see it, and you can trust it, because you know exactly what that software is going to do. Now, that is really a smart contract, that doesn’t have to be smart and most of the time, it isn’t smart, unless you make it smarter, which on Ethereum, is quite difficult to do, because the costs will skyrocket. So, now, actually trying to make a smart, contract smart, whereby the decisions are not just if the standard decisions. So, now, if you bring, for example, different intelligence into the smart contract, so you bring in mathematical computation, which is more in the terms of machine learning or AI, then truly, that becomes a bit smarter, smart contract. Now, the good bits and bad bits of both so for a financial transaction, where you don’t want, and you don’t want this black box approach, because if you know AI, there’s this whole black box approach, we don’t know how the AI is going to respond. If you want, you know, like in finance or anything where you don’t want to have that uncertainty, I think the dumb smart contract is the way to go, because it will take care of it, you know what’s going to come out of it, you know exactly how it’s going to behave, even then there is quite a lot of parameters around it, which people need to be careful about on how the smart contract behaves, because the effect of you know, block delays, or, you know, times it takes off, people jumping the queue to put transactions in the block ahead, you know, all of these things could actually cause manipulation. For Defi space, I mean, we’ve seen three or four disasters, you know, where this has happened, one has to be really quite careful to go through the code and look at, you know, what the outcome is going to be, but to do that in a way that is secure, and more intelligent, clever, and the decision making is done better. I don’t think current architecture enables us to do that. Especially the EVM.

So, you have some other chains, and you know, in fact, we are building this architecture for smart contracts where you can actually integrate machine learning and AI techniques, and you can have much lower cost for some mathematical computation. You know, there’s a fixed-point issue with Ethereum currently, which hopefully will disappear. So, these are a few things one has to be quite careful about, but then with other chains, you have other issues, which is they’re not truly decentralized or, you know, Ethereum is truly decentralized because you know, there’s plenty of nodes, plenty of miners, other chains might not have that. So, if you’re building a solution, and to be fair, you can build a solution on an enterprise chaining, which doesn’t have to be decentralized and you could actually build a smart contract, which is just a piece of software, which you can do on your computer as well and if you are building that kind of approach, then you could use some enterprise chains. Or you could use a deployment of a chain in an enterprise fashion. So, what you have, you know, a limited number of stakeholders who are running it who you can trust, it’s based on the trust element really. So, if you wanted a truly trustless system, then you want as many nodes as possible, and you want the smart contract to be open as possible. So, that’s my view on smart contracts. So, for example, we have a multi agent system where multi agent system can carry multiple transactions from different agents or an agent could a very simplistic way, which is what we’re looking to release very soon, if an agent books a hotel for you, we can come into details of why that would be the case, but if something happens like that, where agents have to transact, they do it on a smart contract. That smart contract might need a bit more computation than what Ethereum can handle, but at this point in time, I think where we are in this evolution, I think Ethereum is doing a great job, building the momentum, but very soon, we will need to if Ethereum doesn’t upgrade to something better. I’m sure there will be competing chains we were building the solutions on.

Ben Sheppard  21:43

So, I’m curious now how is the FET token used. So, if you had an AI that was making a hotel reservation for you, is the FET token, the method of payment for booking that room or is it the incentive that’s given for triggering the agent to make the booking, but actually traditional fiat currency is still used for making the booking.

Humayun Sheikh  22:05

So, the payment could be in either. So, it’s not that much of a problem. Payment could be in FET and FET could be taken and locked. And, you know, the payment to the traditional markets is made in a stable coin, by the FET could be in the background. So, that’s all you could just do it in a stable coin, or you could just trigger a credit card payment, I mean, there’s plenty of crypto credit cards, you could use and do that. So, that’s the last step of the transaction. Before that, if you said you are going to look for a hotel today, what you will do is go on to one of these aggregators, and you’ll input your requirements and the aggregator, you know will find you what you need. So, you’re going from many to an aggregator platform, and then you’re going to manage so for the consumers, many consumers can come to that central point and actually then connect with specific hotels. In our system, what we’re trying to do is give agents autonomy, so both agents can just speak to each other, but can’t speak to each other unless you can find each other. So, there is an element of search and discovery. So, how do agents find each other? So, there is a cost which is associated. So, if you want, rather than paying, let’s say 15-20% of the aggregator, you might have to pay a 10-20 FET, or, you know, five cents, 10 cents for an agent to find you and connect with you the right, let’s say, the right hotel, or the right hotel room, if the granularity allows you to do that. That’s where FET comes in, it kind of gets consumed in the search and discovery because it’s running on a network which needs to be run for this whole thing, session discovery to exist the agents to exist. So, somebody has to run it and the people who are running it the node operators, which is a program we are launching very soon incentivized test nets, which are coming soon, but the node operators will actually earn from running these agents or building these agents. So, that’s where the FET is consumed and then as we discussed the top layer, I mean, if you want to make the transaction and FET and there is an acceptability on the other side, and there’s a gateway, you can pay in FET, but that’s not really where the whole fuel of the system is. So, if you look at where the FET actually sets it’s the nodes, the cost of running agents search and discovery, and then ultimately, the payment.

