Crypto + AI : use cases beyond the buzz words

Transcript

[Music] One two. All right. Good morning. Thank you for joining. Uh my name is Ken Timit. I'm the head of Kronos Labs uh which is the ecosystem entity and startup accelerator uh behind the Kronos blockchain. Uh the Kronos blockchain uh is a top 10 chain uh when measured by total value locked.

Um it's been around for two years now and it's a an open-source layer one which was developed in strong partnership with Crypto.com. Uh so Kronos is a blockchain. It's also a startup accelerator. Uh so we run our own internal accelerator program uh which is currently finishing its second cohort and has uh you know pushed a lot this year to try and um promote boost AI based innovation um on the chain in part thanks to uh partnerships that we've closed with AWS uh Google cloud and

protocol labs who are providing expertise and resources for the start for the startups to explore AI use cases. So today I'm not here to talk about Kronos uh so much. uh I'm here more to share learnings from the last 6 months where we've received uh close to to 400 startup applications as part of the accelerator and through those applications and through the pitch uh presentations from the startups we've sort of tried to understand what's the state of uh innovation at the intersection of blockchain and AI what are the the relevant use cases, what is

ready, not ready. Uh, and so I'm going to share a perspective with you. Um, and I'm also keen to get your your feedback if you have other thoughts. Uh, so first of all, when we think about uh what AI is bringing to blockchain or vice versa, it's important to realize that uh this year we're seeing a revolution in AI. Of course that revolution doesn't impact all areas of AI in the same way. Uh it's really a revolution that comes from the uh invention of the transformers architecture which has enabled the creation of large language models uh and text to image or texttovideo models. By contrast, you

know, AI has been in our lives for a while when it comes to spam detection, recommendation algorithms, uh trading bots. Uh and so um the um uh change that we're seeing this year um is really specific to the introduction of large language models as we see it. Of course, AI can be used for algorithmic trading, but no breakthrough has happened this year in that particular area. And then there are also lots of problems that AI hasn't solved yet and is still working on like how robots interact with the real world, how you put GPUs in secure enclaves. And those are clearly problems that are not solved at the moment.

So when it comes to the uh impact that this AI revolution is having on blockchain and vice versa um the way that we've uh sort of um clustered the ideas from startups is around three buckets of innovation. The first one is uh so we call it 10 times better productivity and user experience and uh our thesis right now uh and I'm going to elaborate on that is that this first bucket is really where most of the value lies right now if you're a blockchain builder. So if you're working on a startup or thinking about a startup using AI to improve the productivity and user experience even though it sounds a bit generic and um uh practical that's

where we see most of the value. Uh second bucket of innovation is everything around identity provenence uh proof of humanity. Uh we're going to discuss it. It's not yet clear. I mean there's clearly uh a huge area of innovation for AI around identity and agents. It's not clear that blockchain really has any contribution to bring to that bucket of innovation. And number three is using blockchain to power the AI economy. There are some very interesting uh projects in that area. uh some of them are likely to materialize uh in the coming months, but most of them are more like five years uh you know um on a fivey year horizon. So

we'll talk about that a little bit and talk about where you can build if you're interested in that in that innovation bucket. So first of all uh let's talk about uh productivity and user experience. So the promise of blockchain from the start was we're going to reinvent the financial system. We're going to streamline it without intermediaries and we're going to make it much more productive therefore passing the cost savings to end users. The reality is that uh if you measure the productivity of crypto and defy and you measure it by looking at millions of dollars managed per employee in those startups. Uh it's a mixed bag.

Uh so on this page what you have are millions of dollars managed per employee in various startups and companies. Uh so you can see that overall DeFi and crypto startups are more productive that you know Bank of America or JP Morgan. So Bank of America and JP Morgan that are on the right hand side are between 20 and 40 according to that metric and then crypto and D5 protocols are somewhere between let's say you know 50 and 400. Uh however uh you can also see that first of all uh if you look at Binance Binance is at the far right. So once you're uh you know even if you're in crypto once you once you start getting hit by regulation and you have to adapt

to all kinds of local markets then your productivity drops quite a bit. Uh and then on the other hand when you look at Black Rockck an asset manager you can see that they are way more productive than any crypto startup even though uh they are clearly trfy and that's the same for hedge funds. Uh so um crypto aims to be more productive than trafi. Uh so far the outcomes are pretty good but they could be improved and in particular they absolutely need to be improved if you take into account the fact that defy protocols are going to need to localize to various languages are going to be hit by regulation etc etc. So there is a strategic imperative

