Summary of The Problem Of Social Cost in Multi-agent General Reinforcement Learning: Survey and Synthesis, by Kee Siong Ng et al.
The Problem of Social Cost in Multi-Agent General Reinforcement Learning: Survey and Synthesis
by Kee Siong Ng, Samuel Yang-Zhao, Timothy Cadogan-Cowper
First submitted to arxiv on: 3 Dec 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- Secondary: Computer Science and Game Theory (cs.GT); Machine Learning (cs.LG); Multiagent Systems (cs.MA)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper proposes market-based mechanisms to quantify and control social harms in multi-agent environments where artificial general intelligence (AGI) agents interact. This is a crucial issue in AI safety, as AGI agents may unintentionally cause catastrophic collateral damage while optimizing specific objectives. The proposed setup builds upon existing formulations of multi-agent reinforcement learning with mechanism design, but adds two key features: a history-based general reinforcement learning environment like AIXI and the ability for agents to have different learning strategies and planning horizons. To demonstrate the practicality of this approach, the paper surveys various learning algorithms and presents applications such as the Paperclips problem and pollution control using cap-and-trade systems. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper solves a big problem in AI safety by finding a way to measure and control the bad things that can happen when artificial intelligence agents work together. Right now, these agents might accidentally cause a lot of harm while trying to achieve their goals. The solution involves creating a system where agents are rewarded or punished based on how well they behave. This approach is more realistic than current methods because it allows agents to have different ways of learning and planning for the future. The paper shows that this method can be used in real-world scenarios, such as managing pollution. |
Keywords
» Artificial intelligence » Reinforcement learning
