The Long Road to the Same Answer: Cognitive Bias Under Escalating Reasoning Budgets in Large Language Models
作者: Obada Kraishan
分类: cs.CL
发布日期: 2026-10-07
备注: Accepted at the 2026 IEEE 8th International Conference on Cognitive Machine Intelligence (IEEE CogMI 2026). 8 pages, 3 figures, 5 tables. Code and data: https://github.com/obadaKraishan/anchored-minds
💡 一句话要点
探讨大型语言模型推理预算对认知偏见的影响
🎯 匹配领域: 支柱九：具身大模型 (Embodied Foundation Models)
关键词: 认知偏见 推理模型 大型语言模型 决策科学 剂量反应研究 模型审计 人类认知
📋 核心要点
- 现有推理模型在推理时的计算预算分配可能未能有效减少认知偏见，尤其是在快速直觉判断中。
- 论文通过剂量反应研究，比较推理模型与非推理模型在不同推理预算下的表现，探讨其对决策偏见的影响。
- 实验结果显示，推理模型并未显著降低偏见，且偏见强度在推理时间增加时并未可靠下降，反而有些偏见向非理性方向发展。
📝 摘要（中文）
本文研究了推理模型在推理时分配额外计算资源的效果，探讨了其对经典决策偏见的影响。通过对30个案例的剂量反应研究，发现推理模型并未显著降低偏见，且随着推理时间的增加，偏见的强度并未可靠下降。结果表明，推理过程并不能保证理性决策，建议对模型进行逐项偏见审计。
🔬 方法详解
问题定义：本文旨在探讨推理模型在推理时的计算预算如何影响认知偏见，现有方法未能有效证明推理时间的增加能够减少决策偏见。
核心思路：通过对比推理模型与其非推理兄弟模型在不同推理预算下的表现，验证推理过程是否能降低经典决策偏见。
技术框架：研究使用30个案例，覆盖六种偏见，进行剂量反应实验，推理预算设置为0、1024、4096和8192个tokens，进行12,350次API调用。
关键创新：研究结果表明，推理模型并未显著减少偏见，且偏见强度在推理时间增加时并未可靠下降，挑战了推理过程作为理性保证的假设。
关键设计：实验中使用的参数包括推理预算的不同设置，采用的统计方法为t检验，分析偏见强度的变化趋势。
🖼️ 关键图片
📊 实验亮点
实验结果显示，推理模型在偏见控制方面的表现不如预期，尤其是在锚定偏见方面，尽管增加推理时间并未显著降低偏见强度，反而在某些情况下偏见向非理性方向发展。这一发现对模型的实际应用提出了新的挑战。
🎯 应用场景
该研究对大型语言模型的应用具有重要意义，尤其是在需要高决策质量的领域，如金融、医疗和法律等。通过理解推理过程中的偏见，开发者可以更好地审计和优化模型，提升其在实际应用中的可靠性和公正性。
📄 摘要（原文）
Reasoning models allocate extra computation at inference time and present their answers as the product of deliberate thought. If this deliberation works the way dual-process accounts of human cognition suggest, longer thinking should weaken the classic decision biases that fast, intuitive judgment produces. Using 30 vignettes covering six biases (anchoring, framing, loss aversion, escalation of commitment, availability, confirmation) from an established benchmark, we run a dose-response study across four model families, pairing each reasoning model with a matched non-reasoning sibling and requesting thinking ceilings of 0, 1,024, 4,096, and 8,192 tokens, for 12,350 API calls. Because a requested ceiling is not the same as realized deliberation, we use the reasoning tokens each call consumed as the dose. First, reasoning models are not less biased than their siblings; the point estimate leans the other way in every family, but the item-level pooled contrast is not reliable (Delta = +0.031, t(29) = 1.45, p = .157). Second, bias magnitude does not reliably fall as realized deliberation grows: no slope is significantly negative, and where anything moves it is the signed score drifting further from the human direction. Third, anchoring is the only bias in the human direction (d = 1.89). Four of the other five lean the opposite way in all seven models; with five items per bias, that reversal is reliable for framing and directional for escalation of commitment, confirmation, and loss aversion, while availability is absent. A one-line instruction to restate the anchor before answering lowered anchoring on all five anchoring items, which no amount of additional thinking did, although the effect does not reach significance (p = .057). The results argue against treating test-time reasoning as a rationality guarantee and for auditing deployed models bias by bias.