Reflection in Action: Meta-Reasoning for Goal-Directed Autonomy

In this talk, I will review my laboratory’s work on meta-reasoning for goal-directed autonomy. We are exploring how meta-reasoning in intelligent agents enables self-adaptation and how self-adaptation promotes goal-directed autonomy. In particular, we are investigating how an agent’s model of its own design enables to the agent to adapt its design to recover from failures (failure-driven retrospective adaptation) and to achieve new goals (proactive, goal-directed adaptation). I will illustrate these issues with examples from the interactive turn-based strategy game called Freeciv.