AI Now Feeds Interviewers Questions Mid-Interview — and It Feels Weird

Researchers at the University of British Columbia wanted to know what happens when you put an AI in the middle of a conversation between two people. They built a prototype called ProbeAssist that listens to a job or research interview and quietly feeds the interviewer suggested follow-up questions — the kind of "tell me more about that" prompts that keep a conversation flowing. Then they sat 18 people down for three practice interviews each: one with no AI, and two with different versions of the tool.

The results were mixed in a very human way. Interviewers were happy to have the AI as a helper. It reminded them of things they'd forgotten to ask and took some mental load off. But the moment the AI felt like it was judging them — grading their questions, or coming up with better ones — they pushed back. Nobody wants a machine looking over their shoulder and doing their job better than they do.

Four tensions kept surfacing. Agency: who's really running this interview? Ownership: whose questions are these, mine or the AI's? Creativity: am I still thinking, or just reading a screen? And presence: if I'm glancing at suggestions, am I actually listening to the person in front of me? The researchers noted that even helpful AI carries a hidden social cost — it can make people feel watched, distracted, or diminished.

Their advice for anyone building these tools: reduce the social pressure the AI creates, use it to spark ideas rather than dictate them, and never let it crowd out the human-to-human connection. The paper is under review for CHI 2027, a major conference on how people and computers interact. The takeaway for the rest of us: as AI moves from writing emails into live conversations, the hardest problems won't be technical — they'll be about manners, trust, and who gets credit.