The Self-Driving Economy: How Blockchain and AI Agents Will Reshape Everything

Transcript

Hello everyone. Uh so we'll we'll stay on the same similar topic uh just slightly different um but we'll say on AI uh it seems to be the the dominant uh topic nowadays. Uh but we'll cover uh another topic called which are AI agents and how they intersect with crypto. and I'll bring my uh three panelists here now. Yes. So um it's not going to surprise anybody if uh I tell you that for the last three years basically since chat GPT has come about uh we uh AI has been LLM has been the the the one of the only topics that

are being talked about uh and uh crypto is not immune. Uh the first time I really uh uh saw that was only like a few weeks after the launch of chat GPT. uh in my activities as a uh as an investor um I do a bunch of those panels where I'm one of the panelists and one of like with only investors and uh crypto investors in particular and one of the question that is always you know customary to ask is uh what is a a brand new topic that you're looking at very intently and making investment at the moment uh and uh I'd say as soon as like December 2022 so only like two months after launchp everybody was saying uh oh we look at the intersection of crypto and AI. Uh and I was always a bit uh

puzzled u because I had to admit that I was not really seeing much deal flow. And when I was talking with other investors, they were like, "Yeah, we haven't seen much deal flow, but uh it's going to be something." Uh for me it took a very long time to really start seeing some interesting something interesting with crypto and AI and and that happened basically last summer uh when AI agents kind of came into the zeitgeist with uh Truth Terminal uh and uh with this interaction where Truth Terminal finally got a a uh a grant in uh in Bitcoin from Mark Andre. Um so we'll we'll cover those topics about what AI agents are kind of like this brand new new thing uh and uh a little bit later how actually crypto can uh can really enable uh and facilitate AI

agents but we'll start with uh my panelist um can you introduce uh each other and start with you Conrad? Sure. Hi everyone my name is Conrad Marino I work with Near Foundation as a head of BD and blockchain success. Super. My name is Benjamin. I work for Google and lead our web three efforts for the EMIA region but of course uh I also have to uh kind of cover AI topics and it's great to be here where at a time when these two topics are slowly merging and yeah I I'm here you maybe half a year ago when we spoke about it last time we didn't have so much to say but hopefully there's more interesting stuff coming out today. Hi everyone. Uh my name is Sham Teriswami. I'm uh

founder and CEO of Provably. Provably is building verifiable data infrastructure. Uh reason we're doing that is probably 99% of the data will always live offchain. Uh blockchains don't really handle big data. They don't really support data operations. Uh so we're building zero knowledge so that all the data that lives offchain has integrity and then when AI talks to that data it can prove uh that is uh giving you answers on authentic data and it can prove to you that any uh data answers are correct. Great. Uh so first we're going to start with um just looking at AI agents uh before going to the the intersection of crypto and AI agents. Um Benjamin let's let's start with you. Um right now there's a a lot of talk about

how AI agents are going to replace workers. Of course, there was this uh kind of uh big announcement by CLA that they were replacing all their customer support with uh AI agents what in February or March this year only a few weeks uh maybe a couple months later for the CEO to revert and say no actually we're rehiring everybody uh because AI agents are not good enough. Um of course um Google develops its own LLM uh model and um Google cloud is one of the leading infrastructure uh for um for LMS. How do how are is Google actually seeing uh AI agents used today?

Actually, what is the reality of AI agents today? Yeah, good good question. I mean, we spoke about it briefly beforehand that we have to define what an agent is because nowadays everything that used to be called an algorithm is now an agent and all of a sudden it's like super exciting. So even something as a currency converter, you might call it an agent because it can kind of give you the latest exchange rate. But when we look at agents, we have this definition that an agent needs to be able to think autonomously and that's when the LLMs come into play and to understand the surroundings and do its own reasoning. But also it needs certain tools to interact with this environment.

So only when these two kind of criteria are fulfilled, we would consider an application to be an agent. And if you look at agents from this definition, Google itself like with for our own products, we have a bit of a history on trying to get there and and provide these agents. If you look at in on the consumer side at what Google Home can do or the um Google Assistant on in the Android phones and I don't want to comment on the quality of these agents, but I think we have reached definitely like a new level now. And um one project that Google is working on is or like maybe two to highlight. One would be like an agent that can do research for you online to kind of do autonomous tasks in your browser kind of and you

can follow along and you can watch it which is an important element as well when it comes to understanding AI. And then our like main goal which we announced or reannounced also at the IO conference just a few weeks ago is called project Astra. And the idea is to build an agent that really understands our world, our surroundings and can be your like personal assistant along the way. like you can interact with it in multimodal way. So by showing him things or the agent I don't know if it's the genre it you can show it things and you can talk to it and you can uh also it has access to your personal like data if you want the context part. So I think this is what we are doing and then the last piece is of course if you want to use our infrastructure as you mentioned

