The Autonomous Altruist: Agentic Web3 and the Future of Public Good | Laurence Kirk | ETHWarsaw [4]

Hey everyone, thank you for coming to my talk. Welcome to day two. Thank you for inviting me. It's always a pleasure to be in Warsaw. Uh so I'm talking going to talk today about uh web 3 web 3 future of public good.

When I started out in web 3 a long time ago, I was very much focused on the idea of public good and public good projects. I perhaps naively thought that this would be a focus uh as the the time went by. Uh fortunately, we tended not to be. We've ended up more with meme coins uh than public good. But I think there's a change coming anyway now.

Certainly with the use of large language models and aentic web 3, I think this is going to have quite an impact on web 3. And so I want to explore uh what the the role is, how this is going to impact public good projects. If you would like to follow along, these slides are available here. You can you scan the code. Um we have lots of resources available for you as well.

We have courses about agentic web 3 at our academy. So please make use of those. Okay. So I want to yeah explore how autonomous agents can help with collective problems. Uh as I say I think there is a paradigm shift uh going to be happening with the way that web 3 is being used because of the use of autonomous agents.

Of course there is something uh of a sliding scale here. We already have a lot of uh automatic uh interactions anyway. Um, and I think it's going to be particularly interesting when and if we see true autonomy with agents. Now, I'm sure I don't have to remind you of some of the the fundamentals about web 3, but these obviously play into the kind of projects that we are we are building and the types of behavior that we'll uh get from those. And certainly the the focus now has moved.

We've got much more focus now on things like DeFi. I think it's interesting thinking about Dows. Uh again, you know, when I started out, these were an exciting concept and there we had the idea of the autonomy there. The idea that there would be rules that would be governing the way that a company worked and uh well with the governance of projects. uh it had a setback when we had the DAO and they kind of got a bad name and but they they've continued and I think we're obviously seeing them being they're certainly involved much more now around things like governance of DeFi.

So it's going to be interesting to see how these will be impacting as well with the agentic side. So uh yeah the idea of autonomous agents um of course we do have already many automatic processes happening um you know transactions being done automatically without uh without human intervention but there's perhaps more to autonomy than that. I also used to I had this exciting idea um yeah many years ago the idea of uh things being completely autonomous. the idea that you could have something like a a self-driving taxi that could uh run independently. Uh it could make money for itself.

It could store its own crypto and then decide itself what it wanted to do with that or without human intervention. And with uh the the complexity that we get now with uh Aentic Web 3 and with the the AI side, I think we do have the uh ability to have more autonomy and I think it's going to be a very exciting time. Of course, we've evolved from very simple systems to things like uh large language models which are much more complex and do they can do much more complex behavior, much more sophisticated behavior and make much more sophisticated decisions. But they do have limitations. Um, as I'm sure if you've uh worked with something like a large language model for a while, you'll understand that the the context uh that they they can work with can be limited.

And this can can produce problems as you add more and more context with your interaction with them. And of course, it's the the idea of the ethical reasoning. They you know there's there's nothing there to give them any ethical background unless that is built in. And then that is part of the the training that they receive and the data that they are built upon. So yeah, these intersect uh so we're now seeing lots of uh agents on blockchain and uh these uh are are producing their their own uh transactions and as I say it's interesting going to be see interesting to see how much how autonomous these things can uh truly be in in theory they can be.

I think it's going to be very interesting when we see agents that uh are making their own crypto and then they can decide to host themselves. They can use the the crypto they've uh they've earned to uh supply their own hosting and then run completely independently of humans. It's going to be an interesting time. Uh but I don't think the future is going to be agents uh running completely independently of each other. There's going to be multi-agent systems.

we're going to get the idea of um specialization within agents. Um but I'm going to come on to this a little bit uh later. But it's important I think yeah to the idea that we could have these autonomous actors that actually have some interest. They have some skin in the game. So thinking about public goods, what kind of things do I mean?

Well, there are, you know, obviously lots of different types of projects. Um there's this this idea of the non-rivalous non-excludable resources, the things where we can uh take part in these projects, but by individually taking part, we're not taking things from other people. And uh one of the advantages of blockchain and the fact that we can have tokenization is that we can then help to solve some of the problems we get things with things like tragedy of the commons and uh we can use tokenization to incentivize that and to uh make sure that there's uh not concentration uh and and resources not being concentrated just for uh a few people. I'll come on to the uh environmental monitoring example of that in a second. But there's lots of uh topics, lots of projects in this area that could be very interesting.

Public health obviously is uh uh very potentially very good. It's always been problematic because of the privacy aspect. We have to be very careful about having any kind of health data any patients data on the uh on any kind of uh blockchain. Uh so we do need privacy to help with that and that obviously needs to be uh secure but al there's all sorts of legal and compliance uh issues around that as well. Uh as I say yeah an example of uh environmental uh uh monitoring this thing from robomics where they've got environmental uh sensors to to monitor the uh yeah air quality in this case.

