Autonomous Discovery in Empirical Domains

Abstract

We present the results of adapting the architecture used in AM (Lenat 1982) to the task of empirical discovery. We believe the architecture is good for autonomous discovery because of its use of justifications to choose the next task to perform. Our preliminary evaluation of the resulting system, HAMB, showed that it was able to make some significant and novel discoveries in the domain of macromolecular crystallization. HAMB proposes its own discovery tasks, selects and performs methods to accomplish them, and then inspects their results both to report interesting discoveries and to propose new discovery tasks. 1 INTRODUCTION We have tested the hypothesis that the architecture used in Doug Lenat's AM (Lenat 1982), which discovered set- and number-theoretic concepts, can be adapted to perform autonomous discovery in empirical domains. A preliminary evaluation of our adaptation, HAMB (Heuristic, Autonomous, Model-Builder), suggests that the architecture is sufficient for autonomous em...