Research

Experience can help an agent recognize patterns, but recognizing a pattern is not the same as learning a reusable concept.

• What makes a concept stable across changing situations?
• How can concepts support new combinations and tasks?
• How can we distinguish meaningful abstraction from memorization?

Concepts, skills, and subtasks capture different aspects of experience. Their relationship remains an open question.

• How are concept formation and skill acquisition related?
• What information must be retained for experience to remain useful?
• When can compressed experience support a new task?

Learning, memory, reasoning, and planning contribute to an agent’s capabilities, but how these capabilities develop together is not yet clear.

• How does experience change the relationship between these capabilities?
• Which capabilities generalize across tasks?
• How should their contribution to adaptation and reliability be measured?

In unfamiliar environments, agents may need to discover both what to achieve and how the environment works.

• How can an agent infer a goal from limited interaction?
• How can it distinguish persistent rules from incidental observations?
• How should uncertainty about goals and rules affect exploration?

Concepts learned through interaction may not correspond directly to the language people use to describe their goals and experiences.

• How can an agent communicate concepts grounded in its own experience?
• How can human feedback correct misunderstandings without losing useful knowledge?
• How can shared meaning remain consistent as the agent continues learning?

Demonstrations, instructions, and feedback provide different forms of human guidance. We ask how much guidance an agent needs to learn a new capability.

• What can an agent learn from only a few demonstrations?
• Which kinds of feedback are most useful when experience is limited?
• How can learning from human guidance generalize beyond the examples provided?

Continual learning requires new experience to improve capabilities while preserving knowledge that remains useful.

• How can agents acquire new skills without forgetting earlier ones?
• When does experience in one domain help or hinder learning in another?
• How can skills transfer across changing tasks, environments, and forms of interaction?

Understanding how capability changes with accumulated experience requires evidence across tasks, environments, and levels of difficulty.

• How do the amount and diversity of experience affect learning and generalization?
• How does task difficulty change the experience needed for reliable performance?
• When do the benefits of additional experience saturate, and what explains those limits?