Ben Sheppard  25:05

One of the things that, yeah, let’s carry on with this booking is a holiday example, because it’s a nice one that people can get their heads around. You know, when you want to do that, you go to a website, you complete a whole bunch of fields that say, okay, there’s two adults, one child, child is four years old, we’re flying from this destination to this destination, we want to stay for this number of nights, and we want breakfast provided and one of the annoying things that you’d have to go through at the moment, if you want to get the cheapest price, as often as you have to go to each individual website, and you have to input the same information every single time. I’ve made mistakes in the past, I ended up with like, a single bed in the room, and there’s three people staying in there. So, you know, there’s, I guess where I’m going with this is and when I was listening to you, I was imagining the idea that this agent, not only is it you know, doing the transaction of booking the room, but it could also ingest information about me, and it only needs to ingest it once to learn what my preferences are. Then it becomes intelligent, and it can continue to make the same sort of search, if you like search and discovery for different hotels or flights that I’m looking for, without me continuing to have to update this information and it could be consuming that on a real time basis from different data sources.

Humayun Sheikh  26:37

You just hit the nail on the head. That’s exactly the kind of service we are building, which is that it’s not just, you know, entering the data, it’s also finding those places to enter the data. That’s also a problem because why should you be looking? If you don’t go to the aggregator, why do we need the aggregator? That’s the first thing, because what the aggregator does is provide a service at the moment, only because you do not have any means to connect with that service provider directly. I’m going to kind of generalize it, which is a consumer and a service provider. So, the hotel room is a service provider, and you’re the consumer, how do you connect to a service provider? If not for either Google, which is still an aggregator, or any other aggregator which is a second-tier aggregator, which is, you know, like booking.com or Expedia. So, if you now think about new way of dealing with it, you know, don’t have to go, and search for 20,000 websites, you just say this is what my requirement is. You know, no entity is giving you any data which they want to give you, you are going out searching for things you want to search and the result is going to come back in the form of communication between two agents. So, the consumer and the service provider are linked directly. So, it’s like picking up the phone and speaking to them directly, but doing it for 10-20 hotels, and then giving everybody the same information. So, what this does is it actually finds you that the 10 service providers, you can link with them, it negotiates based on your preferences, and then comes back with a few options. So, you know, it’s not dramatically different, but the way it transacts is different, because you’re not going on multiple sites to book things.

Ben Sheppard  28:49

Thinking out loud now. Do you see there’s potentially going to be a rise of like individuals data wallets, that has information about them as an individual that every time they want to use an agent like this, to make these bookings, these transactions, whatever they might be, there’s like this foundation of information that’s stored in their individual wallet that they give permission for these agents to then access so that it can then start making these transactions on their behalf?

Humayun Sheikh  29:20

The agent should be able to dip into your information, it shouldn’t be able to let other agents from other companies dip into your information. So, you own your agent. You own your data. What you of course, you can share it if you want to, but that’s not really what we are after. Well, we were after training the agent and when I say agent, agent is a combination of different things. So, an agent has a communication module, how you will communicate with things. You have a skills module whereby you could actually add reinforcement learning into it. So, you know, it learns from what you tell it, you know, if you say, this is bad, this is good it learns from it, you then have other skill sets, like I mean, you could have you know, just like, there’s the skills you add on Amazon Alexa, right. So, one could be a, you know, taxi skill, the other could be a hotel skill, the other could be a supply chain skill, you know, you’re kind of putting them all together and you’re actually launching the agent to do a particular tasks, you assign the task, and it kind of does it, but it takes your data and leaves it with you and uses it for your benefit, not somebody else’s benefit.