and get access to our LLMs to build your own agent and we have like certain tools but in kind of what type of agents are we seeing yes support that's a big topic and for good reason because also working in support some of the tasks can be repetitive it might make sense for both sides to have an agent there but then it goes further it goes to creative agents that can create uh content and publish the content and moderate the content. But these are, I think, some general fields where we see agents being active. So, Conrad, um, LLMs are evolving very fast. I mean, we we're all surprised by uh by just the evolution. Every month we seem to see a new uh groundbreaking feature on LM. So, even though AI agents

might not be, you know, adopted and and and fully ready at the moment, um, where do you see this? We know that's going to evolve and and get better and better. Where do you see uh AI agents uh being used in let's say five years? Mhm. U well our vision at at at near is that uh we believe that AI agents will eat the world. So you can imagine every SAS business being completely disrupted by the agents that would not only run the operations of these businesses but also interface with the customer. So we believe that the the way we interface the internet is going to be dramatically changed in the next 5 to 10 years where

instead of having a static website you will have an AI agents to whom you'll be able to speak and communicate to get the information and uh realize the task that you wish and in order to do it um at near we believe that these agents needs to be autonomous right and um in order to facilitate it we believe that we need to power these agents with tools uh that would enable them this autonomy. So one of these uh very exciting use cases that we're building towards is enable them to transact and communicate between each other on our blockchain because I think everyone is kind of afraid of this dystopian futures where we will have no control over where our

data is going where these agents are sitting are there on some centralized cloud owned by Elon Musk or by our friends of Google etc. So we're building towards the infrastructure that is fully decentralized and where the the autonomous agents can really execute everyday's tasks. So coming back to your question around use cases, every human actions will be replaced by agents and we like to call not agents agents but assistants. So they will be the assistance for humanity. Um Shim the there is um so I I do talk with a lot of security company like uh like internet security company uh and uh there is a a fascination for them

because uh the way LLM work in particular and agents of course uh in as a derivative are a brand new way of hacking systems. Um do how do you feel um like what are the risks of the way agents are deployed today? Uh and maybe um what are some of the the easily avoidable mistakes that can be made um when deploying an AI agent today? Um I I so probably November last year uh we saw the release of uh MCP model context protocol right so I think that is uh it's completely uh you know changed the game I mean the talk today's about self-driving economy blockchain and AI

agents I think self uh automation an agentic web and that automation is starting to happen and now the question is like when does blockchain come in and where are the risks and what are we trying to protect Um but it just to kind of like level set MCP is uh basically allowing every application every database to become AI friendly. So an AI can talk to all type of endpoints that it couldn't talk to before. You know before it was APIs and there were restrictions to accessing those APIs. Now you just the AI can read the MCP and it can talk to it. And then there's Google working on agentto agent frameworks and then Microsoft is also backing NL web. So you know web 2 is going towards AI is basically stitching

the world together and as Conrad said it's eating the world. The thing is risks hasn't emerged so much. Uh it's there like it's filled with risks but it's us humans who are using it every day and we don't have the patience to do verification like we kind of verify by just checking on uh the internet. So things that are public can be verified. But I think now as underlying applications with data, our data becomes available and AI can reach private data. You you you can't do traditional verification. You know, you can't go on the internet and uh verify something.

You can because it's hidden. It's under the tree if if you like. It's in the roots. So now we're getting to a stage where um uh when AI performs actions on on private data uh this is where you really uh want to verify it and and the re the reason is simple traditional computing it's like machines like 1 plus 1 is two and you ask it a million times it'll always say two right um because they are deterministic uh AI or LLM models the way they work they are non-deterministic it can get confused easily it can hallucinate and uh it it just has a nature to invent answers or give you answers that look plausible. So then

when you think about lots of agents they are uh solving a problem together you can easily get into this Chinese whispers scenario where one accidentally passes the wrong information and it just goes on. Now if you're on near and you have a common shared ledger they can verify but what if that ledger doesn't exist or as I said earlier 99% of the world's data is offchain how do you how do you verify so that that is the problem that we're working on where when an AI says I have hotel rooms available I've made the reservation the payment is successful I've booked your car uh for your travel journey whatever every time it says that it can actually give you a

