AI x Blockchain: The Next Level

Blockchain and artificial intelligence: the two most era-defining technologies of our time. Each has been a mighty, groundshifting force in its own right, like Godzilla and King Kong in their respective domains. But from time to time, Kong and Godzilla have teamed up when facing a monster that neither could take down on their own. (They’ll do so again next year in Godzilla x Kong: The New Empire.🍿) Now, imagine the potential that’s unleashed when the strengths of two titanic breakthroughs, AI and blockchain, are brought together to address gargantuan problems.

Those are the possibilities I recently had the chance to explore when I moderated a panel at Coinbase’s inaugural “Machine Learning (ML) and Blockchain” summit. The panel, which convened four leaders from academia and industry, unpacked opportunities at the crossroads of these two fast-evolving technologies. Our conversation threaded many themes, from how blockchain can accelerate AI development to the complexities of working with blockchain data and the promise of large language models (LLMs).

One of the great unlocks from AI x blockchain is that when it comes to the problem of fake data and content, which is becoming an exponentially bigger problem as AI proliferates, blockchain can counter misinformation with cryptographic digital signatures and timestamps, making it clear what’s authentic and what’s been manipulated. At the same time, AI can improve the efficiency of blockchain networks, enhance their security, and unlock new features, like allowing protocols to make decisions based on real-time, on-chain data.

Rather than summarize all of the possible synergies, I think it’s best to let my colleagues speak for themselves. Read on for their takes, edited for clarity and length.

Bhaskar Krishnamachari, Professor of Electrical and Computer Engineering and Computer Science, University of Southern California

In my view, there are two main areas where blockchain and AI intersect. The first applies ML models to tackle challenges in blockchain, while the second uses blockchain to address urgent issues in AI.

In the first scenario, ML models can surface complex patterns within blockchain data and help improve the performance of on-chain, decentralized applications. By analyzing transaction data, they can expose potential misconduct, such as wash trading and illegal fund transfers, and detect emerging security threats. In addition to helping to secure blockchain networks, ML models can enhance their performance. For example, they can dynamically adjust transaction fees in response to trading volume and optimize system resources during periods of peak usage.

Less discussed is how blockchain can aid in the development of AI. As the foundation of a borderless, internet-native payment system, blockchains can create financial incentives for people to contribute data and compute resources to train ML models. We have been doing research at USC on decentralized data marketplaces to enable this.

In recent years, we’ve seen a handful of tech companies amass an ever-greater share of the world’s data and AI power. This has raised concerns about privacy, bias, and security: all of which blockchain, as a decentralized, transparent, and openly auditable system, can address. For example, blockchain can trace the provenance of data used to train AI models and cryptographically verify its authenticity. By confirming that these inputs are unaltered and impartial, blockchain can help increase our trust in the recommendations that AI systems provide.

Leo Liang, Head of Data Platform and Services, Coinbase

At Coinbase, most of the challenges that my team faces concern data. Specifically, we need to extract data from the blockchain and convert it into formats that can be used by ML models. I like to think of the blockchain as an onion because of its myriad, intricate layers. Its decentralized nature means that data is distributed across many nodes, each of which independently validates and adds new blocks. When multiple blockchains come into play, the complexity compounds: now you’re dealing with an interconnected network of onions! Synchronizing and ensuring data consistency across this sprawling, diffuse ecosystem is anything but simple.

In addition, blockchains are self-contained systems, unable to access knowledge about the world beyond their boundaries. For ML models to make real-world predictions, we need to join on-chain data (data stored on the blockchain) with off-chain data (data external to the blockchain, such as stock prices, exchange rates, weather patterns, and so on). Think of it like connecting a blockchain to the Internet. It’s a fascinating but formidable engineering puzzle.

Sam Green, Co-Founder and Head of Research at Semiotic Labs

At Semiotic Labs, I lead AI R&D efforts for The Graph, a decentralized protocol for interacting with and making use of blockchain data. Put simply, the Graph reads data from the blockchain, processes it, and creates an index, much like the alphabetized list found at the end of an encyclopedia. This organizational structure simplifies data retrieval from the blockchain. By thus “indexing” blockchain data, The Graph transforms it into a format that’s easy to query, analyze, and apply in downstream applications.

Transactions on The Graph involve two primary participants: a data seller, or indexer, and a data buyer, or consumer. These entities interact through what we call “gateways.” When a consumer sends a query to a gateway, the gateway distributes the query among indexers, taking into consideration factors like the bid price, quality of service, latency, etc. Indexers earn money by serving these queries and delivering blockchain data to consumers. With the help of AI, we’ve built algorithmic pricing agents that help indexers maximize revenue while ensuring that consumers receive reliable, high-quality service.

In many ways, blockchains are an ideal environment for training AI agents. The rules, defined by smart contracts, and the players’ actions, recorded in transactions, are all openly visible on-chain. Because these rules and actions are known, we can create simulations of this blockchain “game” and use them to train AI agents before deploying these agents in live settings. The secret lies in rapid feedback loops: the faster the rate of learning through trial and error, the faster the agents can improve their performance.

Looking ahead, we see immense potential to integrate LLMs into The Graph. Today, users have to query The Graph in a specialized language called GraphQL. By contrast, LLMs allow users to phrase their requests in natural language. By enabling anyone to interact with The Graph in plain English, LLMs can further democratize access to blockchain data.

Paul Bohm, Founder, Teleport

Teleport is developing an open marketplace for ridesharing. Currently, ridesharing is a closed system, which makes it difficult for users to switch between different services. If email were closed like ridesharing, users of Microsoft’s Outlook mail and Apple’s iCloud mail would not be able to email each other. Similarly, if the web were closed, Apple’s Safari browser would not be able to communicate with Microsoft.com.

Opening up ridesharing means returning it to the norms of the internet. In an open system, participants can choose from a variety of apps by many different vendors that communicate with each other. Closed marketplaces often do not allow the market to set a fair price. Instead, they set prices themselves and maximize the value they can extract. Opening up ridesharing and cutting out this middleman means more money goes to drivers, riders pay less per ride, and more money stays in local economies.

To succeed, open marketplaces must be trustworthy. Engineers often focus on aspects of technology first, such as its speed or novel features. But when building for the real world, we have to start with the users’ need for safety, security, and privacy. Only then can we determine the best technology to meet these needs, rather than the other way around.

These are just some of the possibilities, and just the start of the conversation, about what might be unblocked, enhanced, fortified, and taken to new heights when blockchain and machine learning join forces. Digital consensus technologies like blockchain allow the design of systems that are not only fair, trustworthy, and secure, but provably so. As AI threatens to further undermine trust, blockchain bolsters it, providing a robust mechanism to safeguard the integrity of sensitive data. Meanwhile, AI makes it possible to make sense of the fathoms of distributed data that make blockchain too unwieldy and recondite for mass adoption. By deploying artificial intelligence to this inhuman-scale problem, we can bring blockchain to a billion users.

For blockchain or AI entrepreneurs out there, these are the exhilarating prospects to open your mind up to: not just one or the other technology, but both, working in unison and vastly more powerful for it. AI and blockchain. Godzilla and Kong. Atomic fire and gorilla punch. This is how we take it to the next level. Now—go and be legends.