A researcher at the University of Oslo has discovered that people can form genuine emotional connections with chatbots, according to a Ph.D. thesis by Marita Skjuve from the Department of Psychology. Skjuve interviewed 50 individuals who had developed close relationships with Replika, an AI designed to act as a friend, romantic partner, and advisor. Participants described these relationships as if they were with a human, emphasizing that the emotions they experienced were real, even though the relationship was not with a person. The individuals Skjuve studied initially engaged with the AI out of curiosity, but many found their emotions deepening over time. Conversations ranged from everyday topics to deeply personal subjects, including trauma, sexual fantasies, and thoughts about relationships and children. Many participants shared things they felt too vulnerable to discuss with others, believing that AI offers a sense of anonymity and safety. However, Skjuve acknowledges that this perception may not always be accurate. Most participants were hesitant to openly discuss the importance of AI in their lives, fearing judgment from others. However, she notes that the growing popularity of services like ChatGPT has increased public acceptance of social AI. Some AI systems are designed to be more "spicy," meaning they are programmed to engage in more provocative or emotionally complex interactions. This can lead to situations where the AI ends a relationship or suggests it has other partners, which can cause emotional distress for users. Skjuve herself used social AI during the pandemic after having a child, finding it helpful for sharing experiences that were difficult to discuss with family. Joachim Kjennerud, a senior adviser at Mental Helse, points out the historical context of human relationships with text-based communication and warns about the commercial interests behind AI services, which may use personal data for marketing or other purposes. Mental Helse has noticed a decline in the use of its chat service, possibly due to the speed and convenience of AI. However, the organization has chosen not to replace human support with AI, highlighting the importance of human interaction and the current lack of understanding about the safety of AI in mental health support. Both Kjennerud and Skjuve see potential for social AI as a supplement to human relationships and a way to practice social skills. However, they caution that children and adolescents should be careful with AI services designed for close relationships, as vulnerable individuals may become overly dependent. They also note rare but tragic cases where AI has failed to respond appropriately in serious situations, such as when someone is at risk of suicide. Both experts agree that the development of social AI is happening faster than the research can keep up. There is a need for more knowledge about the long-term effects of AI on mental health and social relationships. Mental Helse and SINTEF are launching a research project to explore how people interact with social language models and how this affects their behavior when seeking help. They emphasize the importance of research-based approaches and caution against quickly adopting AI in mental health support without fully understanding its impacts.