DERELAB: Probing Defeasible Reasoning and Confirmation Bias in LLMs with a Generative Benchmark
2026-08-31 • Artificial Intelligence
Artificial Intelligence
AI summaryⓘ
The authors created DeReLab, a system that generates conversations where beliefs change as new evidence appears, allowing careful testing of how language models update their ideas. They used it to study confirmation bias, where models tend to believe information that agrees with their current view but resist changing when new conflicting facts come up. By testing nine language models, the authors found most struggle to properly revise conclusions even when evidence suggests they should. Their work provides a tool for better understanding and evaluating how models handle reasoning that can change with new information.
Defeasible reasoningNon-monotonic reasoningBelief updatingConfirmation biasLarge language modelsDefault reasoningInheritance reasoningGround truthMulti-turn conversationsControlled generation
Authors
Jayanta Sadhu, Sayem Shahad, Kenneth Marino
Abstract
Defeasible reasoning is a type of reasoning where inferences are drawn from plausible current evidence, but can be retracted upon the introduction of newer evidence. Although recent studies have examined language-model behaviors in defeasible reasoning, the datasets have been static and lack wide coverage of non-monotonic reasoning categories. We introduce DeReLab, a generative framework that produces multi-turn belief-updating conversations from parameterized graph structures across default and inheritance reasoning, with formally verified ground truth at every turn, enabling controlled measurement of how models respond to confirming and disconfirming evidence. This controlled generation process creates a testbed for experimental designs that isolate specific reasoning demands. Applying this capability to the study of confirmation bias, we evaluate nine open and proprietary large language models and find that nearly all exhibit a systematic tendency to accept congruent evidence while resisting incongruent updates, with several models correctly identifying a weakening update yet failing to revise their conclusion. We believe our work and findings will facilitate future research on evaluating language models in defeasible reasoning.