Sovereign Agents - your own personal AI assistant? | The Crypto Mile

On this week's episode of The Crypto Mile, our host Brian McGleenon discusses the development of sovereign AI agents with our guest, Jamie Burke. Jamie is the founder of Outlier Ventures, the world's leading Web3 accelerator by the volume of investments. As we navigate the online world, which is rapidly being transformed by artificial intelligence, each one of us could soon be accompanied by a 'sovereign agent'. These AI bots would represent us in online exchanges and have access to our private data and preferences, automating mundane tasks and executing orders on our behalf, such as purchasing goods and services or even signing legal documents. AI agents could also be used by corporations for financial trading, supply chain management, and ratifying legal agreements. However, what if these bots go rogue, and could there be an equal amount of malicious agents? We discuss how the topography of the internet is changing, a place where complex algorithms have agency to affect decisions in the material world, welcome to the 'The Agent-verse'.

Video transcript

BRIAN MCGLEENON: On this week's episode of "The Crypto Mile," we are speaking to Jamie Burke, founder of Outlier Ventures, the world's number one Web3 accelerator by volume of investments. Now, today we're going to discuss sovereign agents, and how in the not too distant future, each one of us could have one of these AI bots acting on our behalf in the online world that's going to be supposedly transformed by artificial intelligence. So if you're now picturing Agent Smith from "The Matrix," well, you're not far wrong. Jamie, welcome to "The Crypto Mile."

JAMIE BURKE: Thanks for having me on.

BRIAN MCGLEENON: Now, Jimmy, why sovereign agents, and in the future, is everyone going to have one?

JAMIE BURKE: Yeah, so I think most people already intuitively know that you can't really trust anything online anymore. And of course, with advances in AI, generative AI and deepfake technology, that's just going to get harder and harder and harder. So proving you're a person, for example, is going to be even harder, knowing who you're interacting with.

So as platforms advance their AI, and as you have malicious actors also kind of tooling up with all these advancements in AI, we kind of have this AI war. Our argument is that you need, individuals need, sovereign agents to represent their interests, to provably represent their interests as they navigate the web.

BRIAN MCGLEENON: So a sovereign agent here is an AI bot that you own. How would I own this AI bot? Is there a blockchain element? Is there a Web3 element to this?

JAMIE BURKE: Yeah, absolutely. So if you think functionally what you need a sovereign agent to be, a sovereign agent only really needs to do two things well. The first thing is that it needs to be a kind of conversational agent, so it will be some kind of large language model, ideally where you can even verbally communicate with it in a way that you do with Amazon's Alexa.

The second thing is it needs to know everything about you, or everything you want it to know about you. And so imagine if you knew that an AI would only serve your interests. You could have controls on what it did and didn't do, what it shared your data with. Then you'd probably feel quite comfortable in just having it record everything for this kind of ultimate form of personalization.

And in doing so, when you land at a website or you're kind of interacting with another agent, it can effectively represent you. It can represent your interests. So an example could be, in the world of dating, I'm kind of making a prediction that probably in less than 12 to 18 months, increasingly, people will be using agents, bots, to go on something like Tinder, represent them, because it knows generally how they will kind of communicate and interact in that kind of circumstance, and try and get you a date for Thursday evening.

BRIAN MCGLEENON: OK.

JAMIE BURKE: And that's an example in a social context, but it can do that in a business context or even kind of in a transactional context. And maybe we kind of get to that a bit later.

BRIAN MCGLEENON: So these sovereign agents can act on your behalf in multiple sectors so that there can be one sorting out your dating life. There can be one sorting out your business. Is this linked to like your wallet? Is it link to your bank account? Is it secure as a bank account?

JAMIE BURKE: Yeah. And I think this is where you end up needing some kind of economic primitives. You need some kind of financial system, digital financial system, that can enable these agents to carry out economic activity. So one thing is being able to have a conversation with a particular agent. Perhaps it can go and retrieve some information from you, like ChatGPT does.

But really, you want it to have agency to do the final mile to be able to complete a transaction. So for example, you want to book a family holiday to Japan this summer in June. You have a 9-year-old daughter. This agent knows all of that kind of criteria, and it will not just produce an itinerary like ChatGPT would if you kind of gave it that information, but it could then actually carry out the transactions itself.

It would go and book you the best flights. It would make sure you're seated together. It would know the dietary requirements of your daughter. And even more than that, if something changed, your flight got canceled, it would dynamically correct course. It would book you a new flight. I would maybe get you the money back from the airline. And the rest of that kind of value chain, it would kind of sell your hotel room for you for the night and maybe book you a different one.

BRIAN MCGLEENON: So this agent would need to presumably be communicating with other agents that represent maybe a corporation or other private bodies around this-- I heard you mention Agentverse?

JAMIE BURKE: Yeah, exactly. So the thing is, if you want an experience as close to generalized intelligence as possible, i.e. You could ask an agent anything on any subject, and it would effectively be able to come back with highly specialized knowledge, to kind achieve anything like that in a monolithic structure, i.e. Through a product of OpenAI, that platform would have to know everything about anything. And therefore, it would require highly specialized AI models. And to do that, you would need highly specialized data sets.

So most things, again, like ChatGPT, are trained on open web data. It's very generalized data that's freely available. But increasingly, if you want it to be able to solve for any problem, then it would need to be able to tap into lots of specialized data sets, many of which would be proprietary.

