Human bias in AI models? Anchoring effects and mitigation strategies in large language models

IF 4.3 2区 经济学 Q1 BUSINESS, FINANCE
Jeremy K. Nguyen
{"title":"Human bias in AI models? Anchoring effects and mitigation strategies in large language models","authors":"Jeremy K. Nguyen","doi":"10.1016/j.jbef.2024.100971","DOIUrl":null,"url":null,"abstract":"<div><p>This study builds on the seminal work of Tversky and Kahneman (1974), exploring the presence and extent of anchoring bias in forecasts generated by four Large Language Models (LLMs): GPT-4, Claude 2, Gemini Pro and GPT-3.5. In contrast to recent findings of advanced reasoning capabilities in LLMs, our randomised controlled trials reveal the presence of anchoring bias across all models: forecasts are significantly influenced by prior mention of high or low values. We examine two mitigation prompting strategies, ‘Chain of Thought’ and ‘ignore previous’, finding limited and varying degrees of effectiveness. Our results extend the anchoring bias research in finance beyond human decision-making to encompass LLMs, highlighting the importance of deliberate and informed prompting in AI forecasting in both <em>ad hoc</em> LLM use and in crafting few-shot examples.</p></div>","PeriodicalId":47026,"journal":{"name":"Journal of Behavioral and Experimental Finance","volume":"43 ","pages":"Article 100971"},"PeriodicalIF":4.3000,"publicationDate":"2024-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.sciencedirect.com/science/article/pii/S2214635024000868/pdfft?md5=a59aced4d78dcdb67f3ba973b6b2959e&pid=1-s2.0-S2214635024000868-main.pdf","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of Behavioral and Experimental Finance","FirstCategoryId":"96","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S2214635024000868","RegionNum":2,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"BUSINESS, FINANCE","Score":null,"Total":0}
引用次数: 0

Abstract

This study builds on the seminal work of Tversky and Kahneman (1974), exploring the presence and extent of anchoring bias in forecasts generated by four Large Language Models (LLMs): GPT-4, Claude 2, Gemini Pro and GPT-3.5. In contrast to recent findings of advanced reasoning capabilities in LLMs, our randomised controlled trials reveal the presence of anchoring bias across all models: forecasts are significantly influenced by prior mention of high or low values. We examine two mitigation prompting strategies, ‘Chain of Thought’ and ‘ignore previous’, finding limited and varying degrees of effectiveness. Our results extend the anchoring bias research in finance beyond human decision-making to encompass LLMs, highlighting the importance of deliberate and informed prompting in AI forecasting in both ad hoc LLM use and in crafting few-shot examples.

人工智能模型中的人类偏见?大型语言模型中的锚定效应和缓解策略
本研究以 Tversky 和 Kahneman(1974 年)的开创性工作为基础,探讨了由四种大型语言模型(LLM)生成的预测中是否存在锚定偏差以及锚定偏差的程度:GPT-4、Claude 2、Gemini Pro 和 GPT-3.5。与最近关于 LLM 高级推理能力的研究结果相反,我们的随机对照试验揭示了所有模型都存在锚定偏差:预测会受到之前提到的高值或低值的显著影响。我们对 "思维链 "和 "忽略先前 "这两种缓解提示策略进行了研究,发现这两种策略的效果有限,而且程度不一。我们的研究结果将金融领域的锚定偏差研究从人类决策扩展到了 LLM,强调了在人工智能预测中,无论是在临时使用 LLM 还是在精心制作少数几个实例时,深思熟虑和知情提示的重要性。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
求助全文
约1分钟内获得全文 求助全文
来源期刊
CiteScore
13.20
自引率
6.10%
发文量
75
审稿时长
69 days
期刊介绍: Behavioral and Experimental Finance represent lenses and approaches through which we can view financial decision-making. The aim of the journal is to publish high quality research in all fields of finance, where such research is carried out with a behavioral perspective and / or is carried out via experimental methods. It is open to but not limited to papers which cover investigations of biases, the role of various neurological markers in financial decision making, national and organizational culture as it impacts financial decision making, sentiment and asset pricing, the design and implementation of experiments to investigate financial decision making and trading, methodological experiments, and natural experiments. Journal of Behavioral and Experimental Finance welcomes full-length and short letter papers in the area of behavioral finance and experimental finance. The focus is on rapid dissemination of high-impact research in these areas.
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
copy
已复制链接
快去分享给好友吧!
我知道了
右上角分享
点击右上角分享
0
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术官方微信