Overview
Explore the groundbreaking Cicero AI agent developed by Meta AI, capable of playing the complex game of Diplomacy at a human level. Delve into the integration of language models with planning and reinforcement learning algorithms, enabling Cicero to negotiate, persuade, and cooperate with human players. Examine the agent's architecture, including its controllable dialogue model, strategic planning capabilities, and message filtering system. Analyze Cicero's performance against human players and discuss its implications for AI in cooperative games. Learn about the challenges and innovations in creating an AI that can effectively use natural language in interactive environments.
Syllabus
- Introduction
- AI in cooperation games
- Cicero agent overview
- A controllable dialogue model
- Dialogue-conditional strategic planning
- Message filtering
- Cicero's play against humans
- More examples & discussion
Taught by
Yannic Kilcher