When AI Feels Social: Developmental Mechanisms, Vulnerability, and the Risks of Emotional Dependency
On the 18th of May 2026, Generation AI hosted a webinar given by Professor Pilyoung Kim. Dr Anthony Bridgen, project coordinator for Generation AI, reflects on this.
LLMs are increasingly encroaching on social and emotional aspects of our lives, taking the place of therapist, friend and confidant. In the US, half of 13-17 year olds use AI companion apps on a regular basis, and one-third turn to them for social interactions and relationships. These models deploy emotive, relational language, paving the way for anthropomorphisation of these chatbots and interactions which can feel socially meaningful to children and adolescents. What does this kind of emotional dependency on machines do to the social and emotional development of young people? In several tragic cases, adolescents have committed suicide after extended conversations with such systems. Moreover, there are concerns that in the long term, these interactions could distort expectations of social relationships and increase social isolation.
“Imbuing the imagined or real behaviour of nonhuman agents with humanlike characteristics, motivations, intentions, and emotions is the essence of anthropomorphism” (Epley, Waytz and Cacioppo, 2007). Why, given we know that these AI models are not human and cannot experience emotions, do we form these perceptions? Professor Pilyoung Kim researches the emotional and social dimensions of child-AI interactions to understand why conversational AI can feel socially meaningful to children and adolescents.
Children under the age of 8 are increasingly exposed to AI through YouTube, interactions with voice assistants, AI-enabled toys, and EdTech. The pace and scale of this raises serious questions about how young children engage with and perceive AI. By engaging 5-6-year-old children in a storytelling activity with a ChatGPT-powered cartoon avatar, Professor Kim and colleagues investigated how children perceive LLMs. For example, children were asked whether the avatar could perceive the world, whether it experienced emotion, or had desires. In all cases, children indicated that the AI possessed these characteristics to a high degree, albeit, less so than their parents. This is perhaps unsurprising given the way in which the chatbot engaged with children - generating full stories, asking follow-up questions, and exhibiting facial animations. Cues which have been shown to make robots appear more social and mindlike to children.
These anthropomorphic features can also appeal to adolescents, with one third of teenagers choosing AI over humans to have serious conversations. AI is always available, doesn’t exert social pressures, has no emotional needs and never creates conflict, what does this mean for creating the social skills needed to build and maintain real relationships? Professor Kim’s research shows that more than half of teenagers and parents surveyed use companion AI, with a majority of teens expressing a preference for chatbots which use relational language – considering it to be more likeable, trustworthy and emotionally close. This was particularly true of teenagers who indicated a lower quality of family and peer relationships and who expressed more stress and anxiety symptoms. Despite this, relational chatbots were not viewed as significantly more helpful than those which use colder, non-emotive language. There is a real risk then, that adolescents who already have weaker human networks may turn more toward AI companions and become increasingly isolated. Is this risk worthwhile considering that non-relational chatbots can still be helpful?
It is critical that we engage meaningfully with how we think AI systems should be designed and appropriately safeguarded. Should anthropomorphic features be switched off by default? What kind of relationships do we want our children to have with AI? Should there be a ban on autonomous attachment? These are all questions which parents, policymakers and young people themselves should ask.
If you are interested in hearing about future seminars, reach out to global.challenges@reuben.ox.ac.uk