So you know we can build these projects we can do things that are designed for public good and blockchain certainly has a role in that. Um regenerative finance became quite uh yeah interesting topic perhaps a few years ago. Um and perhaps from there we will get you know that will help bring the some of the DeFi aspects and help bring along the ideas of uh public goods as well. Of course, when we are doing this, we we need to understand whether these uh projects, these public goods are actually providing good. And so we do need things like oracles to to verify this.

We need some kind of metrics. Um and this aspect is one that I think is potentially a little bit problematic with the AI as well. I'll explain that in a second. Now, a lot of frameworks are available. If you want to build agents, there a lot of frameworks are available to help you with this.

Uh, I've just got, yeah, some some here. There are other ones as well. If you're interested in in building agencies, as I say, if you go to our academy website, we've got lots of resources there that can help you, but there are big players getting involved now. So, we've got uh, you know, people like Google who have their agent dev kit uh that that you can use then to to build agency. Not necessarily focused on web 3, but you know, you can use it for that as well.

And there is the more general protocol the MCP um that is there to standardize uh interactions between agents and systems. So uh if we're doing this what are the the challenges? What do we have to be careful about? Uh so scalability is always a problem in blockchain. I don't know if I need to say much much more about that.

Um the Oracle problem uh we do need yeah to to connect things on chain to get uh off-chain data and that needs to be secure um and and reliable and uh this this can be yeah this can be problematic for uh the agentic side as well u making sure that they are acting on reliable data. Um, of course, they're not going to act in isolation. They're going to be acting across different protocols. So, we need to be have the ability to do that. One of the most important things, uh, I think certainly is the, uh, the security.

Um, obviously I' I've, you know, my company is an auditing company as well. We've been involved in security and so we've seen really that you know over over the years that security hasn't is still quite a major problem uh in in web 3 and certainly this is just going to increase um this the uh having a gentic web 3 is an additional uh really attack surface there and privacy as well uh it's interesting um I often give talks about ZK and the privacy aspect uh and it's uh For me, that's that's a really important thing. And maybe for a lot of people in web 3, that's super important as well. But I think that maybe we're in a little bit of a bubble and for a lot of people, they're not so concerned about that. But certainly uh if the if we're doing things with public good and as the example I mentioned with, you know, uh healthcare, this is obviously is a massive uh massive potentially a massive problem if any of patient data leaks.

Uh and then looking at the you know the nontechnical aspects we've got challenges on the the ethical and regulatory side. Um so who controls the agent? How do we do governance? Is it going to be are there going to be humans uh within the loop to to do this? And as a society are we ready for this?

And I think the certainly the answer to that at the moment is no. people are only just beginning to start to think about the impact of AI and the kinds of decisions that uh are being made and how that is going to fit into the established legal and compliance frameworks that we have and that always takes time. We're always going to be a little bit ahead on the technological side from uh where we are on the the compliance and the the legal side. One um one point as well that uh is problematic the the causal I uh point I have here is the fact that uh it's often difficult for the uh current AI to understand uh sort of cause and effect and understand uh what may have started some behavior and what would be the impact of a certain type of behavior and this I think is important for public good projects because that is going to uh we need to understand that we the the the the goods part of it, the good part of it is having the right kind of impact and um I was given an example that with with charities this can be uh sometimes the sort of regulators have to come into charities because the uh there's a concentration um there are too many trends and certain charities don't get the the funding that they need just because certain things are super popular. Um so it's that kind of thing is very difficult for AI to understand and that is something that we need uh help with on the on the research side and then the the data biases of course with AI we have the potential problem that the the type of decisions that are being made well of course they're very much dependent on the the data that they are trained with but then how do you take care that that data doesn't have biases so all of these are are problems uh which moves in then to Yeah, the accountability aspect as well.

If you have an agent making a decision, it does something in a project, who is responsible for that? If it's acting at all autonomously, uh presumably the agent is responsible, but then you know how can we build that into our framework that we have of our sort of, you know, traditional legal and compliance framework. Um so uh what kind of things do we need to do to to get a a better kind of architecture to how can we build something and and promote something that is going to work and help agents do something uh in public good? Well, I think initially we've got humans in the loop that uh that is obviously going to be true and they're going to be needed for critical decisions with maybe agents doing the lower risk decisions but then that's not autonomous. So over time we're potentially going to move away from that.

But there are some uh projects, there are some things we need that will help with this things like decentralized identity. Um we've got a lot uh of standards around that. Um so potentially that is something that that we could have also very interesting um the idea of reputation systems. Um these maybe have a a mixed reputation as a as an idea. People sometimes don't like the idea of introducing a reputation system into a project thinking that shows a failure in the project.

But the the way that this could work with on the agentic side is that agents could then see each other's reputations and decide to interact with each other based on some type of reputation. But of course, if you do that, then how how do you you judge the reputation? And this comes back to the idea of having reliable oracles and reliable metrics that agents can can pick up on. Uh an interesting idea also the the intent uh centric architecture is the idea that we could have some uh behavior that is where we specify a high level intent. We we say we want the the agent to do something at a very high level and then we have specialized agents that are solvers that go in and decide how that can be achieved.

uh and I think you know this uh this term I've I've come up with here computational philanthropic ethics uh is a a new uh idea that I think we need to research into. We need to understand how autonomous agents can act in potentially an ethical way. So um I haven't got much time but I'll just finish off with some uh further questions. Uh can agents generally be altruistic? Will they be altruistic?

