6thWave: AI News Hub

A recent paper by researchers from Microsoft Research Asia and the Okinawa Institute of Science and Technology introduces a groundbreaking theoretical framework that integrates habitual and goal-directed behaviors using variational Bayesian methods. Traditionally viewed as separate, habitual behaviors are automatic and fast, while goal-directed behaviors are slow and flexible. However, the new framework proposes that these behaviors share neural pathways and can synergize to enhance decision-making processes in both biological and artificial agents. The core innovation involves the Bayesian intention variable, which bridges habitual and goal-directed actions. Simulation experiments in vision-based sensorimotor tasks, such as a T-maze environment, demonstrated the transition from goal-directed to habitual behavior over repetitive trials, adaptation after reward devaluation, and zero-shot goal-directed planning for new tasks. This research has profound implications for cognitive neuroscience and AI, providing a comprehensive model that balances efficiency and flexibility, and could inform the design of more adaptable autonomous systems.

Source.