The AI Daily Brief: Artificial Intelligence News and Analysis - Can Blockchains Solve Deep Fakes and Other AI Challenges?

Episode Date: May 14, 2023

This is an excerpt from a discussion that originally happened on The Breakdown crypto/macro podcast. It features Chainlink founder Sergey Nazarov.  The AI Breakdown helps you understand the most important news and discussions in AI.  Subscribe to The AI Breakdown newsletter: https://theaibreakdown.beehiiv.com/subscribe Subscribe to The AI Breakdown on YouTube: https://www.youtube.com/@TheAIBreakdown

This AI breakdown bonus episode is an excerpt from an interview I did with Chainlink founder Sergey Nazaroff about a month ago. Now, for those of you who listen to the main breakdown, you'll have already heard this, but for those of you who have just joined the AI breakdown, this is the part of my conversation with Sergey, where we talk about how the rise of adversarial AI potentially changes the need for cryptographic truth systems in the world. It's a really interesting line of thought, and I think that you'll enjoy it. Happy Sunday and see you back here tomorrow for the regular AI breakdown. I think the thing with AI is that it's very exciting and has a lot of great applications, but it's the type of technology that can get away from people without the right controls and without some of the right things in place to manage risk.

And also it's accelerating at a very fast, fast pace. I've had an interest in it for many years, but at the end of the day, the way it relates to our industry is what I call AI resistant. So I think that one of the main questions is what are the countermeasures? to AI misbehavior and adversarial AI thinks, basically. Both the misuse of AI, which is in the example you described, as well as purely adversarial AI, both of these questions are things I don't see really enough work being done in.

And at the end of the day, from the point of view of decentralized infrastructure like blockchains, adversarial AI would just be another adversary, be another type of adversary. So the real question is, would blockchains be the safe haven to create a reliable world in a world that was made less trustworthy and less reliable by the misuse or through adversarial AIs? That's the way that I would formulate the question. And I think the answer to that question is yes, that blockchains would be quite useful. oracles and smart contracts would be quite useful, not only for verifying what's a deep fake and not a deep fake,

but also for maintaining the ability to transact in markets, maintaining a view of reality in relation to asset prices and assets, because one of the simplest things that AIs could do, frankly, is they could go manipulate and crash markets. And then you would be left with the need for a market, but no existence of a market. and then if you had an alternative environment, such as a blockchain environment, like a Dex or derivatives, a DeFi project, or something else like that, and that was resistant to AI manipulation,

AI resistant, then you would migrate there. So for me, AI has a lot of great potential that I'm excited to utilize in risk management, even within Oracle networks and in other ways. But at the same time, it presents unique risks and a unique profile for what an adversary could be and could do and a new type of adversary. And that adversary is one that could make the world very unreliable, unpredictable. And that's once again what I feel blockchains in our industry does, is it creates a predictable world, right? It creates a world where things are deterministic and the outcomes of markets, the outcomes of information is reliable. So the degree to which AIs get misused or there's an adversarial AI that goes on to manipulate

markets or information, I do think that encryption technologies will be quite useful in minimizing that risk. And if that risk grows large enough, and if there's enough manipulation by AIs of markets, that are not AI resistant or of information channels that are not AI resistant, then I think encryption technologies will be able to resolve that crisis of faith even more. So I think it's not really just about blockchains. I think it's about encryption technologies. And blockchains are an encryption technology that will have a very large role to play in a world that would become unreliable due to AI misuse or misbehavior.

And so that's the thing that's interesting to me, in addition to how can AI be used for risk management and to increase the security of Oracle networks. Because really, it's a tool like anything else. You can misuse the tool or you can use the tool properly. And with any new tool, you'll have people that use it properly and people that misuse it. And if our industry is focused on creating a deterministic, reliable, predictable world, then our industry will have to create that against all forms of unpredictability. And if AI becomes a form of unpredictability, then it's possible that our industry becomes the solution to that problem in certain cases. In your deep fake example, I think it's, I wouldn't say it's a relatively simple fix, but I think you probably need a public

infrastructure that can identify people and then can allow them to sign information as accurate or inaccurate. Personally, I really wouldn't be surprised if Twitter became this. So if Twitter, they had the checkmarking everybody, on the one hand, it makes them money. On the other hand, they're getting people to verify themselves into a private key infrastructure where they could sign things as fake or real. And I think blockchains would have a role to play in that. But I think blockchains would actually have an even bigger role to play in various systems where transactions were getting manipulated. So, for example, markets.

And that's that, while deepfakes in this type of stuff is valuable, really markets are very valuable. Markets are what the whole world runs on. And if you get to a world where AIs can manipulate markets in adversarial ways, that's a pretty scary world. And if you had another alternative world where you had other markets that weren't manipulated, like defy markets weren't as easy to manipulate for the reasons that we see. said with encryption technologies being ingrained in there at the very base level, that could indeed happen. But it once again depends on this event, right? It depends on this negative event. The negative event could be global financial collapse irrespective of AI. Could be AI accelerating

global financial crisis. Could be, I don't know, I don't know what it could be, but basically it could be a ton of stuff. And I don't think the world's going to get more reliable. I think it's going to get less reliable. And whether that's AI-driven or global markets, fun time comes to an end-driven, I think our industry will be the answer to both or either of those scenarios. I want to just spend a moment maybe playing out this example, because I think it's something that people are going to start to really try to wrap their heads around. Let's maybe try to hone down in on two things. One is, what might adversarial AI market manipulation look like in a simplest form of that. And then how would a blockchain-based or encryption-based market be less subject to that

adversarial behavior? I'm actually reticent to give anybody too many ideas about this. Yeah, I thought about that actually as I was asking the question. From my point, maybe let's do the second, why don't we do the second, I think that the second part of the question might be able to be answered without the first part of the question. So rather than thinking about what the, what the attack are, what makes blockchain or encryption-based systems more inherently able to resist or AI-resistant, or manipulation-resistant? Sure, makes sense. So blockchain-based systems and smart contracts and define all these things are built from day one with security audits and security and encryption baked in to allow it to be resistant to various anonymous adversaries.