And if so, what form is that going to take? We tend to think of the altruism being towards humans, but why would that necessarily be the case? And if they're autonomous agents, would they be altruistic to to humans or altruistic to other agents? And is there a difference anyway? Uh, you know, how do how would we deal with that?

Um, and the part of this could be dependent upon the way that the agents are trained and the the AI is trained. And so uh as a as a society, but certainly as as developers and developers of these types of systems, the way that we train them and the data sets that we we use is going to influence this and potentially we can influence this. We can influence the way that agents and that AI is going to make decisions by the data that they are provided with. So I I think that is um an area that that perhaps is not I'm not sure how much is recognized at the moment but I think could potentially be quite useful and there could be uh almost races on to to get data to be picked up and used uh for training uh to influence uh AI and to influence the types of decisions that they make. And the question you know is is this any kind of altruism a likely outcome um unless we we do something to develop it.

And I think the answer to that is no. I mean the agents are going to be completely neutral. Um and so if we want any kind of altruistic behavior and then or any particular behavior in any direction really then as uh developers uh as builders of systems we need to be building that into the the system. So developers have a a very uh large role there. So where are we going with this?

I think uh in the near term yeah we're going to have humans in the loop. they're going to be providing uh oversight of the way that agents work. Moving on, uh we'll get more autonomy. So, we'll have uh agents uh systems of agents that are are self- sustaining. And then hopefully in the long term, uh I uh there will be autonomous agents that can help make the the public good projects more efficient and can help them to deliver their their public goods.

But in order to do that, as I say, we do need to to build this in. So it needs to be built in and the idea of public goods needs to be built in uh as a uh a first principle and not just an afterthought. So yeah, we we're at a very exciting time. There's lots of uh challenges uh in the way that uh the agentic uh sphere is is moving and how this is going to develop. Um it is evolving very rapidly.

Uh I it's going to be very important how we develop incentives. Uh one one idea that that we are working on is the idea or on the security side of incentivizing agents to uh be careful about security and to look for security problems again. But that's you know just on the security side. You could do this uh in any kind of incentive. Uh but how this develops at the moment this is very much up to us.

Uh I'm I'm thinking particularly of the the developers but obviously in the future as they become more autonomous then maybe humans will be taken out of the loop. Okay. Uh that's that's everything from me. Thank you very much for coming to my talk. I hope you enjoy the uh rest of the the day at the conference.

Thank you.

Do we have any questions? Oh, lots of questions. How exciting. Thanks.

Hi, this is Machi. You said that uh regarding this ethical questions of agents, they might be neutral. I might say that they would rather be efficient. Uh is this is probably the metric they would be act upon. And uh here the question you brought regarding acting for for community good um is actually philosophical question because uh we know how it works with people.

People are driven by different values be it religion be constitution be it uh values coming from literature you know from role models and so on but we don't know how it works with agents so far and from my perspective this is currently mainly efficiency and what do you think what um set of values or guard card rates can we set up in terms of somehow common um commandments whatever we call it. uh in order to not let those agents be uh ethical or rather um inbuilt somehow ethical constitution for them or commandments for them like there is aim of laws of robotics and that robots cannot harm humans and so on.

Yeah, thank you.

Thank thank you for the question. Uh yeah, I think that's that's that's quite true. uh certainly the way that people are developing agents they're going to be looking at probably efficiency but there is I think I said you know they're going to be neutral there is perhaps perhaps a background though with the way that uh AI things like large language models work that come from the data that they are trained on and so that potentially is going to influence the way that they they make decisions at the moment they are trained to you thinking of large language models they are trained to work in a certain way and to avoid certain situations and avoid acting and doing things. So there are some kind of ethics built in there. But this is very much dependent upon the the training and the the companies that produce these and there we have a great centralization risk.

We have very big companies who are deciding this and if they decide they want to train them in a different way um make them different then yeah there's a potential problem. But I I think you're quite right. I think a lot of the the development of agents is going to be focusing on efficiency. Uh and if we want ethics, then we're going to have to think of ways to build that in. One way is to to provide incentives, but how we can do that at the moment, I don't think is is 100% clear.

Thank you.

Awesome. Thank you. We have time for one more quick question. So there we go.

Thank you. Uh hi, I'm Nico from Octant. We've been uh funding public goods for two years using uh yield from Ethereum staking. So this is a very interesting uh talk on what the future of that will look like. Um so I was wondering if you've seen already any experiments in that direction that use AI to fund public goods.

Um not no not particularly not on the the autonomous agent side at the moment. I it's uh fairly fairly early days. I think the I think what the way that it's been picked up has tended people have tended to use these for things like DeFi and you know we're getting the bots doing lots of trading and that kind of thing and uh the public good side has perhaps been left behind a bit but I please do please do integrate them please.

Yeah I actually know a couple of them so I'll uh come to you and uh tell you about it later.

Thank you.

Thank you.

Amazing. Thank you very much. Florence has given one more large lands reports.