The Implications of Large Language Model Integration On Bias in Lending Practices

Dalimar Flores-Torres

By analyzing how popular LLMs such as ChatGPT-3.5, ChatGPT-4.0, and Microsoft Copilot categorize loan applicants into “strong” and “weak” profiles, this study evaluates the potential demographic biases in these models. The results of this study reveal that significant biases are expressed by these models, with strong profiles being overwhelmingly assigned to white, male, U.S. citizens while weak profiles were disproportionately non-white, female, and non-citizen.

Major: 
Computer Science
Exhibition Category: 
Engineering
Exhibition Format: 
Poster Presentation
Campus: 
Harrisburg
Faculty Sponsor: 
Jeremy Blum
Poster Number: 
182