Ben Sheppard  30:38

Yeah, interesting. I was thinking about how, you know, data monetization is a hot topic at the moment and the different ways data has been monetized, whether it’d be right or wrong. Some people are just doing it for the wrong reasons, but I was imagining a scenario where I have an agent, [Inaudible 30:58], my data, I’m asking my agent to make various different transactions on my behalf, but one of those transactions could be okay, I’m willing to sell this personal information in return for some level of income, I’m happy for this information to go to these different sorts of services. So, for example, one of them might be a health care service, because actually, by sharing information on how effective drugs were on me, I can actually help the pharmaceutical companies improve the quality of their drugs, I can help the health centers actually understand what drugs worked, what didn’t and you know, they can get more information than for their own AI models, and hopefully provide better services in future so. Probably, I just do that for free, because it’s a benefit to me, but there is obviously those other opportunities out there where, you know, advertising companies that want to ingest lots of information about us to be able to do targeted sales, you know, maybe you can use an agent to be the one that’s making that transaction, you say, okay, fine, they can have my data, this is how much I want for it and I receive FET in return for actually, you know, selling data to those third parties for that purpose.

Humayun Sheikh  32:16

On offering work, you could certainly do something like that. We are trying to veer away from that kind of model, because I don’t feel that this whole data economy, I think it’s a false economy. I give you an example, we make a lot of noise about, you know, we don’t want to share this data with, you know, big corporates, right? Big corporates are giving us a facility. Now take that facility away. That’s the value of the data. So, what are we going to do with the data, we’re going to sell it? We’re going to earn how much from it I mean, you know, how’s the valuation going to be done? Because this data is not useful unless it’s in millions and millions of data points. So, you’re giving your data? I mean, you’re not going to earn that much from it and you’re going to lose the service. Is it really worth it? I mean, yes, in medical stuff, that’s a different discussion. I agree with you, [Inaudible 33:21], you know, the drug discovery, drug companies and all of that. That’s okay, but that’s a lot more contained, but to get a service, and then saying, we’re going to charge you for the data. I don’t feel that’s going to fly.

Ben Sheppard  33:44

Yeah, I mean, I agree, I don’t think it works in every industry. I think, for me, it’s specific use cases where that has value and the healthcare industry, is one of them because of the nature and the value of that data, but for sure, I mean, I love convenience, I don’t want to lose out on convenience, because I wanted to get two euros a month for [Inaudible 34:08].

Humayun Sheikh  34:11

You can get in the car, and you can set the Google map and it knows your location, and he can fix it for you. Rather than go and buy another TomTom from the old fashioned, keep upgrading it every six months. You know, that’s why I don’t think there was a lot of talk about, you know, I want to hold my own data. I want to give it to you for machine learning, but I don’t want you to own it. That’s all. You know, I’ve got my question marks on it. I’m not a fan of that approach, because I think what is more fundamental, which we need to change is how do we learn from that data? How much data do we actually need to learn? Because I think if you Look at the prediction models, which work the best are where there is an actionable prediction, you know what action you took, at the moment, it’s all about, you know, just somebody saying something or somebody, you know, kind of doing their action online, but online actions are very different to the actual, physical actions. I think that’s why Google does so well because it has your, you know, all the actions and it gives you actionable predictions. That’s the model we are taking and saying, okay, you don’t need huge amounts of data, you can actually train your agent, which is kind of data that you can train it, and you can then get a service out of it, because what it’s doing is what all the big corporates do in a big data center, you can actually train your agent, which doesn’t need that mistake to that, and it could do the service you love, it can still deliver you that without actually giving everything to the big corporates, and you know, I’m not discounting the facility they provide, but it’s just becoming too much. I mean, it’s only like three companies in the world who just know everything about you. Then you can actually even enter that game.

Ben Sheppard  36:19

Yeah, and then there’s the misuse of the data when that happens because of that as well and you’ve only got to watch programs like social dilemma on Netflix, and it’s about four nights, so they should have that out on the 31st of October, Halloween, it would have been a good one.

Humayun Sheikh  36:40

I mean, we could see from all the past, you know, how the data has been used for people to change their opinions and everything. So, that’s, you know, I’m not condoning that. I’m not saying that’s right. I’m just saying that this whole, you know, we’re going to have this data marketplace. I think that’s old. That’s all we need to move to the new. I think we’re already kind of data marketplace is probably, you know, we want the models the marketplace for machine learning and AI rather than the data and we need to stop learning just from the historic data we need to start teaching AI, you know, we need to take that step forward.