But a generalized experience through an app owned by a platform, that platform would somehow have to convince all of these proprietary data owners to give them access to those data sets, which we know is highly unlikely. What's more likely is, if we kind of leverage what are called agent-based systems, you can have billions, trillions of highly specialized agents that will just be a deep expert at a very, very particular thing.

And so when I go to my sovereign agent, that could just be a platforms agent, by the way. It might not have full sovereignty. And I know Meta's already talking about sovereign agents which aren't full sovereignty for the user, but really a kind of a platform agent.

BRIAN MCGLEENON: OK.

JAMIE BURKE: It can then effectively outsource tasks to more specialist agents. And somehow, in this marketplace of agents, you'll get back this complete itinerary. There are already technologies coming to market in Web3 that enable this to happen in a decentralized way, whereby effectively, yes, a company could create a series of specialist agents that represent its domain expertise in a certain industry, and it could make those agents available in a marketplace.

And whenever I had a requirement for something, my AI would effectively outsource that task to these network of specialist agents, and presumably remunerate them, right. And this kind of then begs the question, well, how do you do that in a decentralized way? How can you have trillions of micro-transactions being carried out on a kind of singular platform? It's not going to happen, and that that's why for us, all roads lead to Web3, and especially with AI.

BRIAN MCGLEENON: Right, OK. So in the future, we're going to have like micropayments. These agents will be presumably deployed on the blockchain?

JAMIE BURKE: Exactly, yeah. So they might not necessarily be carrying out all their transactions say on Ethereum Mainnet. There'll be lots of different solutions that will allow them to do that in kind of more economically viable environments. So for example, we have one investment called Fetch.ai. They have something called the Agentverse.

This is an environment where any independent developer or corporation can develop an agent, train it, and deploy it. Within that agent, there'll be certain Ts and Cs, certain parameters. When it's used, there'll be some royalty payment, maybe, to whoever trained the AI model. There'll be kind of a direct payment to whoever owns that particular agent, percentage of a transaction or something like that.

BRIAN MCGLEENON: So lots of micropayments. So do you think in the future, everyone's going to have an agent? It could be as ubiquitous, as say, the amount of people that have websites?

JAMIE BURKE: Yeah, exactly. I think that's a really good way to think of it.

BRIAN MCGLEENON: Now, will we not seen nation-states and probably big blocs like the EU coming into this space and saying, we want to develop digital IDs for everyone, every citizen in the EU, but these digital IDs are actually going to be your sovereign agent, and we will control it for you on our blockchain? Is that a foreseeable thing?

JAMIE BURKE: Yes. So to realize the true potential of a sovereign agent, you do need a kind of regulatory-compliant form of ID. So you actually need two things. Firstly, you need a decentralized form of identity. So it's ultimately an identity I own and control. I give permissions to who can see what. You know, maybe my permissions are, certain agents can interact with it and certain agents can't. Certain agents could see certain things about myself, and others can't.

But ultimately, that has to be anchored in a real-world identity. That somehow needs to be linked to how you're identified as a citizen of a nation-state, of a particular country. Your sovereign agent won't be controlled by a government, I would imagine. It would just link to some form of government identification, and it would be able to verify certain things about you without revealing the underlying data.

So it would know that my agent can interact. It could book me a table in a restaurant where it's serving alcohol, because it knows that I'm over 18, but it doesn't know my date of birth. So ideally, you would be leaking as little value as possible in any transaction.

BRIAN MCGLEENON: OK, so to book a flight, your agent would have access to your passport details and stuff like that.

JAMIE BURKE: Exactly.

BRIAN MCGLEENON: So if we're looking into the future, conceivably, like these agents could have some sort of legal weight. So they could maybe sign a contract on your behalf. But there might be risks to this as well.

JAMIE BURKE: Yes, so it's a really good point around personhood. So how do you extend personhood to a sovereign agent? And there's a lot of stuff that's actually quite analogous anyway. So the interesting thing is, in a US context, for example, companies have personhood, believe it or not. They are regarded as legal persons as much as you or I, as people are.

So there's already concepts of things that aren't necessarily people that have personhood, that can engage in forming legal contracts. And yeah, you're absolutely right. If you think about as a user, well, I don't want to have to worry about renewing my insurance policy, that's a great use case for my agent.

You know, your policy is about to need to be renewed. Go find me a better deal and just auto-contract. But anybody that has gone through the pain of renewing insurance or even a new mobile phone contract, it's just lengthy papers and disclaimers. So you would need your agent to say, I understand the terms and conditions. I enter into this contract.

BRIAN MCGLEENON: OK, so it could conceivably-- you can automate it to go and fulfill forms to contractual obligations on your behalf. Are we saying it's kind of the convergence of your wallet, your mobile phone, maybe all your certificates from your qualifications that you've done in the past, all along to one digital thing that just moves like a smart wallet?

JAMIE BURKE: It's a great way of thinking about it. Effectively, you're right. You can combine not just identity and socialness, you know, the ability to represent yourself in a social context, but also your legal personhood. You're kind of extending that to this agent. So you could think of it as a smart wallet, like a really smart wallet, or a really smart contract, in the context of Web3. It's kind of a smart contract that is on a chain and is carrying out economic activity, or legal contractual activity, on your behalf.