Playing Extensive Games with Learning of Opponent's Cognition.

Area of Science:

• Artificial Intelligence
• Game Theory
• Computational Economics

Background:

• Decision-making in intelligent agents is crucial, with cognition influencing outcomes.
• Extensive games model interactive decision-making scenarios.
• Previous models like cognitive perfect equilibrium (CPE) incorporated agent cognition of game structure but not opponent cognition.

Purpose of the Study:

• To propose a novel framework for extensive games that incorporates agents' cognition of their opponents.
• To introduce a new equilibrium concept, adversarial cognition equilibrium (ACE), that accounts for opponent modeling.
• To demonstrate the enhanced realism and strategic implications of ACE compared to CPE.

Main Methods:

• Developed a framework to evaluate opponents' cognition within extensive games.
• Designed and analyzed an algorithm for playing games incorporating opponent cognition.
• Introduced the adversarial cognition equilibrium (ACE) concept.

Main Results:

• ACE was shown to be more realistic than CPE by including learning about opponents' cognition.
• Demonstrated the effectiveness of the ACE framework through a running example.
• Presented findings on the computational complexity, soundness, completeness, and existence of ACE.

Conclusions:

• The proposed framework enhances agent strategic ability by evaluating opponents' cognition.
• ACE represents a more sophisticated and realistic approach to modeling interactive decision-making.
• This work opens new avenues for developing advanced AI agents capable of complex strategic interactions.