Primate-Inspired Specialized Learning in an Agent Architecture: Safe and Robust Adaptive Action Selection

Harvard University, Cambridge MA

Investigators

Abstract

EIA-0132707 -Harvard University-Marc D. Hauser-Primate-Inspired Specialized Learning in an Agent Architecture: Safe and Robust Adaptive Action Selection The intended multi-year research project would build a library of computational modules or idioms for representing learning and action-selection in non-human primates. We will particularly focus on modeling why well-learned new behavior does not always over-ride more established behavior patterns, and under what circumstances it can come to be exploited. This modeling could be key to designing fail-safe mechanisms in highly adaptive intelligent systems. In this exploratory project, we would adapt an existing agent architecture to model the specialized learning in Cotton-Top Tamarins, focussing on several completed experiments already documented in our laboratory. Conducting this research under an SGER will give us an estimate of the scale of the software project, the level of interaction we can expect between the modeling and our ongoing research, and the educational potential for this system. This will help us determine the level of staffing and funding requested in the full proposal.