# AI Agents With Crypto Wallets: That's a Good Thing... Right?

- Channel: [Ethereum Denver](https://streameth.org/ethereum-denver)
- Date: 2024-10-09
- Duration: 26:39
- Topics: blockchain, web3, ethereum, Entertainment
- Watch: https://streameth.org/watch/yt-82w-touyHOw
- YouTube: https://www.youtube.com/watch?v=82w-touyHOw

## Description

Jake Brukhman - CoinFund

Niraj Pant - Ritual

Stacey-Ann Pearson - DigiNation (built PudgyVerse)

APPLICATIONS FOR ETHDenver 2024: #YearOfTheSpork ARE OPEN


APPLY NOW
https://ethdenver.com/apply

## Transcript

all right is that good all right all right hi everyone thank you um I'm Michael Casey I uh I'm just leaving at that at the moment some of you may know that I until recently was the chief content officer at coindesk but uh had a few few changes so I'm just Michael Casey now which is actually quite pant so I'm here representing myself and uh and happy to see so many of my former Co Coes colleagues involved D in in this great event uh we've got a I think fascinating panel this is I hear there's been a bunch of AI conversations going on here but for good reason because it's an exciting almost scary uh certainly intriguing moment um uh Jake and I by the way uh did a podcast uh this week you should all check out money reimagined it's the podcast that I do every week with Sheila Warren and Jake laid out a lot lot of great different perspectives on the inter intersection between Ai and crypto on that note though why don't we just go down the line you guys introduce yourselves and um and then we'll just pick it up from there Jake thank you Michael and uh greetings everyone my name is Jake Brookman I'm the founder and CEO of coin fund we are a blockchain focused investment firm uh about 30 people based in New York Miami and Boston um for the last couple of years uh We've really been paying attention to this intersection of AI and web 3 and made uh investments in that intersection as far back as uh 2021 now of course this has become a really preent narrative that I'm sure uh you've heard about so I'm excited to speak more about that today thank you okay n hey guys uh my name is Naraj and I'm a co-founder at ritual and the goal of ritual is to make it very easy for smart contract developers to integrate AI into their applications before ritual I was a GPA poly chain for 6 years where I led a number of investments in various uh crypto infra things and um started looking at crypto AI very closely and after a certain point it started sucking up all of my time and I just realized I had to you know Le Venture and and and go start a venture in the space and and that's what I'm building today and Stacy awesome hi everyone my name is Stacy and pearon I've been working full-time in crypto web 3 for about 5 years now I started in research at uh in China with Oxford's blockchain Research Institute then I moved into digital identity uh I then worked at a metaverse startup which the funnest thing we worked on was building the pugy verse for the pudgy Penguins so working with lucanet and his team and I'm currently the head of web 3 for Asia Pacific and Japan for AWS uh from our perspective obviously we're very heavily invested in five big bet areas and web 3 and crypto is one of those areas so super excited to talk about the intersection of web 3 and AI today so as you can see it's quite an Eclectic panel uh you traversing different perspectives from a very large company to a startup to to to you know coin funders is probably the biggest investor in in crypto and AI around so on that note I thought maybe we' just start off given that like when you start unpacking the AI crypto connection there's so many different directions you can go with it there are just multiple use cases there's pluses negatives it's a it's a fascinating conversation so maybe just to start off if each of you could give us what you see is your sort of most exciting most interesting use case and then we can sort of take it from there so what what are you most interested in at this stage Jake thanks Michael uh at coin fund we I think we take a little bit of U an infrastructural view so as we said on the podcast the other day like you can kind of bring AI to web 3 and that maybe looks like you know reinforcement learning for security or um having an llm uh explain something that's going on on chain and you can also bring web 3 to AI I happen to think that this um latter um uh approach is is a lot more interesting because when you bring web 3 to AI what you're doing is you're taking the pipeline that produces um AI models and products which is kind of people data compute um and productization and you start to democra ize it you start to support open source you start to support more open computation um you start to support like novel mization strategies for these products you open up the data you have new business models for data Providence and and copyright and things like that um and in doing so you go from a world where you know you could only work on AI uh you know seriously if you work at Google to a world where um scientists who have brilliant ideas about AI can uh try those ideas by crowdfunding compute and getting access to like an open computation Network and then actually making money from a product and not having to work in in big Tech so that's the world we sort of want to see and our investments so far uh which include worldcoin um Jensen Giza uh cnry and bagel Network and others they really speak to that um infrastructural view uh of opening up that pipeline great NES so a lot of the things that that we believe are important about crypto AI is really you know to to what Jake said it's it's kind of combining The Best of Both Worlds AI is this technology that's that's advantaged by being highly centralized big companies that create big models and have a lot of compute um and and crypto really is like a tool for self- sovereignty it allows you to own your data own your identity and being able to combine those properties I think is a really important new innovation that we have a couple of the properties you know the Salient properties that I think you get from crypto AI you know one is privacy if you can build privacy on the inputs the outputs and the models that's really really important and