Researchers at MIT and the University of Washington have developed a model that accounts for the sub-optimal decision-making processes in humans, potentially improving the way artificial intelligence can predict human behavior.
Named ‘inference budget,’ the model infers an agent’s computational constraints, whether human or machine, after observing a few traces of their past actions. It can then predict their future behavior based on these observations.
It is suggested that this breakthrough could be used in AI systems to understand human behaviors better, thus aiding in the prediction of future actions. Athul Paul Jacob, an electrical engineering and computer science graduate student and lead author of the paper, argues this model could make AI systems more adept at assisting humans, especially if the AI can determine when a human is about to make a mistake, based on their prior actions.
The team used chess games for their inspiration, noting that players’ thinking time varied based on the complexity of the move and that skilled players tended to spend more overall time planning their moves in challenging matches.
The team believes this finding bodes well for planning in a variety of scenarios, beyond chess games, where decision-making is not ideal and said the new model’s performance matched or outperformed a popular alternative across different tasks.
They plan to continue their work in utilizing this model across multiple areas, such as in reinforcement learning, commonly applied in robotics. The ultimate goal of this work is to aid in the development of AI systems that can work more effectively with their human collaborators.