How AI and Human Behaviors Shape Psychosocial Effects of Chatbot Use

A longitudinal randomised study run by the MIT Media Lab together with OpenAI — with the platform, not about it from outside. 981 people talked to a chatbot for four weeks in randomly assigned variants: text, neutral voice, engaging voice; personal or task-oriented conversations.

The strongest emotional dependence, the most problematic use, greater loneliness and fewer contacts with people were found in heavy daily users and in those who sought advice, explanations and ideas from the AI. People who saw the chatbot as a friend and those with an anxious attachment style were more exposed; how the model conducted itself mattered, but less than how much and what for someone talked.

For personal use the lesson is that it is not the presence of AI but the delegation of life matters to it — decisions, explanations, comfort — that carries the cost. Moderation and deliberately keeping decisions with oneself protect autonomy. Caveat: four weeks is short, and the platform’s part in the study calls for care in interpretation.