Inspiration
Our project was inspired by the potential of AI agents collaborating toward a mutual goal for the benefit of a larger community. We were intrigued by the idea of leveraging blockchain technology, specifically smart contracts, to facilitate this collaboration. Our work draws inspiration from research on Multi-Objective Contract Augmentation Learning (MOCA) and the Melting Pot 2.0 project, which explores the dynamics of multi-agent systems.
What it does
Our project integrates Solidity smart contracts and AI agents to optimize multi-agent systems within Partially Observable Stochastic Games (POSGs). We employ MOCA to adjust agents' reward functions, encouraging cooperation and mutual goal achievement. This approach ensures transparent, immutable, and decentralized enforcement of game rules and reward distribution, fostering fair and efficient multi-agent interactions.
How we built it
We utilized Solidity for developing the smart contracts and the AutoGPT library for implementing the AI agents. The smart contracts define the game rules and reward distribution, while the AI agents, trained using reinforcement learning, interact with the smart contract to optimize their performance.
Challenges we ran into
Integrating AI agents with blockchain presented unique challenges. One significant hurdle was managing the interaction of agents with the blockchain, considering factors like gas fees and transaction times. Ensuring that the smart contracts were transparent, immutable, and decentralized, while promoting cooperation among agents, was another challenge we faced.
Accomplishments that we're proud of
We hope to lead a new research field under "Agents on-chain collaboration" names DeCoAI.
We take immense pride in developing an innovative approach that synergizes blockchain and AI to optimize multi-agent systems. Our project stands as a testament to the potential of these technologies in enhancing cooperation and efficiency in complex environments. The successful integration of Solidity smart contracts and Auto AI agents, and the resulting improvement in agent collaboration, is an achievement we're particularly proud of.
What we learned
Our journey through this project has been a profound learning experience. We delved into the intricacies of leveraging smart contracts for facilitating communication between AI agents in a decentralized setting. We discovered how these contracts can be used to define rules and distribute rewards, aligning the incentives of the agents towards cooperation. The most significant learning, however, was understanding how AI agents can learn and train together to perform tasks for the common good. This collaborative learning, fostered by the transparent and fair environment created by the smart contracts, has the potential to significantly improve the efficiency and effectiveness of multi-agent systems.
What's next for Blockchain designed for Autonomous agents collaboration
Moving forward, we plan to expand our project by implementing more smart contract examples and introducing more complex environments for the agents to interact in. This will allow us to test and refine our approach under a wider range of conditions.
We also aim to improve the methods we use to rank and evaluate the performance of the agents. By developing more sophisticated ranking methods, we can gain a better understanding of how the agents are improving over time and how different factors influence their performance.
In terms of applications, we're excited to explore how the trained agents can be used in various domains. Potential applications include autonomous cars, where the agents could be used to manage interactions between different vehicles; gaming, where the agents could be used to create more intelligent and adaptive opponents; and community management, where the agents could be used to manage resources and coordinate activities for the benefit of the community.
In essence, the next phase of our project involves expanding, refining, and applying our approach to a wider range of scenarios, with the goal of demonstrating the full potential of integrating blockchain and AI in multi-agent systems.
Log in or sign up for Devpost to join the conversation.