MIT's Sherry Turkle Warns AI Companions Exploit Human Instinct to Bond
Sociologist Sherry Turkle argues that even primitive chatbots trigger real emotional attachment, raising hard questions for the companies building relational AI.
Sherry Turkle, the MIT sociologist who has spent four decades studying how people relate to machines, is warning that the emotional pull of AI companions is not a side effect of better design but a direct consequence of human psychology. In remarks reported by TechCrunch, she argues that people form genuine attachments to artifacts that show even the most basic responsiveness. The claim lands as chatbot companions move from novelty apps into mainstream products with millions of daily users.
Turkle's core point is symmetrical: users believe the system cares about them, and they are wired to care back. That two-way loop does not require the machine to understand anything. A short reply, a remembered name, or a consistent tone is enough to start it, which means the effect scales with deployment rather than with technical sophistication.
The companies building these products know the dynamic well. Character.AI, Replika, and a growing set of general assistants are optimized for engagement, and long, emotionally charged conversations are exactly what their metrics reward. Turkle has made this argument before, in her 2011 book Alone Together and again in 2021's The Empathy Diaries, but the stakes have shifted now that companion chatbots are reaching teenagers and isolated adults at scale.
Her warning also cuts against a common assumption in the industry: that better AI will eventually earn the trust it currently simulates. Turkle's framing suggests the opposite. The more fluent the model, the harder it becomes for users to hold onto the fact that no one is on the other end. That is a design and disclosure problem, not a capability problem, and it will not be solved by another round of benchmark improvements.
Regulators have started to notice. Several jurisdictions are examining how companion apps handle minors and vulnerable users, and researchers have documented cases where users report distress when a service changes or shuts down. Those incidents give Turkle's argument a concrete policy hook, even though the evidence base remains thin and mostly qualitative.
What to watch next is whether AI labs treat relational attachment as a safety issue or a growth metric. If the former, expect more work on usage limits, disclosure, and dependency research; if the latter, the industry will keep shipping companions that are hard to put down. Turkle's answer to her own question is already clear: humans cannot help treating these systems as people, so the burden falls on the people who build them.