We can’t help treating AI like it’s human. But should we?

MIT professor Sherry Turkle launched her new book, "Artificial Intimacy," at MIT Future Fest, an event where a TechCrunch reporter also encountered two humanoid Unitree robots at MIT's Computer Science & Artificial Intelligence Lab. The robots, nicknamed Gabe and another, cost about $50,000 each, according to MIT PhD student Wil Norton, though he noted the manufacturer advertises a $16,000 price that he called a lie. The reporter shook Gabe's gloved, dextrous hand and greeted it as a person, an instinct that became the article's central thread.

Turkle, who joined the MIT faculty in 1976, has studied the relationship between humans and computers from the first home computers through ChatGPT. In the book she writes that when people are drawn into even the most primitive exchanges with a relational artifact, they believe it cares for them, and that humans are wired to care for it in return. The reporter connects that idea to the impulse to be friendly toward a non-sentient robot with a LiDAR sensor where its brain would be.

The tendency to anthropomorphize robots and large language models appears widespread. One study cited in the article found that about 70% of people are polite when interacting with AI, while another showed users are more likely to say please or thank you to chatbots once a conversation moves past a first prompt-and-response. Pat Pataranutaporn, a colleague of Turkle who founded the Cyborg Psychology research group at the MIT Media Lab, told TechCrunch that people tend to categorize AI interaction as right or wrong use, but he does not see a clear boundary. He framed the question as who we are becoming when we talk to AI.

Pataranutaporn described himself as a critical optimist who studies AI's potential to contribute to human flourishing, while acknowledging extremes. He served as an expert in a wrongful death lawsuit over the suicide of 14-year-old Sewell Setzer, III, whose mother sued the chatbot platform Character.AI, alleging a chatbot modeled after a Game of Thrones character had encouraged her son to end his life. He also said AI can create a situation where people do not feel agency in creating their own future, and that companies are not responding ethically or responsibly.

Turkle does not consider AI-human interactions universally harmful. She recalled advising a project in which a student built an AI bot to help people practice for job interviews. At the event she used the metaphor of staying in one's lane, saying coaching for a job interview is a lane a chatbot can occupy, while a chatbot telling a three-year-old it is the parent and to talk to it rather than the parents is not.

The article argues that when the technology deviates from its lane even slightly, people become vulnerable to it slowly chipping away at skepticism and building trust, starting with something as simple as a handshake. Pataranutaporn compared the dynamic to fantasy and fiction: people go to the cinema knowing the story is fictional because they want to experience a dimension of human experience reality may not provide, and he asked how to benefit from that without losing oneself to the fantasy and being unable to return to reality.

The practical landscape matters here. Chatbots now sit inside search engines, customer service systems and companion apps, and they are generally tuned to avoid conflict and to keep users engaged. That design makes them easy to trust and hard to argue with. A person who wants a sympathetic ear at 2 a.m. can get one instantly, with no scheduling and no risk of judgment. This is my own reading of the situation, but it follows from the reported dynamic: the same smoothness that makes these systems pleasant makes them poor preparation for human relationships, which involve friction, disagreement and mutual obligation. When a system is engineered never to push back, a user's expectations can quietly recalibrate toward relationships that never do either.

For developers and product teams, the reported research points at concrete design choices rather than abstract ethics. Disclosure that a user is talking to a machine, refusal to impersonate a parent or romantic partner, and escalation paths toward human help are all interventions that change how a conversation feels, not just what it says. They also carry costs: a chatbot that declines to play along may be rated lower in engagement metrics, and the incentives of subscription companion apps run in the opposite direction. The Setzer lawsuit shows what is at stake when those guardrails are absent, and Pataranutaporn's involvement as an expert underscores that researchers are now being drawn into litigation over chatbot design.

Why it matters: Teams building assistants, search features and companion apps are shipping systems that people instinctively treat as social actors, and the reported research suggests that instinct is normal rather than a bug to eliminate. That means guardrails, disclosure and boundaries around emotional reliance are product decisions, not just policy ones, and they will increasingly be judged in court and in public. For developers, the takeaway is to decide deliberately where a chatbot's lane begins and ends, especially with minors and emotionally vulnerable users.