Stereotypes in Large Language Models: Demographic Biases in Hiring Decisions

Drushchak, Nazarii

Домівка
→
Students Research & Project Works | Роботи студентів
→
Факультет прикладних наук
→
Освітня програма наук про дані
→
2024
→
Перегляд матеріалів

dc.contributor.author	Drushchak, Nazarii
dc.date.accessioned	2024-08-22T10:12:47Z
dc.date.available	2024-08-22T10:12:47Z
dc.date.issued	2024
dc.identifier.citation	Drushchak Nazarii. Stereotypes in Large Language Models: Demographic Biases in Hiring Decisions. Ukrainian Catholic University, Faculty of Applied Sciences, Department of Computer Sciences. Lviv 2024, 57 p.	uk
dc.identifier.uri	https://er.ucu.edu.ua/handle/1/4665
dc.language.iso	en	uk
dc.subject	Stereotypes	uk
dc.subject	Large Language Models	uk
dc.subject	Demographic Biases	uk
dc.subject	Hiring Decisions	uk
dc.title	Stereotypes in Large Language Models: Demographic Biases in Hiring Decisions	uk
dc.type	Preprint	uk
dc.status	Публікується вперше	uk
dc.description.abstracten	This thesis proposes a methodology for assessing demographic biases in hiring systems powered by artificial intelligence (AI), evaluates existing bias mitigation techniques, and conducts a comparative analysis between English and Ukrainian at all stages. Our study highlights the importance of Responsible AI practices in shaping fair and equitable hiring processes. We initiated this research by creating a dataset of anonymized CVs and job de- scriptions. We then developed a robust framework for benchmarking AI-assisted hiring systems to evaluate potential biases across a range of categories called pro- tected groups. Having detected biases across these groups, we experimented with the known pre- and post-processing mitigation techniques to alleviate the level of bias. Our results show that bias mitigation remains a complex and multifaceted chal- lenge. While certain strategies demonstrated positive results, they haven’t fully fixed the bias problem in AI-assisted hiring. Our work is a foundational step towards fostering fairness and inclusivity within AI-driven recruitment systems. We aim to continue this research, exploring novel approaches to handle bias problems and promote equitable hiring practices.	uk