As a first step, AIs need to understand how they can use smart contracts coordinate with each other (change the incentives of an interaction).

Meanwhile, we have a proliferation of AI applications and agents (There are 600k models on hugging face now.
Those agents will inevitably interact with each other
Traditionally, it has been demonstrated that contracts/commitments significantly increase multi-agent AI systems’ performance (e.g., Amazon robots, Google ad auction auto-bidders) add research paper on multi-agent performance
We should expect the same for future AI systems that are gonna involve even more (and smarter) participants
Now, who makes the best contracts/commitments for those AI systems to use?
Crypto, we are building blockchains for AI agents (whose population will far exceed human users)

We built an environment/benchmark to evaluate and train generative AI agents’ ability in using contracts to coordinate/collude. Specifically, an on-chain game based on a popular manga: Liar Game.