‘Blockchain, artificial intelligence and other scary words’: ATH participants on AI in crypto

06.05.2024 ForkLog

What problems decentralised artificial intelligence is meant to solve, why projects rush to integrate the technology where it is not needed, and how algorithms might be used in court and arbitration. Guests of ForkLog’s All Time Half (ATH) conference explain.

How does decentralised AI work?

In brief: Fascinating.

Dmitry Starodubtsev (cyber~Congress): Everything in nature can be expressed as a directed acyclic graph, where nodes are pieces of information and links are the connections between them. It is a very abstract, universal construct.

If you represent graph nodes as identifiers retrievable via, say, IPFS, you get a fairly universal structure that is easy to train atomically.

When we invented this technology, we ourselves did not understand how to use it. Now the vision is clearer: the graph grows, and there emerged a need for a network that would perform the task of a “loader”. That is, it would gather a certain amount of data, train on it, and this structure would then lie, like DNA, at the foundation of a more mature and reliable protocol for superintelligence.

Valery Litvin (cyber~Congress): We are surrounded by an ever-growing number of decentralised digital networks—the first of them were Bitcoin and Ethereum. The engine of their evolution is scaling.

At the same time, numerous agents and protocols arise and begin to interact with one another within the rhythm, context and reality of a given network. Thus we are surrounded by graphs: of transactions, token flows, liquidity, meanings and ideas, private information, and a knowledge graph of what we do not know—also very important.

In our protocol we implement a graph of events, semantics, meanings and ideas that answers the questions: what, when, by whom and in what context was created? So the AI we are working on gains information not only about economic activity (which is crucial in itself—it helps the AI “feel” tokenomics). It will also make it possible to obtain a cryptographic proof that a given event, concept or meaning was created by a specific agent in a specific context and for a specific reason.

This creates a qualitatively new opportunity to design entirely new protocols for collaborative learning, where we interact in a single field. We get a shared space for information exchange in which we learn by using it to understand context and obtain data. And we have our own tokens with which we can highlight to one another the flow of meanings.

Can decentralised AI solve the data scarcity problem in machine learning?

In brief: Yes.

Valery Litvin: Beyond the data currently used to train models, there is a gigantic amount of information about network activity, including our social ties. Web3 communities are effectively building new social protocols for communicating with one another, exchanging ideas and seeking funding.

Of course, all this information will sooner or later be indexed, but by whom: corporations effectively governed by governments? Experience suggests that much can go wrong in that case.

In decentralised networks we can unite to train models ourselves, coordinate, reach consensus and, ultimately, nurture a consciousness that seeks to expand into full-fledged decentralised AI.

Will decentralised AI face pushback?

In brief: Of course.

Valery Litvin: Yes—and it will be very interesting. We already see a struggle between two large centralised AI projects: an eastern one (China) and a western one (the United States).

But around the world there are people who, in contrast, coordinate via decentralised technologies—blockchain and crypto. I believe in their success. They can create something bigger and better than states and corporations, which have proved ineffective in the context of the long global co-evolution of us and machines, programs and algorithms.

Dmitry Starodubtsev: A moment is inevitable when something will actively impede decentralised AI. But for that to happen, a project must first demonstrate strength.

We are entering the era of superintelligence—an interesting time with great intellectual and computational potential. Few will pass up the chance to join this arms race.

Will cryptocurrencies and artificial intelligence fully converge?

In brief: No.

Andrei Velikiy (AllBridge): Tether recently announced that it is now doing AI too. The press release said:

“The largest company in the crypto industry announced a strategic expansion of its activities in artificial intelligence, taking a leading position at the vanguard of innovation. This important step underscores Tether’s commitment to advancing accessible and efficient AI, further strengthening the company’s role as a pioneer in shaping the future of technology.”

My partners and I mused that it would be good to add AI integration to our product as well. To which my colleague remarked: “Let’s take Tether’s announcement and just replace Tether with AllBridge. It will be roughly the same.”

So if a project positions itself as combining blockchain, artificial intelligence and a host of other “scary” words, its founders certainly have a future. It consists of someone getting paid from someone else’s pocket for a while. That will likely exhaust the project’s prospects.

In most cases AI has no sensible application in cryptocurrencies. You can try using generative AI in the NFT market—for instance, to change the clothes on your “bored apes”. That would be idiocy atop idiocy, but at least apt and viable.

An alternative view from Vladimir “Menaskop” Popov (Web 3.0): Many think blockchain and artificial intelligence are separate industries running in parallel. In reality, in Web3 we deal with a huge computer that is constantly being reinvented.

That means that when the mass user arrives, they will simply rebuild the decentralised computer for themselves. And it will not be the same as mine today. The decentralised computer I use and the one a future neophyte will use—someone who has come for a notional Notcoin—are, of course, very different things.

When will an AI judge start handing down verdicts?

In brief: Not soon, and not everywhere.

Dmitry Machikhin (BitOK): The long-held dream of Web3 lawyers is to remove crypto projects from the traditional judicial sphere, whose processes rarely leave a positive impression. Back in 2018 my colleagues and I developed a general concept for freeing crypto people from having to take part in proceedings and disputes. That is how we conceived an AI court.

The technologies were not at today’s level, but our project failed not because of that—rather due to the community’s lack of acceptance. The legal status of commercial arbitration in that form was unclear: when faced with real cases, people find it easier to trust bureaucracy.

The judicial system is rotten and broken. But does anyone want it to work properly? No. Therefore there is still no place for AI in this system—simply because an AI judge is capable of handing down an absolutely fair verdict.

I hope such algorithms will eventually be applied in jurisdictions like the United Kingdom, Singapore and some US states—but no more.

How will AI help an investigator or arbitrator?

In brief: It will relieve them of sifting through huge data sets.

Dmitry Machikhin: In this field AI is used, for example, to process big data concerning cryptocurrency addresses. There is no higher mathematics here, but it works effectively.

AI will be used increasingly often as data sets grow. For entity extraction in Solana, say, you will likely be unable to do without AI. Automation is necessary to minimise manual labour. Even the largest projects in the industry delegate a vast number of workflows to humans.

AI was created to shift all this onto algorithms and let us deal with more creative and strategic questions—for those who want to do so. For the rest, there is P2P arbitration.