چکیده مقاله
This systematic review assesses the degree to which large language models LLMs mitigate confirmation bias in strategic decision making by senior executives in technology and finance industries we identified o top relevant studies and only studies qualified, predominantly in finance five , with others in business two , operations management, security, and high stakes decisions; none directly involved real senior executives or technology applications, relying on simulations, analysts, or vignettes Results show LLMs e g , GPT 5,0, GPT 2, ChatGPT yield mixed bias reduction In structured, data rich tasks, they produce conservative forecasts with smaller errors Li et al , YYY , higher earnings prediction accuracy 7 % vs 0% human; Kim et al , Y ε , and alignment with professional benchmarks via hierarchical agents Chen et al , Y TE Debiasing via retrieval augmented generation or self help prompting minimizes hallucinations and enhances consistency Esposito et al , YYE; Echterhoff et al , Y Yε Conversely, in ambiguous or unstructured scenarios, LLMs mirror human biases e g , disposition effects; Tan & Li, Y ε , exhibit persistent confirmation bias resisting counter evidence Lee et al , Y Yo , or display overconfidence Chen et al , Y YT Moderators include task structure, model capacity, prompt design, and human oversight, positioning LLMs as complementary tools Limitations encompass LLM inherent biases, technical inconsistencies, and risks of introducing new errors in complex contexts The absence of technology specific or executive focused evidence underscores gaps, urging targeted empirical studies for ethical, high stakes deployment
کلیدواژهها
نویسندگان
Bahram Rezvani
Msc student of Information System Management Advanced Information Systems, College of Management, University of Tehran, Tehran, Iran
شیوه ارجاع
Rezvani, Bahram,1404,The Role of Large Language Models in Countering Confirmation Bias among Senior Executives,The 13th International Conference on Management, Accounting, Banking and Economics in Iran,Mashhad
ارائهشده در
مجموعه مقالات سیزدهمین کنفرانس بین المللی مدیریت، حسابداری، بانکداری و اقتصاد ایران21 آبان 1404 · مشهد