Explore an in-depth paper review and author interview on the role of arbitrary social norms in enhancing learning and enforcement behaviors in artificial agents. Delve into a multi-agent reinforcement learning study that investigates why "silly rules" exist in society and how they may contribute to faster adaptation to genuinely useful norms. Learn about the experimental setup, results, and implications of introducing arbitrary rules in a simulated environment. Gain insights into the research process, methodology, and potential real-world applications of this study. Discover how this research contributes to our understanding of social norms, rule enforcement, and the development of artificial general intelligence.
Spurious Normativity Enhances Learning of Compliance and Enforcement Behavior in Artificial Agents
Overview
Syllabus
- Intro
- Paper Overview
- Why are some social norms arbitrary?
- Reinforcement learning environment setup
- What happens if we introduce a "silly" rule?
- Experimental Results: how silly rules help society
- Isolated probing experiments
- Discussion of the results
- Start of Interview
- Where does the research idea come from?
- What is the purpose behind this research?
- Short recap of the mechanics of the environment
- How much does such a closed system tell us about the real world?
- What do the results tell us about silly rules?
- What are these agents really learning?
- How many silly rules are optimal?
- Why do you have separate weights for each agent?
- What features could be added next?
- How sensitive is the system to hyperparameters?
- How to avoid confirmation bias?
- How does this play into progress towards AGI?
- Can we make real-world recommendations based on this?
- Where do we go from here?
Taught by
Yannic Kilcher
