Values in Interaction between Humans & Large-Language Models

Do Large Language Models have Values? Can they form a coherent value system, similar to the values of individuals? And what values do we learn when interacting with them?

In the Educational Values in Human-LLM Interaction project, funded by Google, we collaborate with Prof. Amir Globerson and Prof. Gal Elidan. We investigate large language models, computerized systems such as Chat GPT, Bard and Gemini. These systems stormed our lives with their advanced capabilities. We ask whether LLM's can recreate a complex value system, that resembles the value system humans hold. We further ask whether LLM's can produce value systems of multiple individuals, resembling a human populations.

As many individuals now interact with LLM's often, we ask whether such interactions can influence peoples' values, leading to value shifts.

The Team