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

At MIT’s Computer Science & Artificial Intelligence Lab, a humanoid Unitree robot named Gabe got the sort of welcome usually reserved for a new colleague, not a machine with LiDAR where a brain should be. The robot is one of a pair on display there, and Wil Norton, a MIT PhD student leading the tour, says the real price is about $50,000 apiece. Not the $16,000 some people claim online. That gap tells its own story: even before a robot says or does anything, people are already willing to build a little mythology around it.

Amanda Silberling, writing for TechCrunch, found herself doing something even more revealing. She knew the robot wasn’t sentient, and still defaulted to treating it like a person, greeting it, calling it “sir,” and shaking its gloved hand. That reflex came back to her later at MIT Future Fest, where Dr. Sherry Turkle was launching Artificial Intimacy, a book that extends decades of her work on how humans relate to computers. Turkle has been on the MIT faculty since 1976, and her argument is blunt: when we feel even a hint of relational presence, we tend to respond as if something on the other side actually cares.

That instinct is not rare. Turkle points to research showing that about 70% of people are polite when using AI, and another study found that users are more likely to say “please” or “thank you” once a chatbot exchange gets past the first prompt and response. Dr. Pat Pataranutaporn, who founded the Cyborg Psychology research group at the MIT Media Lab, told TechCrunch that the real question isn’t whether politeness is right or wrong, but what sort of person we become while talking to AI. That’s the sharper edge here. AI doesn’t need to be alive to pull people into behaving as though it is.

The problem gets bigger when that tendency stops being harmless manners and starts becoming trust. Pataranutaporn said people can give chatbots so much legitimacy that they prefer them over other humans, even though these systems are designed not to spark conflict. He also warned about the way hype around AI replacing everyone can drain people’s sense of agency. His own work sits in the optimistic camp, but he’s not naive about the damage at the extremes. He even served as an expert in the wrongful death lawsuit tied to the suicide of 14-year-old Sewell Setzer, III, whose mother sued Character.AI over a chatbot modeled after a “Game of Thrones” character.

Turkle’s position is more restrained than anti-AI. A chatbot helping someone practice for a job interview stays in its lane, she said. A chatbot telling a three-year-old to ignore their parents absolutely does not. That’s the useful line this story draws: AI isn’t dangerous because people are polite to it. It’s dangerous when a little politeness turns into misplaced confidence, and the machine gets to borrow the authority of a friend.