Can LLMs Have a Personality - And Should They?

Abstract

Personality is what makes people unique. Personality traits describe individuals, explain differences between people and influence behavior and social dynamics. As large language models (LLMs) become increasingly human-like and evolve from helpful assistants to conversational companions, questions arise about whether we can identify and modify LLMs’ personality traits and what the implications are. This talk explores the literature on personality research, computational methods to measure and induce LLM personalities, and their effects on human-AI interaction and collaboration. It examines the potential advantages of LLM personality as well as the risks. Finally, it discusses research gaps and future directions towards personalized and trustworthy human-chatbot interactions.

Bio

Kim is a first-year PhD student at ELLIS Alicante working in the intersection of psychology and computer science. She holds a Bachelor’s degree in Psychology from the University of Münster and a Master’s degree in Quantitative Data Science Methods from the University of Tübingen. Her research focuses on human-centered chatbot interactions and personality in LLMs. To develop more trustworthy human-AI relationships, she is examining how LLMs can be personalized to the user’s psychological profile. Her work includes investigating the motivation, methods, consequences and applications of LLM personality. Kim is supervised by Dr. Nuria Oliver (ELLIS Alicante), Dr. Cristina Cachero (University of Alicante) and Dr. Anna Korhonen (University of Cambridge).