Critical or Compliant? The Double-Edged Sword of Reasoning in Chain-of-Thought Explanations

TL;DR

This paper investigates how Chain-of-Thought explanations in multimodal moral reasoning can both aid understanding and foster overconfidence, revealing that tone and presentation influence user trust and error detection.

Contribution

It systematically examines the double-edged role of CoT explanations in moral reasoning, highlighting how delivery style affects trust and error recognition in vision language models.

Findings

Users often trust outcomes over reasoning correctness.

Confident tones reduce error detection despite flawed reasoning.

Delivery style can override actual reasoning accuracy.

Abstract

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.