So from the beginning, the whole premise of the system is actually to be resistant to manipulation. So that's one big advantage. This is not how many centralized systems are built. The way that centralized systems are built is if you find one flaw and you get into the system, once you're past the security measures of the system, you can do a lot of crazy stuff, basically. And the amount of internal checks and the amount of internal checks and balances is not as large. So I think the first thing is that DFI protocols have been securing hundreds of billions of dollars in value in an already very adversarial environment that it constantly tries to take and steal from them. The second reason, which I think is actually the more important reason, is once again consensus.

Consensus is what underpins all of these applications, whether it's at an Oracle network basis or a smart contract state or the blockchain trance. transaction itself, all of those are generated through some form of consensus. And the reason that consensus has better security properties is because of having one entity to compromise, you have tens or hundreds or thousands of entities, basically. So your ability to compromise one entity, research that entity's security practices, identify issues and holes in its infrastructure is significantly easier. then it is to do that to over 50% of the Bitcoin network, right? So you basically are left in a place where you have multiple places to compromise.

And then as you have systems that have more and more independent nodes, you essentially arrive at a place where it becomes harder and harder. So it's essentially the centralization property. I think it makes sense. And thank you for taking some time. Like I said, I think this is an area of growing discourse. And I genuinely believe I said this, I've said this a couple times on Twitter, first, that a meaningful number of people who have been in the crypto space for whatever it is the last couple years and maybe came into the door of interesting tokenization projects or came in the door of NFTs are going to find themselves feeling like the big purpose is in fact having this sort of alternative to some of the challenges that sort of new AI world will be get. Super interested to have that conversation.

I guess by way of wrapping up, Sergey, it's been a great conversation. As you look over the next, I don't know, call it six months, 12 months, what are the most important or exciting steps on the pathway to this more deterministic world? I think Defy has a big future and will continue to become more and more widely used. Blockchain gaming, I think as the scalability properties of the core underlying infrastructure improves, will continue to migrate, and I see more and more large real studios looking at making blockchain-based games for some of the reasons that I described. I think those are the two main categories right now, Define Gaming. Insurance is something we're always

working on, and it moves as slowly as an industry, but has a lot of impact. I think what we're going to do in order to facilitate all that is have greater and greater quality data, allow people to do more advanced computations in a consensus-driven way outside of a blockchain. and enable cross-chain communications and value transfer so that you can build applications that are essentially functioning across multiple chains. So those are the three big pillars of what we plan to achieve is more and better quality data, more advanced computation, so you can build more advanced applications still using consensus, but without relying on the computational limitations of chains.

And then also cross-chain communications and value transfer. so that you can build applications across multiple chains and what we're calling cross-chain applications. I think those three infrastructure improvements will in the coming months and years really accelerate what people can do in defy and gaming and insurance. But then there's always other use cases, right? Like it's very possible deep fakes become important. And then there's a lot of value for some version of proof of reserves that proves the origin

of a piece of content, how NFTs came out of left field and very quickly became a social phenomenon. So I think that there's always that kind of unexpected element. And when that comes, I think what we're going to do is we're going to repurpose Oracle networks to address the need for cryptographic guarantees and cryptographic truth in that new unexpected element as well. Do you think maybe that's another great question, does it feel to you like the information, infrastructure is starting to be in place now to deal with unexpected occurrences that happened. You were just mentioning that you could repurpose oracles for this purpose. Do you see that sort of not just in the context of oracles, which obviously you spend your time on,

but other parts of the crypto or blockchain ecosystem as well? Are we prepared? Maybe another way to ask it is, are we prepared for unknown unknowns or are there things that we're lagging on that need to be built? I think AI and quantum computing are things. that are unclear in terms of their capacity to influence our industry. I think both of those have the ability to make what our industry does either more valuable or less valuable, depending on how they progress, basically.

And I think that our industry will need to adapt to the emergence of both of those technologies in the next AI right now, quantum computing, That's not clear to me. But I would say those are the two big unknown unknowns, depending on how far they go and how fast they get there. Other than that, there's the usual question about consumer demand and governments and all this type of stuff. But that's something that our industry has become experienced and more experienced at dealing with. Because as long as the consumer demand is there, people will make applications that service it. I think the unknowns and unknowns are really around.

Yeah, AI and quantum computing. I would say those are the two big questions that I have. Those are big questions, I think, for everyone right now. And always awesome to have you here, Sergei, to dig into this stuff. We're going to need to have these conversations on a more frequent clip just to keep track of everything. But I appreciate you taking the time out of building to come talk with us today. Absolutely, my pleasure.

Thank you for having me.