Levels of Learning in Natural and Artificial Agents
Abstract
This project will investigate the hypothesis that there are two distinct levels of learning in natural and artificial agents. We initially define Level 1 (L1) and Level 2 (L2), where L1 are fixed, innate, automatic (architectural) learning mechanisms, and L2 are (knowledge-based) learning strategies that are deliberately controlled by the agent to create experiences such that L1 mechanisms can learn useful knowledge. In humans, L1 include at a minimum forms of perceptual, procedural, reinforcement-based, episodic, and semantic learning. L2 strategies include simple rehearsal and practice, self-explanation, after-action review, and in the limit, scientific research. The purpose of this project is to explore and refine these distinctions across both artificial and natural intelligent autonomous systems. Our research will include literature reviews of relevant research in neuroscience, animal behavior, psychology, and artificial intelligence. We will analyze existing learning methods and strategies described in these fields, including specific AI systems in terms of L1 and L2. We will also attempt to develop a general description of the computational processes that characterize these levels.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- May 30, 2018
- Source ID
- FA95501810180
Entities
People
- John E. Laird
Organizations
- Air Force Office of Scientific Research
- United States Air Force
- University of Michigan