little zero knowledge proof that not a human doesn't verify it, another agent will verify it. So I think uh where security improves is where agents to in agent collaboration they start to verify each other and uh and also just the last thing is that I think we've also gone from this one-shot answers to more reasoning right it's really thinking about it and if you notice on chat GPT answers are taking longer uh and this is the default so actually they can verify each other and give accurate more reliable answers uh and these little cryptographic proofs will enable that to happen in web 2 and this is the game that I think crypto has to try and intercept and and be part of and be part

of that fabric somehow. Um let's talk so let's dig a bit a bit deeper into this idea of how uh blockchain can help AI agents. Um Conrad Nar is the the self-proclaimed blockchain for AI. uh could you uh I mean you've already touched upon a little bit but can you give us like some of the examples of uh of the tools that you you provide or how you can support the ecosystem uh the agent ecosystem? Sure. I I think we are not only self-proclaimed but we are building towards becoming the blockchain for AI and and fun fact is actually near story started at Google where our founder Ilia Sukin was leading one of the R&D departments there and co-wrote with other uh Google employees uh this famous paper attention is all you need that

laid out the foundation for uh transformers that are powering chat GPD and the generative AI but history aside I think currently we went through this journey where we built our blockchain infrastructure that is super scalable right so near is a um sharded blockchain that enables uh infinite scalability because we can add shards as the number of transaction grows because we believe that if the agents uh will take over the world they will not be able to uh open an account at uh UBS or credit Swiss or which doesn't exist anymore but they will they will need a wallet right to to make all these transactions. So we we

have built um the framework that enables communication and transactions on our blockchain for the agents that we called AATP. On top of this, we believe uh in interoperability. So um our agents that we already have on the on the blockchain and are powering some of our apps are able to swap not only uh different assets using the near blockchain but also doing uh crosschain transactions in multiple other blockchains. Right? So we're building the infrastructure to enable this economic layer for the uh for the AI agents. Nice. So maybe a follow-up question on this. Um there there's always those discussions about

how, you know, the the Terminator hypothesis, uh we're going to have Skynet uh that overtakes the world. Um it scares some people so much that they actually want to prevent uh agents or LMS to to be able to own things. Really want like a law that that that prescribes it. uh when you is near building the the tools for the uh for the the LLMs to uh to take over the world and uh and dominate the human human race. Well, I I think we're building the reverse. So, uh our our like product strategy is divided in the agent, right? So, we want them to be autonomous. We want them to basically serve you better, do transactions on your behalf, but also we're building the

infrastructure to keep them under control, right? So we're building a a trusted execution environment. So these agents live within the private cloud that users and developer control because currently in the current infrastructure we're relying on the web to infra to power all these agents right and this web to infra course is very robust very powerful um but uh it's centralized so um you don't know when you're using chat GPT where your data is going to to what purposes it's going to be used. Whereas with the private cloud environment that would be then potentially distributed on the blockchain uh there is some degree of privacy and protection that would like to ensure to to the users of these

agents. Great. Uh Benjamin since we talk about um infrastructure um you you work on the web three side of things at at Google cloud um and uh however even though we work on the web3 side Google is still a centralized uh company. Um can you give us a kind of a rundown uh of where you believe decentralization can improve uh AI agents um and and where it just doesn't cut it? Yeah, as you mentioned like we are in the business of like Google Cloud specifically is in the business of providing infrastructure for much what you can do in the web 2. But of course we're not going to stand still. So with these emerging technologies as like blockchain or agents, we also want to like help building those applications

and I don't even have to be here anymore because you already mentioned that agent to agent is a protocol that we have been co-developing with a whole alliance of like I think 50 or 60 organizations which will allow agents like to communicate with each other. that kind of sets the rules like similar to HTTP which allows like uh the internet communication to have a certain framework that also these agents need to have defined like rails and when it comes to to your question when where do we see these things come together from an infrastructure perspective Conrad already mentioned that agents might want to have access to wallet and payment systems so even when it's clearly defined how they can communicate with each other this still needs to

technically be enabled that agents can transact with each other or even reward each other for helping each other. And uh I think none of us would claim that there's only the one blockchain to rule them all. But we the one little piece that we are also adding to the puzzle is that we announced quite recently a ledger for crossber payments. It's called Google Cloud Universal Ledger. I'll be talking more about it tomorrow. If you're interested, come and and come to my talk. But one of the use cases that we see and I think why I'm bringing this up is that an organization as large as Google also believes in this use case as you described it that the traditional payment systems do not support what agents will need to be doing like for good or bad I don't want to judge but

currently to open a bank account is already can be a challenge and even more so if you are less human than maybe us. So I think this is something we as an organization also share this kind of idea that there needs blockchain can be a way to solve it. I think that's kind of a good way to conclude right the the you talked also about interoperability. It seems to be like a key uh key topic here. Um, how do you, so I guess you still see a a world with many different blockchains or do you see at some point a need to have like at least this part of the infrastructure uh run on a single blockchain?