if you're in DeFi or crypto to improve your productivity and AI can help you with that. So at a very basic level I think you know if you're an a blockchain startup right now and you're not using uh AI for marketing or AI for code development and code security you're probably uh you know building on the wrong foundations. Uh and um the the correlary of that is also if you're thinking about building tools for uh blockchain developers, you have to incorporate um uh products that make it easy for your users. So for those blockchain developers to take advantage of AI features. So in marketing, everyone knows uh the the you

know uh the tools that you can use that are coming from uh you know AI um AI products. I think it's interesting to see how Binance used their project called Bicaso which is you know just using enabling their users to um to create generative art uh or generative AI not art for free so PFP is for free it's pretty dumb right it's uh it's just a use of an existing technology but clearly it struck the im imagination of users and as a marketing tool was something that was very powerful uh and then of course in terms of code development and testing open zeppelin is doing uh some interesting work around uh

code security using AI tools and then there's this project called chat GPT which aims to enable developers to generate solid solidity code from from text there are other tools as well I think that's pretty interesting of course human oversight is required uh but you should definitely look at that or build on upon those tools And then the other area where um you can use AI to improve your productivity and the user experience is of course everything around the interface with users. Uh so Crypto.com launched Amy which is their AI enable enabled bot. Uh so primarily right now it's a bot that's uh used to answer general knowledge questions about crypto uh and to help

with product navigation. Uh um there's a a project called Omni by Margin FI which uh is a text box uh so it replaces the usual extremely dense interface of D5 protocols and uh in that interface you can ask for the bot to find yield farming opportunities for you and it will u find it for you. Uh it's you know again just a more um pleasant uh and simple interface for you to deal with the D5 protocol. What I think is interesting as well uh with those chatbots is that they provide kind of social you know human comfort. They can they provide social proof uh which we know is extremely important uh in the

minds of crypto users. You know if you look at how you know the extent to which they're always uh listening to influencers or what's happening on cryp crypto Twitter to help with their investment decisions. Uh so in our accelerator there are also some interesting experiments happening. So there's this project called Solas. It's an account obstruction a wallet for account abstraction. They're using uh AI to generate human readable explanations of the transactions and human readable notifications. Uh there's a project called earn in our accelerator. uh they are using AI to create risk risk scoring using uh onchain data in order to uh attribute a

risk score to borrowers on a on a lending and borrowing protocol. Uh and then there's another project called DI which is a social platform using uh Kronos as well uh and blockchain and they're using AI to um uh enable autocomplete for users. So it's just a way to make the user experience more fluid. Again, I think if you look at those use cases, it's really using AI technology that's really already around and incorporated into crypto. So it can sound not breakthrough. Uh but uh from what we've seen, that's really where the a lot of the value is. And also like I said before, there's a strategic imperative to do it otherwise you won't deliver DeFi, which is more productive

than Trafi. Second bucket of innovation. Uh so we call it provenence and digital identity. And uh I think this bucket usually starts with the realization that the notions of truth and the notions of identity are changing. Uh on one hand when you look at the blockchain space uh so you have account abstraction which is changing the notion of wallet. uh you have uh lots of incentives for inauthentic account to create activity on chain. Uh the a speaker earlier this morning talked about how most of the

activity on new layer 2 is driven by airdrops and most of those wallets taking advantage of airdrops are bots. Uh so of course so our space right now is uh swamped is sorry is flooded by uh robot robot wallets. Uh and at the same time there's also a very interesting uh trend which I think is that you're seeing lots of platforms being developed that help understand who the wallets are create CRM tools etc. So the notion of identity we we we choose to be completely like anonymous human users is changing in blockchain and at the same time of course AI is changing the notion of identity as well. Uh first of all because it's much easier to create

humanl looking or you know uh uh robot agents that um pretend to be humans but also of course AI is used a lot uh for any kind of KYC or identity verification. um you know look at onfido for example which has been been using those technologies for a while to help uh a lot of um uh igos or ICO projects uh u verify the identity of investors. So those trends are happening and are uh changing the uh notion of truth and identity. And the undercurrent of that is also of course that people know that uh LLMs tend to hallucinate. And so

