Can We Be Friends with GPT? Exploring the Dynamics of Relationship Development towards a GPT4-based Social Chatbot

Emotional relationships with social chatbots might have become possible due to recent technical breakthroughs. However, the literature shows contradictory findings between group-level laboratory studies and participant-level ecological investigations. We argue that the dynamic and complex nature of friendship is hardly described by classic analyses and might require a complex-systems approach. Therefore, we conducted N = 11 case studies, of which three are presented here. With a self-made GPT-4-Turbo-based social chatbot on WhatsApp, we investigated whether friendship development would take place within a two-month time frame. With our rich quantitative time series data, we determined recurring phases within each participant. These phases are discussed and evaluated using rich qualitative data, taking inspiration from previous work on mixed-methods complexity analyses. In conclusion, our resulting data set shows an enormous variety of experiences with our chatbot and introduces novel analyses to research into human-machine interactions.Emotional relationships with social chatbots might have become possible due to recent technical breakthroughs. However, the literature shows contradictory findings between group-level laboratory studies and participant-level ecological investigations. We argue that the dynamic and complex nature of friendship is hardly described by classic analyses and might require a complex-systems approach. Therefore, we conducted N = 11 case studies, of which three are presented here. With a self-made GPT-4-Turbo-based social chatbot on WhatsApp, we investigated whether friendship development would take place within a two-month time frame. With our rich quantitative time series data, we determined recurring phases within each participant. These phases are discussed and evaluated using rich qualitative data, taking inspiration from previous work on mixed-methods complexity analyses. In conclusion, our resulting data set shows an enormous variety of experiences with our chatbot and introduces novel analyses to research into human-machine interactions.