InductionBench: LLMs Fail in the Simplest Complexity Class

2025-02-20

AI Summary

This work addresses the weak inductive reasoning capability of large language models (LLMs). We introduce InductionBench, the first benchmark specifically designed to evaluate inductive reasoning. Methodologically, we systematically incorporate subregular complexity classes from formal language theory—such as strictly local (SL) and strictly piecewise (SP) languages—into LLM evaluation for the first time; tasks are constructed using finite-state automata to ensure syntactic rigor, minimal sample size, and controllable difficulty. Our contributions are threefold: (1) we fill a critical assessment gap for scientific discovery—the ability to infer latent rules from finite observations; (2) we empirically demonstrate that state-of-the-art models (e.g., o1, o3) perform below chance on the most basic subregular tasks, revealing a fundamental deficit in their inductive reasoning; and (3) InductionBench establishes a new paradigm for diagnosing and advancing systematic reasoning in LLMs.

Technology Category

• Natural Language Processing: (Large) Language Models
• Machine Learning: Large Multimodal Models (LMMs)
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning

Application Category

• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Large language models (LLMs) have shown remarkable improvements in reasoning and many existing benchmarks have been addressed by models such as o1 and o3 either fully or partially. However, a majority of these benchmarks emphasize deductive reasoning, including mathematical and coding tasks in which rules such as mathematical axioms or programming syntax are clearly defined, based on which LLMs can plan and apply these rules to arrive at a solution. In contrast, inductive reasoning, where one infers the underlying rules from observed data, remains less explored. Such inductive processes lie at the heart of scientific discovery, as they enable researchers to extract general principles from empirical observations. To assess whether LLMs possess this capacity, we introduce InductionBench, a new benchmark designed to evaluate the inductive reasoning ability of LLMs. Our experimental findings reveal that even the most advanced models available struggle to master the simplest complexity classes within the subregular hierarchy of functions, highlighting a notable deficiency in current LLMs' inductive reasoning capabilities. Coda and data are available https://github.com/Wenyueh/inductive_reasoning_benchmark.

Problem

• Assessing inductive reasoning in LLMs
• Introducing InductionBench for evaluation
• Highlighting deficiencies in LLMs' inductive capabilities

Innovation

• Introduces InductionBench for LLMs
• Evaluates inductive reasoning ability
• Highlights LLMs' deficiencies in induction