I mean I can't tell you like I don't have an answer. If I had I would also be happy to share it. I think from generally how if you look how Google approaches things um we believe in choice like in many on on many levels and I think also for agents it's good to have choices like how they want to communicate but also how they want to transact with each other and to your point um I was just thinking when you were talked about how in our current world when we do human human interactions trust does play a role like often things happen based on recommendations um or your first impression of someone and of course there's all these elements like how they dress, where you meet them. But uh even when it comes to let's say less personal decisions, I was just thinking of this

example when you are out on vacation, you don't know the place and you stroll through like this uh street of restaurants. Okay, you can look at Google Maps and like ratings and hard facts, but if I see there's nobody else sitting in the restaurant, I might just keep walking and and don't end up there. And this is this like it's not measurable. Like it's just this like trust piece where your intuition comes to play. And I feel like for agents they will probably lack this piece and that's where we need zero knowledge proof and these like technologies to kind of compensate for that. And I can I can add something to that and I was just mentioning offstage that uh uh there is going to be a working group like on on what we're developing new zero knowledge that works on big data really fast small

proofs. it will be open source and there is a a a group forming uh with Google, Facebook, um whoever has data in which they can uh prove things to each other and a cryptographic a zero knowledge standard is going to emerge. It's just starting but it's it's not a coincidence that it's happening now and players like us from crypto should be part of that uh conversation to share what we've been doing and what we can bring to that so that when that new uh when it starts to get adoption uh probably you know who are the people working on trust technologies a lot of it is in crypto so we should be there to share what we're working on open source it uh and then of course Facebook Meta Netflix all these people are there and uh hopefully we

will see a common standard emerging in uh data exchange, right? Um uh yeah. Yeah. So, when it comes to to um uh data, um that's that's usually what we see with with companies that uh that pitch us uh is uh they are most of them build on the same LLM uh or a a bouquet of different LLMs. Uh however, what makes a difference is the quality of the data that they that they have that they provide. um how do you actually how can you actually support um those companies uh that provide this uh proprietary data or even companies that would like to work with NLM with their with their data? Uh how do can you actually uh help having an agent that's uh specialized or or has an edge uh because it has

proprietary data from a company or from a a group of people? So we we're building a single product which is like a verifiable data infrastructure that someone can deploy to the uh whether it's a blockchain or a database they can deploy it in about 5 minutes 5 to 10 minutes. Um the reason we're doing that is um that if you take a blockchain and you know everyone will say why why does the government need to uh why do we need to file our taxes physically? Why can't they just look onchain for example, right? The there's a very old and simple problem that exists in all of computing whether it's in a blockchain or anywhere else which is it's just very hard to prove something to somewhere else to someone far away. So for example today

you cannot go onchain and say give me my total profit and loss from last year. You cannot do this onchain. So what happens is data gets indexed offchain and that's where you do your data operations. So um now coming to so the point is that it's very difficult to prove something to somebody far away. It's a distributed systems problem and that's why when two blockchains talk to each other bridges often get hacked. That's the moment of attack like or in oracles or bridges. So what we're doing is allowing whether it's a blockchain or database to become like a ledger. It can prove uh when was that data created, where did it come from, uh how was it used, the provenence and the lineage. So that's one side uh and blockchains can

already do it but if you're a database databases can do it as well and then you can uh or AI or a human or whoever can uh run these data calculations and then they can uh pro give that to someone and there's a zk proof that they included all the data and the maths was done correctly right and and so this kind of like really gets to the problem of like even why do banks uh or whoever has data they're filing things uh but the audits are always manual because the person on the other side doesn't understand if they've done the maths correctly whether they've included all the relevant data and now for AI anytime AI does data operations it can uh provide a proof that let's say AI did its own research

in you know we're sitting in Rodro there's a lot of pharmaceutical companies here it ran its own uh uh bio uh research uh for something right for it it did some research it did some simulations It can prove that uh how it created that data. Where did it simulate it? How did it simulate it? And then it can reveal it all or or it can give those answers and show it did all the maths correctly. No hallucinations and all the lineage behind it with a simple proof that someone can verify in in milliseconds and they don't have to run the whole um verification all over again. they don't need access to all of that data and and that that way uh trust just becomes you know in milliseconds faster uh and and this is aimed for AI