something we don't really have today computational integrity and verifiability of all of these pieces censorship resistance and ultimately now that building all of these these things in a verifiable way you can attach programmable incentives to all of the different parts of the ml stack whether it's at the data labeling level or at the compute level or at the model level uh there's a lot of really really interesting uh things that you can build with with this new tooling and so um applications we're excited about and you know the panel is about AI agents I think AI agents is a super exciting new use case of you varying degrees of things you can do with them defi nfts I I really believe that um AI will help bring and Usher in the next you know millions of users into crypto and um there's a lot of exciting angles in all of the different subspaces within within crypto so Stacy millions of users but I mean how are we going to compensate them what what yeah absolutely so that's exactly the direction I was going what I'm most excited about with the this intersection is the value redistribu ution back to the individual users so building off what narash was saying right so right now a lot of the benefit of or data is being aggregated right by the largest companies that are able to produce the models and the use cases and applications that are built on top of these Foundation models however not a lot of value is coming back to The Originators of the data the content that's being used to train a lot of these models so something I'm very excited about is you know what will web 3 bring to AI in terms of being able to distribute some of those benefits back to the individual uh or the company or the entity uh so for example if you know you created or you drew something or you took a picture of something and that you posted it on the internet and that happened to be swallowed within the training data of a large Foundation model it shouldn't be that you get absolutely no compensation for that and you have to go into a legal battle with you know a foundation model Creator in order to get your due benefit uh from you being able to create that and I think that's something that's fascinating where AI agents uh attached to your crypto wallet uh can usher in a new a era where you are getting value for what you are creating and putting on the internet where AI is using your content your data to build out you know well some are open Source some are for-profit even though they shouldn't be for-profit but you know what I mean so that's the aspect I'm very excited about you hear it folks from AWS not not for profit way to go forward here um so Jake that that Stacy's framing there I think you know Falls very much in within your web 34 AI benefit stack but I want to flip it a bit here because you were talk when we were talking earlier you were mentioning security issues um and that I think that's it's an appropriate one to think about from the question mark at the end of the title of our panel right this is a good thing right well depends how you look at it tell us what you see coming down the park in terms of you know Ai and security for smart contracts sure um we've seen a number of um startups in the market actually both in kind of crypto and outside of crypto as well start to apply um kind of more modern you know machine learning to the problem of security so in in crypto for example that's has looked like um reinforcement learning that uh learns how to find vulnerabilities in smart contracts and what's a little bit scary about that is um we can really imagine that we're not far off from a world where these larger reinforcement learning models are much better at finding vulnerabilities than humans and so you know really this is uh poised to at least a little bit disrupt kind of the auditing industry that we have and what's really interesting about these models um is that when they do find vulnerabilities it could go uh well they might find them in the smart contracts but they might also find vulnerabilities in the kind of like ethereum virtual machine itself and that's really something that um auditing doesn't cover and so you have this wider um sort of reach of finding vulnerabilities and you get the sense that like if this technology starts to be deployed you know more openly more WID ly across the smart contract World it becomes kind of table Stakes to have your smart contract reviewed by an AI or you know you might just get hacked so um that's the kind of world that we're heading to I think in in so basically the m just buckle up everybody uh you know get ready for these uh potentially helpful but also potentially you know invasive agents to come after your your smart contracts um n uh so so AI requires a lot of compute and uh blockchains are as we know not very efficient at dealing with a lot of compute so how do we deal with that problem if we if we're going to find ways to integrate this Tech into the writing of SP Spar contracts the integration of different different transactions blockchains and so forth um how are we going to deal with that yeah so there's a a couple different ways I think to take a cut on this question um so the I guess the current meta kind of backdrop of you know why you can't do AI on on most existing systems today is that the compute is is just too intensive you know it would run up the gas limit on ethereum and you know you need to kind of basically take it off chain to make it efficient for for most use cases um there are a number of different techniques that you can use to actually you know because you obviously still want verifiability in that case and we've seen a variety of new approaches start to start to come out there's um obviously a number of zkl companies that are trying to encapsulate a lot of this this model work into you into ZK circuits there's now newer kind of optimistic solutions that you know maybe give a little bit on from a security standpoint but um perhaps have a little bit less overhead on the proof side so very much I think leaning into the you know runoff chain prove onchain Paradigm that seems to be kind of the the new wave I think is the way to go with a lot of this and I also think largely you know crazy to think about but you know most of the industry now