Ben Sheppard  37:26

Yeah, I understand what you’re saying, and I think we don’t want to make the situation worse than it already is. So, we don’t want to go to the point where everyone’s locked their data down and then actually, no one can learn from it, because the economic impacts of that will be massive, it will slow development down so much and I think there is a genuine risk because of what has happened with the misuse of our data. It could go completely the other way and actually, we just made things worse now in terms of economic development, and that has impacts as well. Whether, you know, we’re obviously in the business of data marketplaces in TX and Streamr and I understand your point around, you know, are they the right tool for now or is it something else I see is a combination of things still.

Humayun Sheikh  38:17

I don’t disagree with you. Sorry. I mean, what I’m saying is, we need to look forward, but that doesn’t mean the data marketplaces of today are useless. I’m saying we need to use them, but we need to come up with new techniques to make an improvement. So, you have the data you learn from it, but now you have to keep teaching it further, not just by data, but by also creating these different models and different methods of actually making use of the machine learning and AI techniques which are coming out now.

Ben Sheppard  38:48

Yeah, definitely, and I think, from the EU as well, we’re seeing a lot of grant money coming out at the moment for different ways of creating marketplaces for AI models as well. So, how to build these models, get them into marketplaces, so that people can start accessing them. So, not just data marketplaces, but AI marketplaces and then how these tools could even work together in some instances?

Humayun Sheikh  39:16

Also, deployment. I mean, so you’ve got an AI model. How are we deploying it? Where do we deploy it? What benefit do we get out of it? You know, for example, and I think we’ve discussed it before, let’s say you have this taking example of COVID. Let’s say you got an x-ray, you know, model which detects something is wrong with the lungs and detects the COVID issue. So, let’s say you’ve got 20 or 30 hospitals, they have to train a model. Now they can all train their own models and then do this whole cumbersome thing of trying to put that together, or they could train the whole model as one, they can train it together without giving your data away, there’s no need to give the data, you can still train the model. That’s where the collective learning side of things come in. It’s collective learning where there is no. So, there is a little bit of difference between federated and collective, we use collective. There’s federated where there’s a single entity, which kind of says, okay, you know, everybody’s training, but we’re gonna upload it into one place. Now, in collective, you’re more like the model is spread, and everybody trains it belongs to everybody.

So, collective learning, what you could do is you could actually train the model, but how do you then deploy it, so you ideally want, if one region has more cases, and you want to learn from that, the other regions could actually query that model and get a result back. Even if they’re not training the model, you know, not necessarily how to train it. So, you know, that’s really where I was kind of going towards, which is, we need to have that fabric where you can deploy such a solution. You know, and that’s where, again, Blockchain is very good tool, to do that in an open system, rather than a, you know, I run my server, I’ve got a business which protects all of this and allows people to use, it’s more we should evolve towards collective learning models, because that’s when it becomes is the same effect. So, if you think about what’s happening in Defi and blockchain, it’s a collective learning model, people write code, other people use it, write more code, the other people use it and, you know, you’re kind of open source, we need to do the same with AI, which is, you know, we need to say, here’s the models, everybody can use them, because it’s an exponential increase, if you allow everybody to use them, and train them, but it has to be the right mechanism. So, it doesn’t get, you know, some parties don’t take undue advantage of it.

Ben Sheppard  42:06

Yeah, that’s actually where my next question was gonna lead to it because I was imagining this scenario where you’ve got several car companies that say, okay, we’re finally going to open up and allow each other to learn from it, it’s not gonna happen in our lifetime. Hypothetically, let’s assume a bunch of car companies actually decide to open up and say, right, we’re going to share some information, so that we can work together to build a new AI model, that’s going to help us solve some of the problems to do with connected autonomous vehicles. My question was going to be who owns that model, when data from multiple companies has been ingested to build the model? Is the model owned by everyone that participated or is it owned by the company that builds it and then they give a license out to the other users or how do you see that actually working in practice?