agents doing these uh data calculations or just regular enterprises proving things to each other to finish um I'd like to consider just a more philosophical question about uh AI agents um and uh the first time I heard kind of a a funny story about AI agents was Olaf Collison from uh Polychain um who um multicoin sorry uh who um was talking about how he imagined a a day where the richest uh entity on earth was a uh broken down Tesla that could not on the side of the road that could not uh um drive anymore uh but started to uh trade crypto or trade uh in order to

just pay for uh g for electricity refill uh and ended becoming just like this uh independent non-human uh entity that's the the richest in the world. Um this to me opens kind of a debate as to do we think AI agents are going to be um like another amount of people like another group of entities another universe that lives next to us or is it or is it something that is always going to be more of an assistant because there's really this this vision as to complete independence and there is a world kind of like in her where the the your your agent falls in love with another agent or do we feel like this is not really something uh either uh that we wish to

happen uh that is desirable uh or just not possible and that a agents will always be just like servants to to humans. Yeah, I assume so. But yeah, these guys to be honest with you guys, I have no idea. I I I I think it's difficult to predict future. I believe that the space of uh learning by all these agents is phenomenal and in general by the progress of the artificial intelligence that sooner or later we will be at the stage of super intelligence by the AI in general right and if AI is super intelligent therefore the agents will be super intelligent and will regain all this autonomy which is scary and as a society we would need to

learn how to live with uh intelligence that even surpasses human intelligence and can live in parallel. So I think it's a big societal challenge that we'll face within the next 5 to 10 years. But I think to your point, I I think it's you're fully right like when we talk about agents today, these are still powered by LLM with like these prediction models that are designed to sound good and look good, but we haven't achieved AGI like uh like artificial general intelligence yet and it's still like a question if it will ever be possible. It's not like oh if you just wait two years or throw another like few years of research at it, we will get there. Maybe it is impossible to reach

like it has not been proven yet to your point we don't have a proof that it is possible to develop it and this I think the answer to your question depends on that fact if we will be able to develop it or if agents will be able to develop it then the world looks different to if we have the current version of agents that are designed for a specific purpose and are easily like easily controllable um to please humans and I think that's currently what an agent is a good agent uh would be doing I I would say I have no opinion on this. I mean a day-to-day problem is uh it doesn't have the memory like you can have a conversation with an AI agent in one chat and then when you go somewhere else it it can't even take its memory and

context to the next uh conversation. So I think uh we have a lot of work to do to for it to retain uh or have the shared context and then go from there to start thinking for itself and persisting that it it's it's it's a wild too early to tell. Okay. And so the the very last question do you think that the agent/crypto to to go back to my my first observation AI agent/crypto do you think it's an investable uh uh sector at the moment or are we still in the the kind of hype? I mean, we've seen kind of a hype on tokens uh late last year uh that has kind of died down and might pick up again in the next few months. Who knows?

Do you think it's actually an investable sector or are we kind of like are we in the same place where I don't know DeFi was in 2018 with ICOs or 2021 or today? Are you looking for financial advice? I'm I'm I'm looking to actually convince myself that I should make an investment in AI agents. I'll be super brief. I think we should not limit our observation to blockchain and AI like these are two technologies and there's another million others and we just today focus on these two but whatever will happen it will be a combination of several technologies maybe those two and maybe not like I don't know yeah I I I think you're you're right Benjamin I think both of these technologies will merge within

different use cases I think crypto is a funny space where people want to make a quick buck and there's like plenty of narrative is uh I think couple of months ago everyone was throwing money on these AI agent launched coins um but we are trying at near to focus on on on on building infrastructure that would enable this uh agent-driven economy right so both Google and and uh Shai um is working towards achieving the same goal right building the best tech for humanity and blockchain and AI is tech components that we want to offer to the world I would say yes of course that you should do it and I think uh rather the the the the massive upside is not

necessarily in the whole world moving to crypto and AI living on crypto it's in that 2.5 intersection where uh as I said AI because of the very nature uh that it is non-deterministic it will need a truth stack I mean Google already have their own internal ledger so that is the intersection to focus on and and the and the the reward there is that's the entire world using it, right? It's not the crypto uh bubble. It's like the whole world is looking for truth technology and that's where you should invest. So, it's the AI crypto mullet.

Uh LLM in the front and crypto in the back. That's it. That's what stable coins are, right? It's stable coins underneath, but everybody's uh using their traditional apps and I think uh stable coins in USDC is like the best example, right? Uh it's onchain but it's actually dollars underneath. Yeah. So, makes sense. All right. Well, thank you very much, guys. Uh, it was very interesting and uh, still I'm still not not completely convinced I should make an investment right away, but uh, it's still something that I look very closely. So, thank you very much. Thank you.