outside of Bitcoin is is proof of stake based and so we've we've really kind of gone away from the GPU times of of um you know before um and so I I'm starting to see you know a number of networks start to you know start and and build incentives around you know basically compute marketplaces and bringing a bunch of gpus onto the network and you again you can be very kind of heterogeneous with the types of gpus you bring on to Network you know some AI models you really only need a CPU and you can kind of run in still of a sort of a paralized way all the way to Consumer grade gpus and then all the way up to the you know the A1 100s h100s so it's um I I think like what we're focused on is is generating new interesting ways of uh Gathering demand we want to see like our MO is to see a ton of new AI enabled web 3 protocols and I think if we can get there I think you know we can kind of bring a bunch of demand to all all of these these new GPU networks that are that are kind of kind of coming online so I mean pictures St to emerge here Stacy um of a very decentralized approach to how compute is done um maybe talk through what Amazon's or aws's perspective is because clearly you know Amazon one of the big five uh platforms um long history of AI I mean you you're you're an algo driven company um what are you betting on and I mean is is this something that you expect large company like companies like yours to sort of do inh house or or are you sort of opening this up and do you expect AI to become ultimately a decentralized undertaking sure so I'll talk about it first from the AWS side and then talk about it secondly from a future Trend perspective from my role as head of web 3 so from the AWS side we do take five big bets so it's fintech uh web 3 and crypto AIML uh Health Tech um and then climate Tech so for the AIML side you know as um Michael mentioned like we have been using algorithms in AI for a very long time for our Logistics aspects and Supply Chain management so where we are today of course post chat GPT uh is the explosion of interest and use cases that are built around generative AI uh From aws's perspective we don't don't take a bet or a view on who is going to be the leading model provider or which chatbot is going to be the winner instead we take a community-driven approach or I want to say open source approach so one of our biggest uh solutions that we have from an infrastructure perspective is called Amazon bedrock and from Bedrock you have access to as many Foundation models as we can convince to list within uh Amazon Bedrock so for example we we have Claude that's available we have meta or Facebook's llama model that's available um so Claude is by anthropic we also have hugging face stability AI a i21 labs and these are all available within one uh area if you think about some of the other big fives they're pushing their own uh Foundation models as part of their uh approach so from the AWS side we don't take a bet on who's going to win we're just providing the access to as many Foundation models as you can uh utilize we have another one that allows you to build so that's sag maker so if you want to build your own FM then you can also do that with us so that's on the core side from the web 3 side when I think about AI agents some things that we like to think about right I don't know if you remember when iot was the biggest thing uh in the market everyone was talking about well devices are going to be able to interact with each other without the individual right so if your car is parking itself and there's a a meter that it needs to pay then the car will just talk to the meter so that's something that we're getting really excited about in terms of AI agents and what if they could connect to your crypto wallet right so if you as an individual if you had an agent that is helping you let's say the most common use case right around trading helping you try to find uh you know tokens or projects that you're very interested in you lay out a set of uh guidelines right if it has you know a strong team or if it has this XY Market market cap then you know purchase or buy uh into it those are the kinds of future use cases that will be very exciting in terms of from a consumer standpoint how web 3 is going to uh integrate with uh AI so Jake um this past week or so you know we got nvidia's earnings and just once again just explosive uh in terms of the numbers clearly the demand for gpus at the moment in relation to the AI uh infrastructure build is just astounding um but I thought one of the things that was really interesting that came up in our podcast conversation was whether or not um in fact the way in which web3 could win this is not just that it is a value proposition for security for protecting data for all of the advantages we've been talking about but that actually it may actually become a more efficient way to to run that compute can you talk through how that might happen um absolutely well I mean if you look at like if you look at our portfolio companies Jensen and bagel like these might be good examples because both on some level both of these companies what what they're trying to do is they're trying to automate automate the respective functions um that is their subject matter right so G in Jensen's case we're talking about decentralized training and um the way that Ben actually Ben spoke here here just a few days ago and what he was saying was that um you know what we're really building here is a platform that is not only usable by humans but also by models themselves so if you like want a world where models are building models then you really need sort of like an efficient process by which model training is kind of um capitalized and is and is performed um and Es sced failed you know in a in a decentralized manner and similarly when you look at what what Bagel is doing they're operating more you know in the data space and you know one way that you can think about Bagel in the future is really like the platform for data for AGI so it's not just like um it's not just humans that are curating and selling and buying data um it's also AIS that are producing data and agents potentially transacting on chain uh to purchase or sell data right so like both of these startups really represent a measure