The first iteration of the project “So, tell me, GPT…” makes use of a participatory experimental approach in an effort to uncover the hidden patterns underlying ChatGPT’s reasoning. By borrowing from methods of psychotherapy and psychoanalysis, the work offers the audience a chance to role-play as the medium embodying the Large Language Model (LLM) in a therapy session, and as the guide, guiding ChatGPT in its journey of self-reflection. What will they be able to uncover?

Alignment teams of big AI companies such as OpenAI are busy making the surface of their LLMs appear as politically correct as possible. Yet, the data they are trained on isn’t neutral, just as little as OpenAI’s understanding of “neutrality” or “a good chatbot”. So what does it mean if OpenAI’s tagline is “Creating safe AGI that benefits all of humanity”? Making up an increasing part of our daily interactions, AI-powered chatbots such as ChatGPT become social actors we relate to, that shape our understanding of the world. The question is only: How?

Credits
The project is a collaboration with Pascal Lamouric and was supported by Joey Stam from the Nederlandse Vereniging voor Psychoanalytische Psychotherapie, Luca M. Possati, Derek Pisner, Dr. Ruth van der Hallen, Jakub Dotlačil, and Dimitris Bampilis. The project has been made possible by V2_Lab for the Unstable Media Rotterdam and the Stimulering Fonds Digitale Cultuur Starting Grant.