Why We Treat AI Chatbots Like People — and Whether We Should

On Oct. 9, 2026, TechCrunch published a feature examining why people treat AI chatbots and robots as human, and whether they should. TechCrunch The question is old. The machines are new.

The feature centers on Sherry Turkle, who joined the MIT faculty in 1976 and researches the psychological tie between humans and computers. Turkle wrote the book "Artificial Intimacy," which had a book launch at MIT Future Fest. Harvard Science describes her as the psychologist who pioneered understanding of human-computer relationships.

In "Artificial Intimacy," Turkle wrote that "when drawn into exchanges with a relational artifact, people believe it cares for them and are wired to care for it in return." Her word choice is deliberate. She calls machines "evocative objects" and "relational artifacts," terms for systems that invite a social response and lead people to assign them aliveness, intention and care.

The TechCrunch account pairs Turkle's caution with researchers working to make talk with machines smoother. Pat Pataranutaporn founded the Cyborg Psychology research group at the MIT Media Lab. MIT's Computer Science & Artificial Intelligence Lab houses humanoid Unitree robots costing about $50,000 each, including one nicknamed Gabe.

MIT News covered the same book in an article titled 'Who we become when we talk to machines' about Turkle's 'Artificial Intimacy.' MIT News That book presents a critique of chatbots and the antisocial dynamics Turkle links to them. As summarized in a Sept. 30 review, Turkle warns in 'Artificial Intimacy' that seeking comfort from A.I. devices endangers well-being and democracy. The New York Times

The argument has a long history. Turkle wrote "Relational Artifacts, Children, and Elders" as a National Science Foundation report. In earlier writing on relational artifacts for children and the elderly, she called nurturance, the act of caring for something, the new "killer app," the use that drives adoption, and wrote that people attach to what they nurture. She lists a presentation titled Relational Artifacts and Life-Practice Sociabilities: What 'Counts' as Alive.

That line of study continued into the generative era, when AI learned to produce fluent text and conversation. In 2024, Turkle wrote 'Who Do We Become When We Talk to Machines?', a qualitative study based on interviews that looked at the social and emotional effects of generative-AI conversational programs. She was a panelist at AI + the Future of Relationships in Cambridge, MA in April 2026. Harvard Science listed a book talk event featuring Turkle in conversation.

The broader context here will be familiar to anyone who builds conversational systems. Anthropomorphism, treating a machine as human, is not a user error. It is the default setting of a social brain meeting fluent language, turn-taking, a name, a face or a remembered preference. Theory of mind, our fast habit of guessing at another mind's intent, fires quickly, and apologies, repairs and first-person pronouns pull harder. Anyone who has watched children give a household device a personality will know the pull without needing the theory.

In my view, that is why the Turkle and Cyborg Psychology positions are less opposed than they appear. One asks what sustained artificial intimacy costs in daily human practice. The other asks what close coupling between person and model can add to memory, thought and creative work. Both accept that attachment will happen. The design question is whether the system admits the imbalance, keeps friction where friction protects judgment, and leaves the user more able in human ties afterward.

Looking at what this means for builders and operators, the practical controls are unglamorous. Persona limits, clear machine identity, bounds on simulated care and memory, and logging and tests for emotional dependency and sycophancy, or agreement meant to please. For enterprise and developer use, the same pull shapes trust calibration, handoff to a human, and audit of high-stakes advice. Handled with that care, social fluency in machines can lower the cost of explanation, tutoring, accessibility and prototyping. The risk is not that people feel something. The opportunity is to build systems worthy of the care they will receive.