there's this dream that uh if you can use blockchain to track the provenence of what LLM is telling you uh you can maybe prevent or combat hallucinations or make sure that if you're you you know what the training data is when uh you're looking at at the output of a generated um uh image or video. for example. However, I think the when so I think the problems are the trends and problems are well identified. Uh the question is more like is there a solution to those problems and can blockchain really help and here uh our view so far based on startup applications is that it's a mixed bag. It's not clear that

blockchain has uh is going to contribute a lot. Uh so there's a bullish case which is to say uh AI is going to increase the demand for agent and agent identification uh as well as proof of providence. This is going to generate huge demand for decentralized identity and verified credentials and finally all those identity and credentials protocol uh which are decentralized and were until now kind of searching for a for a use case you know will have found product market fit and are going to explode. Uh so that's the bullish case. The bearish case is that you know every time that uh the blockchain community has tried to work on supply chain provenence, proof of authenticity, it's been a huge

disappointment because no one who generates credentials has economic interest in putting those credentials on chain. Uh and so in the end it's always centralized entities who control um and distribute those credentials which is why we still have social login on Twitter and Facebook uh as a means of mainstream adoption. Uh so no clear uh no uh no strong conviction uh in that uh area. Uh we're of course really eager to see how more R&D can generate valuable solutions for the ecosystem.

Let me just check the time. We're good. Okay. And finally, we have uh a third bucket of innovation. How does blockchain enable the AI economy? And so here when you're looking at uh the press, often times the thesis is that there's a shortage of GPUs. uh there's a shortage of um high-end um uh computers and server farms that can run AI models. Therefore, blockchain is the solution. It can um provide developers of AI models with a wide installed base of decentralized computers, you know, from gamers and other people around the world who have

unused compute capacity and can put that capacity at work uh to train AI models. So that's the thesis that you read. The reality uh that we need to be aware of is that so far the AI economy is not doing too badly without blockchain. And so when we're I think when we think about the problem those problems, we really need to put ourselves in the shoes of developers of AI models and truly understand what their unmet needs are. So far when it comes to creating you know uh AI models or fine-tuning them in a decentralized way hugging face uh which is a very web two platform has performed spectacularly and provides by

far the best experience and the most powerful contribution to the community in terms of freeing us up from uh the influence of you know Facebook and Twitter. And of course also when when it comes to renting GPUs on demand to run your models, you know, there's collab for small developers. There's run pod and vast.ai. Uh all kinds of tools that are already uh providing access to those distributed resources without necessarily being decentralized um uh and they work quite well. Uh so therefore we have to ask ourselves really what does blockchain bring to the picture and how it can make a difference.

The first very important uh way that I think blockchain is going to make a difference is uh through um by uh combating censorship. So by censorship resistance uh what we're already seeing today is that there's a phenomenon that we call regulatory capture where the top companies in AI are going around the world meeting with heads of state. So it's Biden on the left hand side um and telling them what AI is way too dangerous. You have to leave it to us to create regulations uh and make sure that those models stay in the right hands. We can have debates as to whether uh the risk is real or not. uh but regardless I think the implication of that is that

when you look at the European Parliament uh French authorities American authorities you know they are meeting with Elon Musk who's telling them it's too dangerous you can't leave it run free uh and uh you have to let us do the work uh so this is going to lead to what we call regulatory capture whereby you know those big tech companies are going to lobby governments to say if you want to create and commercialize an AI You have to submit 500 pages of documentation, register, uh, you know, be approved by some kind of agency, which is going to take two years. And of course this is going to restrict the ability of uh random developers or 18-year-old m machine learning

scientists to contribute to the innovation and also what is contributing to regulatory capture is that there is a huge uh uncertainty around the legal implications of AI uh particularly around IP management intellectual property management and so you know increasingly when you launch a product you run the risk of having a hund00 million Um, so of course you blockchain is well suited for that. I mean, I'm not saying that blockchain is well suited to do uh to to do unlawful things at all. Uh, but I think blockchain is well suited to encourage and enable innovations in situations where there is a lack of regulatory clarity or the regulations or

laws are not adapted uh for the fruitful development of open source innovation. And so I think it is in that context uh in particular that blockchain can bring some contribution. Uh also uh you know one thing that blockchain does well or crypto does well is uh orchestrating commerce auto automating commerce. And so when it comes to monetizing all kinds of models and um services uh in a global way across you know for machine learning machine learning scientists who are living in Indonesia to sell their stuff to people in Mexico I think you know um uh it's just more convenient to use crypto initially for payments once you you know until you are big enough to

