Abstraction and Analogy-Making in Artificial Intelligence

TL;DR

This paper reviews various AI approaches to conceptual abstraction and analogy-making, highlighting their strengths and limitations, and proposes new challenge tasks and evaluation measures to advance progress in this area.

Contribution

It provides a comprehensive review of symbolic, deep learning, and probabilistic methods for abstraction and analogy-making, and suggests future directions for research and evaluation.

Findings

Current AI systems lack human-like abstraction and analogy-making capabilities.

Symbolic, deep learning, and probabilistic approaches each have unique advantages and limitations.

Proposes new challenge tasks and evaluation metrics to measure progress in AI abstraction and analogy-making.

Abstract

Conceptual abstraction and analogy-making are key abilities underlying humans' abilities to learn, reason, and robustly adapt their knowledge to new domains. Despite of a long history of research on constructing AI systems with these abilities, no current AI system is anywhere close to a capability of forming humanlike abstractions or analogies. This paper reviews the advantages and limitations of several approaches toward this goal, including symbolic methods, deep learning, and probabilistic program induction. The paper concludes with several proposals for designing challenge tasks and evaluation measures in order to make quantifiable and generalizable progress in this area.