Patterns Over Principles: The Fragility of Inductive Reasoning in LLMs under Noisy Observations

TL;DR

This paper evaluates the robustness of large language models in inductive reasoning under noisy data, introduces a new task and method to improve reasoning stability, and reveals their susceptibility to pattern overfitting and hypothesis drift.

Contribution

The paper introduces Robust Rule Induction and Sample-steered Rule Refinement, advancing understanding of LLMs' reasoning stability under noise and proposing methods to enhance their generalization.

Findings

SRR outperforms other methods under noise with minimal performance loss

LLMs show instability under noisy observations despite stable accuracy scores

Counterfactual tests reveal reliance on memorized patterns over true abstraction

Abstract

Inductive reasoning, a cornerstone of human cognition, enables generalization from limited data but hasn't yet been fully achieved by large language models (LLMs). While modern LLMs excel at reasoning tasks, their ability to maintain stable and consistent rule abstraction under imperfect observations remains underexplored. To fill this gap, in this work, we introduce Robust Rule Induction, a task that evaluates LLMs' capability in inferring rules from data that are fused with noisy examples. To address this task, we further propose Sample-steered Rule Refinement (SRR), a method enhancing reasoning stability via observation diversification and execution-guided feedback. Experiments across arithmetic, cryptography, and list functions reveal: (1) SRR outperforms other methods with minimal performance degradation under noise; (2) Despite slight accuracy variation, LLMs exhibit instability under noise (e.g., 0% accuracy change with only 70% consistent score); (3) Counterfactual task gaps highlight LLMs' reliance on memorized patterns over genuine abstraction. Our findings challenge LLMs' reasoning robustness, revealing susceptibility to hypothesis drift and pattern overfitting, while providing empirical evidence critical for developing human-like inductive systems. Code and data are available at https://github.com/HKUST-KnowComp/Robust-Rule-Induction.