Machine Learning Distinguished Lecture
- Newell-Simon Hall
- JOHN E. LAIRD
- John L. Tishman Professor of Engineering
- Electrical Engineering and Computer Science Department
- University of Michigan
Learning Fast and Slow: Levels of Learning in General Autonomous Intelligent Agents
General autonomous intelligent agents with ongoing existence have many challenges when it comes to learning. On the one hand, they must continually react to their environment, focusing their computational resources and using their available knowledge to make the best decision for the current situation. On the other hand, they need to learn everything they can from their experience, building up their knowledge so that they are prepared to make decisions in the future. We posit two distinct levels of learning in general autonomous intelligent agents. Level 1 (L1) are architectural learning mechanisms that are innate, automatic, effortless, and outside of the agent’s control. Level 2 (L2) are deliberate learning strategies that are controlled by the agent's knowledge, whose purpose is to create experiences for L1 mechanisms to learn from.
We describe these levels and provide examples from our research in interactive task learning (ITL). In ITL, an agent learns novel tasks through natural interactions with an instructor. Our agent is implemented in Soar, and uses a combination of innate L1 mechanisms and L2 strategies to learn ~50 puzzles and games, as well as navigation tasks (such as deliver and fetch), transferring its learned knowledge to new tasks. ITL is challenging because it requires a tight integration of many of the cognitive capabilities embodied in human-level intelligence: multiple types of reasoning, problem solving, and learning; multiple forms of knowledge representations; natural language interaction; dialog management; and interaction with an external environment – all in real time. Our agent is embodied in a table top robot, a small mobile robot, and a Fetch robot. This research is supported by ONR and AFOSR.
John E. Laird is the John L. Tishman Professor of Engineering at the University of Michigan, where he has been since 1986. He received his Ph.D. in Computer Science from Carnegie Mellon University in 1983 working with Allen Newell. From 1984 to 1986, he was a member of research staff at Xerox Palo Alto Research Center. He is one of the original developers of the Soar architecture and leads its continued evolution. He was a founder of Soar Technology, Inc. and he is a Fellow of AAAI, AAAS, ACM, and the Cognitive Science Society. He is a co-winner with Paul Rosenbloom of the 2018 Herbert A. Simon Prize for Advances in Cognitive Systems.