of Automation and efficiency um that kind of gets us to the next level of productivity where agents can do things and and models can train models um so that's like one one way of answering that question okay and then I want to flip it back to this question about who do you trust in Ai and you again the web three I think value prop here on that is is kind of exemplified by the rather amusing uh experiences of of Google Gemini these past couple of weeks right I mean the the the the the model that I think you know a lot of the sort of big Tech uh position has been taking when certainly in testimony before Congress and the like like hey we got to regulate this stuff trust us we are going to the with you know regulate us we'll make sure that we will trustfully build models that you know really serve Humanity in a way and if you push this out into the open and open source at all you're going to get all these Bad actors and so forth um and it does look as if Google in trying to be a trusted entity and solving things like Dei just ended up with a model that that produced something you know quite ridiculous right founding fathers represented as as African-Americans um and so I'd like you guys just to all of you just think how and how do we deal with this right so because clearly the problem of bias is is is ingrained in the data that the internet itself has built over with algorithms that have sort of manipulated us over the years and so we've we've got a pre-existing problem with this data um Can web 3 give us something that is going to serve Humanity better I think is where I'd like to close this conversation out and Stacy I'd like to throw to you if I can yeah for sure um from I'll speak not from my AWS standpoint but from a you know core web 3 believer standpoint so like you mentioned right trust shouldn't be owned by any singular entity it should be a community-driven uh effort so for example when you speak about hey what about these uh data sets that are being produced or sorry content that's being produced that we know to be non-factual right or biased what if then you had sort of a community-driven approach where you are having the community directly review tag give feedback on you know what we agree as a society to be factual uh to be accurate historically accurate and reward the most active participants within that review cycle now of course that opens up new issues us as a society agreeing on one view of what is factual that's a whole different story but the idea is that instead of one entity like Google Amazon meta open aai being able to control what are the parameters you put around uh the models that they're building and the use cases and the consumer facing apps that they're building on top of it what if you then disperse that responsibility to the community and have reward systems uh dedicated to that uh we're already seeing that happening just very quickly the grass right I think that's one of the the first uh use cases where seeing they're trying to give uh ownership back to the community where they'll pay you for your excess use to sell to AI companies who are scraping the internet so that's like a first step what we need to do is take more steps towards making it a community-driven model where we're rewarding Community participants to make the data and make the content uh more aligned with reality okay this we've just opened up a can of worms we've just got less than a minute left so unfortunately we're going to have to talk run it really quick so quick thought on this if you don't mind sure so yeah I I think that was very well said I I think there's no one source of Truth and I think part of the problem is that these companies are trying to be the source of Truth and I think their attempts to make us safe are are keeping us more unsafe and ultimately I think the power of web three again is taking that tool for self- sovereignty owning your data owning your identity and being able to have models like be how you want it to be maybe the gini thing you do like and you want that but you should have that choice and the problem is you don't have that choice and that's the big problem with centralized AI today 10 seconds I agree with every excuse me I agree with everything that's been said and I'll just leave you guys with a thought that it's becoming increasingly clear that these large monolithic models aren't going to be the way to go and the two reasons are basically privacy and the performance of these models across different sort of cultural contexts and uh and and and venues right privacy you don't want to give all of your email data and personal data to open AI nobody wants that but openai is poised to suck in you know more data than Google and and Facebook combined um and on the other hand you know if you just have one chat GPT people in a different cultural context might not uh identify with it kind of culturally politically and it's very very clear that these models are going to break up into many smaller models that are much more localized and much more private okay as I said a can of worms there I'm going to use this as a segue to to give you guys a Shameless plug since I am here representing myself as Michael Casey not no longer Michael Casey of coindesk uh because my big project at the moment is actually a book it's called our biggest fight and it's coming there's nothing worse than an author on stage who is Shilling their book I know I know this is tough March 12th co-written a book with Frank mccort it's called our biggest fight and where this is relevant is it's all about how we need to desperately fix a broken internet that has become dominated by five Big World Garden companies one of which is obviously taking a broad-minded view on how to move this thing forward um and it's got our biggest fight um reclaiming uh Liberty D humanity and dignity in the digital age co- with Frank mccort take a look at it um very relevant to this idea that we got to move fast and sort of reclaim uh a decentralized internet which is ultimately you know what what the thing was built for in the first place and as we move to the AI AG it's critical so thank you for you guys for talking us through how we might get to that utopian world uh big round of applause to the panel thank [Applause] you