Abstract

Conceptual abstraction and analogy-making are key abilities underlyinghumans' abilities to learn, reason, and robustly adapt their knowledge to newdomains. Despite of a long history of research on constructing AI systems withthese abilities, no current AI system is anywhere close to a capability offorming humanlike abstractions or analogies. This paper reviews the advantagesand limitations of several approaches toward this goal, including symbolicmethods, deep learning, and probabilistic program induction. The paperconcludes with several proposals for designing challenge tasks and evaluationmeasures in order to make quantifiable and generalizable progress in this area.

 

Quick Read (beta)