start using credit cards and other means of payment. So overall what we're seeing uh is a very interesting field of innovation which also comes with some strong technical challenges right. So right now when people say that they can uh create u decentralized uh GPU networks the reality is that no one really knows how to run uh decentralized GPU networks efficiently uh to train a model. Um and uh also no one knows how to uh put a GPU in a secure enclave in in such a way that you can distribute your model uh without you know while at the same time making sure that no one has access to it

and can steal your um uh secrets or industrial secrets or competitive advantage. Uh so it's it's still very uh there's still a lot of R&D around how you build uh decentralized network um to power the EI economy. But when you look at the right hand side of this slide, you know, where we're we're less talking about uh heavyduty server farms, but we're more talking about commercialization, distribution and monetization. I think that there are some very interesting examples. So, Fetch AI which is working on which is using crypto to um enable AI uh agents to pay each other for services. Uh

there's this project which we have in our accelerator Corgi AI which aims to create a marketplace for prompt engineers. So you know people around the world who are able to create prompts or who are able to fine-tune models and the idea here is simply to use crypto to commercialize and distribute those services around the world in a very easy uh and light way. Uh and then there are also some um uh projects which are um uh at the you know still on the right hand side but at the bottom who are simply trying to incorporate uh the ability uh for smart contracts to run very simple AI models. Uh so that's what that's what people call ZKVM sorry ZKML sometimes.

Uh so the idea here is just simply to find solutions where the smart contracts can make use of um uh AI models that are sitting offchain in in a you know in a simple and effective way. And then at the left hand side of this slide, you have more like projects that are uh of course um uh very visible in the news that talk much more about the use of decentralized server farms to train models and the ab you know making decentralized compute readily available to machine learning scientists. But you know our view right now is that the technology to support those use cases is not really um ready. Uh however those are very interesting and fruitful areas

for R&D. Oops. Okay. So we're done. In summary, if you're a builder, uh first of all, you should look at how you make use of AI to improve your productivity and the user experience or you should also you can also build tools uh for developers and for ecosystems so that they can make use of AI uh easily and there's a huge amount of value to be created there. Uh secondly, when it comes to identity and provenence, uh huge there's a huge need um to support um emerging trends and problems, but it's not clear how

blockchain can help. And number three, when it comes to powering the AI economy, there is also huge potential uh which is driven either by the risk of regulatory capture or by the need to um uh you know by using AI blockchain to create commercial models for AI models. Thank you. So I think we have time for two questions. Questions as um a startup founder actually looking at the interchange of blockchain and AI.

We want to make it easy to apply to jobs. Um, and we're looking to verify jobs on the chain, but make it basically machine readable to do the matching between company what they need and a professional what they offer. Um, and that falls in the bucket in the middle. Do you see that possible in the next future or shall we just focus on one side and then implement the other at at a later stage? Of course, I'm I'm uh I would it would be irresponsible of me to uh comment on your project, you know, from a 10-second introduction. What you know what what I'm doing here is I'm sharing some experiences from a few hundred applications that we've

received. I think the idea of using uh blockchain to uh prove that CVS are correct or work experiences are correct or referrals are correct, that idea has been around for a while. uh so it's very important just to you know even before AI so it's very important to understand what those other project why those other projects have failed and then come up with ways that you can create product market fit that overcomes those challenges but um yeah the I don't have all the answers another question yeah um instead of what uh blockchain brings to AI u what do we have going on with what AI brings to blockchain in

terms of like agency uh like the A and the DAO being AI. What's the what's the what's the current uh trend in that regard? I don't fully understand the question. Maybe you want to give an example or Yeah, I arbitum might be a good example like where the uh you know governance was completely ignored but if there was a an autonomous agent that enacted the will of the DAO then you might actually have a decentralized autonomous organization.

Um yeah, so I think that again we're at a period where in 2023 what has really changed from a technology standpoint is the introduction of large language models underpinned by the transformers architecture. Uh so uh to the extent that um you're considering a breakthrough that you know wasn't possible before LLMs and is possible now with LLMs uh I think that's hugely valuable and in particular when it comes to governance I think we all know that most people don't read governance proposals most people don't understand governance proposals don't know what they should be be paying attention to and the rate of participation uh on blockchain governance is singledigit

percent. Uh so I think anything that uses LLMs to distill for end users what they should be paying attention to uh and understand the pros and cons of a governance proposal I think that's hugely valuable. Mercy again. Mercy. [Applause] [Music] [Music]