Project description
Many games that contain hidden or partially observable information require some element of deception in order to play effectively (Stratego, Poker, StarCraft, etc.). This project will focus on one of several tasks relating to the implementation or detection of deceptive strategies by AI agents. - Game/Scenario Design: Develop simple games & scenarios where deploying deceptive tactics are required to play effectivity. For example, one player may be trying to accomplish certain hidden goals, while the other player is trying to prevent them from accomplishing this without knowing what its objectives explicitly are. Research will involve identifying and categorising different kinds of "deception" that may be encountered. - Predicting Agent Intentions: Look into detection techniques for identifying the intentions and/or objectives of a given AI based on its behaviour. This can be done by observing its prior moves and deriving plausible goals, which can in turn then be used to predict future moves - Hiding Agent Intentions: Look into agent game-playing techniques for concealing your intentions or objectives from an observing player. This could be done by implementing the concept of "bluffing" or making sub-optimal moves in an effort to keep information hidden.
Assumed knowledge
Knowledge of basic AI, Machine Learning and/or Data Science techniques strongly desirable
Note: You need to register interest in projects from different supervisors (not a number of projects with the one supervisor).
You must also contact each supervisor directly to discuss both the project details and your suitability to undertake the project.