InfoChess Game Framework Introduced for Studying Information Control Through Adversarial Inference
InfoChess introduces a symmetrical adversarial game focused on competitive information gathering, removing conventional piece capture elements that might obscure the significance of information. Players manipulate pieces to enhance visibility across the board, with points awarded based on their probabilistic guesses regarding the opponent's king's position during the game. A tiered system of heuristic agents was created, each with improved capabilities for modeling opponents, to investigate strategic options within this setup. An agent developed using reinforcement learning outperformed these initial strategies. The game’s discrete nature allows for analysis using information-theoretic measures such as belief entropy, oracle cross entropy, and predictive log score, which distinguish epistemic uncertainty from calibration errors in information control contexts. The research was published on arXiv with the identifier 2604.15373v1, contributing to a laboratory environment for measurable information control research.
Key facts
- InfoChess is a symmetric adversarial game focusing on competitive information acquisition
- Traditional piece capture mechanics are removed to eliminate material incentives
- Players score based on probabilistic inference of opponent's king location
- Pieces function primarily to alter visibility across the board
- A hierarchy of heuristic agents with increasing opponent modeling was introduced
- A reinforcement learning agent was trained that outperforms baseline strategies
- Game analysis uses information-theoretic measures including belief entropy and oracle cross entropy
- The research paper was announced on arXiv with identifier 2604.15373v1
Entities
Institutions
- arXiv