Humayun Sheikh  43:06

So, the model is a live evolving thing. So, let’s think of it as a live evolving thing and let’s say 10 companies are going to train this model. So, again, this is not what the companies will do, this is what I think we should do, it should not be owned by anybody. You know, the flip side is it should be owned by all the 10 people which have trained it, right? So, you could take both approaches, why I don’t think it should be owned by 10 companies is because if we want innovation really to take, it’s kind of the accelerated view, which is happening in the Defi in the blockchain space, then you need it to be open. So, people can actually use that model. You know, they can then build something better on top of it, because you know, maybe all these corporates can’t build it, maybe somebody who has gone a lot low resource can actually build a better thing on it. So, we are suppressing innovation by not making it open. Yeah, and I think that’s what we should do. So, especially like things like traffic models. I mean, you know, if you have 10 car companies who have so many cars on the road, and they can all feed into this model, where the model can learn, you know, we don’t just need Google or Apple or, you know, two or three companies to have that ability. We can all have that ability, maybe we can do something more with it, maybe, you know, it could affect people’s lives better. I mean, you could do dynamic taxation, if there’s congestion somewhere you could dynamically impose taxation. Again, this is you know, coming out of this model, which belongs to everybody, or nobody, you know.

Ben Sheppard  44:59

Traffic model is a great example I used to own a company that did traffic modeling, and you know, the software that’s used and the way that the data is captured for traffic models by doing physical counts on the side of the road and stopping vehicles to interview. I mean, in this day and age is just insane. It should be coming directly from the car companies, from the mobile operators, but I don’t know if you saw it, but we had Llewellyn from Oxfordshire City Council at Oxfordshire County council and he was commenting on this that he loved the access to this data, but the reality is it’s too expensive to acquire. Again, that’s the issue.

Humayun Sheikh  45:44

I think about it, if it’s too expensive to acquire, what I’m saying is if all these cars are feeding into this model, and you’re training this model continually, you know, it’s the model is gaining value, and everybody can gain value from it, right? You could say, I mean, there’s different ways of doing this. I mean, we do this in crypto all the time the project comes along, there’s a token, you know, the value of the token improves. Well, here’s the model. We tokenized it. If you’re really that interested in it.

Ben Sheppard  46:17

Yeah. It’s a completely different way of looking at the transaction. It really turns it on its head.

Humayun Sheikh  46:22

So, it’s costly, yet. It’s not costly, because it’s costly, when it’s trained, and it’s developed to that level, but it’s not costly when you’re starting. So, if you really want to be part of it, invest in it, and take a chunk of it.

Ben Sheppard  46:35

Yeah. This is where the challenge of educating the community [Crosstalk46:40], this is even possible, isn’t it?

Humayun Sheikh  46:47

I am trying to convince the Council to give a talk now stake in a model that will be interesting.

Ben Sheppard  46:54

Well, Llewellyn is one of the more forward-thinking people in a council that I’ve ever met, you know, he might actually, you know, be quite interested in that, but whether you then get the EMNOs, you know, the telcos of this world to sign up in the car companies to sign up to this, that I think that would actually be the trickier.

Humayun Sheikh  47:11

I don’t think that’s necessary. To be fair, I mean, the car companies don’t need to sign up to, you know, the individual users need to sign up to. So, if there’s a county council, who can convince or bring people to carry this, you know, we all are carrying apps all the time, the biggest sensor boxes, your phone, if people find easy to have an agent, which can train the model without giving any data. You know, it’s the individual consumer who needs to do this not the car companies. We don’t need to go there.

Ben Sheppard  47:49

I know what our next pod is gonna be, we’re gonna get Llewellyn back on here. We’re gonna work out the use case, and we’re going to do a project together, I’m gonna get him back on.

Humayun Sheikh  47:58

That will be a good conversation. We can have together I mean, again, you know, we can have a County Council who can convince it, you know, constitutes to actually carry it out, which trains this model, they can own the model. They can sell the model. We can all on the model. You know, that’s the new way of looking at it.

Ben Sheppard  48:22

Yeah, now, I totally agree. Humayun, it’s been great having you on the pod. This has been a really interesting conversation.

Humayun Sheikh  48:31

It’s been a pleasure. You’ve been great. It’s always fun to speak to you. It’s a thought-provoking exercise.

Ben Sheppard  48:41

Indeed. Okay. Well keep in touch, and no doubt we’ll have you on the pod again. Yeah, thank you very much.

48:50

If you enjoyed listening, don’t forget to rate, and subscribe to our podcast and please do share your feedback with us. Thanks for joining the TX podcast.