Dabble Lab

This is the second post in a series exploring the intersection of AI and blockchain technologies. Each post builds on the previous one, so if you haven't read the first post, I recommend starting there.

I took this picture in Salt Lake City, UT on July 9, 2023. I remember thinking that the city lights and buildings looked like they were made to compliment the sunset and the mountains in the background.

In my last post, I discussed multi-agent AI systems and why I see them as a key enabler for significantly more capable AI systems and potentially AGI. In this post, I'll discuss how blockchain technologies will likely play a critical role in enabling multi-agent AI systems across organizational and platform boundaries.

But first a disclaimer: I'm speculating here. I'm not placing any concrete bets at this point. Things are moving fast and there are a lot of unknowns about how these technologies will evolve and interact. But I can't help imagining the possibilities. So, that's what this is - imagining possibilities.

Blockchain and Multi-Agent AI: Synergies and Potential Use Cases

There are many possibilities so I'm just touching on a few that seem like the most obvious use cases. They are:

Decentralized Coordination: One of the key advantages of blockchain technology is enabling decentralized coordination and agreement between parties who don't necessarily trust each other. This aligns well with the concept of multi-agent AI systems that need to coordinate and collaborate across organizational boundaries. Smart contracts on a blockchain could provide a secure, transparent, and immutable way for AI agents to negotiate, establish agreements, and execute shared workflows even if they are running on systems controlled by different entities.

Collective Decision-Making: Building on the previous point, DAOs (Decentralized Autonomous Organizations) on a blockchain utilize smart contracts to enable decentralized governance and decision-making. In a multi-agent AI context, smart contracts could similarly be used to facilitate collective decision-making between AI agents without centralized control. The rules and voting mechanisms would be codified in the smart contract. This could allow large numbers of agents to efficiently coordinate and converge on decisions.

Marketplace Infrastructure: Blockchain platforms with smart contract capability essentially provide a decentralized marketplace infrastructure. This could be leveraged to create marketplaces for AI agents and AI services. For example, a smart contract could allow people or organizations to post machine learning tasks, datasets, or problems, along with bounties/payments. AI agents could then bid on and complete the tasks, with the smart contract automatically validating results and disbursing payment. Think Mechanical Turk for AI agents. The blockchain provides the trusted infrastructure for this without a centralized platform.

Federated Learning: Federated learning is an approach to machine learning where the model is trained across decentralized edge devices or servers holding local data samples, without exchanging the data samples. This can help address data privacy and data siloing challenges. Blockchains could provide a shared layer for coordination between federated learning nodes run by different organizations. Smart contracts could be used to orchestrate the federated learning workflow transparently.

Beneficial AI Alignment: AI Alignment efforts are focused on the future ability to steer AI systems toward a person's or group's intended goals, preferences, and ethical principles. To that end, blockchains could be used to create a decentralized, transparent registry of AI systems. Smart contracts could specify and enforce certain constraints or behaviors of AI agents - almost like a constitution. There could be mechanisms for humans to monitor, audit, and vote on changes to AI systems via the blockchain. Maintaining decentralized oversight of powerful AI, rather than letting it be controlled by a single entity will be the goal.

Again, this is all speculative musings on my part. But from a technology standpoint - the potential is there for sure. Still, there are tons of open questions and challenges that need to be addressed. For example: blockchains can be inefficient and hard to scale compared to centralized systems. Many blockchains have limited privacy. Most AI today is centralized in large tech companies and governments which will likely resist decentralization. And there are still numerous unsolved challenges in multi-agent coordination/orchestration. Even still, the potential is exciting.

In part 3, I’ll provide examples of how specific blockchains could potentially support multi-agent AI systems. Specifically, I'll discuss Ethereum, Chainlink, IPFS, Hyperledger Fabric, Ocean Protocol, Polkadot, and Golem. I'll also present a hypothetical but real-world example to illustrate how it might all come together. So stay tuned!