Critical or Compliant? The Double-Edged Sword of Reasoning in Chain-of-Thought Explanations
This addresses the problem of misleading explanations in NLP systems for users, highlighting risks in transparency tools, but it is incremental as it builds on existing CoT research.
The study examined how Chain-of-Thought explanations in multimodal moral scenarios can foster confirmation bias, finding that users often equate trust with outcome agreement and that confident tones suppress error detection, sustaining reliance even when reasoning is flawed.
Explanations are often promoted as tools for transparency, but they can also foster confirmation bias; users may assume reasoning is correct whenever outputs appear acceptable. We study this double-edged role of Chain-of-Thought (CoT) explanations in multimodal moral scenarios by systematically perturbing reasoning chains and manipulating delivery tones. Specifically, we analyze reasoning errors in vision language models (VLMs) and how they impact user trust and the ability to detect errors. Our findings reveal two key effects: (1) users often equate trust with outcome agreement, sustaining reliance even when reasoning is flawed, and (2) the confident tone suppresses error detection while maintaining reliance, showing that delivery styles can override correctness. These results highlight how CoT explanations can simultaneously clarify and mislead, underscoring the need for NLP systems to provide explanations that encourage scrutiny and critical thinking rather than blind trust. All code will be released publicly.