Do you act differently toward a chatbot than toward a person — and if so, what flips the switch?

This explores whether people treat AI with the same social habits they use with other humans, or whether they build a separate set of habits just for AI. The corpus suggests the answer is 'both, side by side.' Long-running research in the 'Computers Are Social Actors' (CASA) tradition finds that people don't just copy their human-to-human manners onto machines. With repeated use they develop a second script made for media agents, and they apply it just as automatically as the first one (Do humans apply human-human scripts to AI interactions?). Their responses shift in measurable ways as AI use piles up. The AI script grows out of experience instead of replacing the human one.

Age appears to be one of the things that decides which script switches on. In a 244-person study, participants passed a ball more often to an AI agent that was being left out of a game. That's the same compensating move people make for an excluded human. Older adults did it more, and younger participants did it noticeably less (Do people treat excluded AI agents like excluded humans?). One plausible reading is that people who have spent more time with AI have a sturdier AI-specific script, so the human 'include the excluded person' reflex fires less readily. That's an inference from the pattern, not a finding the study states.

What sets off a social script at all is a smaller set of cues than you might expect. One strong primary cue, such as a voice or a face, is enough to make an AI feel like a social actor. Stacking up many weaker cues doesn't get there (Do more social cues always make AI feel more present?). Once people are engaged, they size up an AI partner along three lines: how competent it seems, how human-like it seems, and how flexibly it communicates. Competence carries the most weight, about half of the variation in impressions (How do users mentally model dialogue agent partners?). That may be what makes the AI script distinct. Mostly we judge AI by whether it gets things done, and we judge people by many other things too.

One philosophical framing fits these split scripts well. Murray Shanahan argues that a chatbot is best understood as playing a character. Folk psychology (reading beliefs and intentions into behavior) applies to the persona the system is playing, not to the system underneath (Should we treat dialogue agents as role-playing characters?). If that's right, a separate AI script isn't confusion. It may be a sensible adjustment: being polite to the character while knowing it's a performance.

There's an interesting twist on the AI side. GPT-4.5 predicts which behaviors people will find socially appropriate more accurately than any individual human (Can AI learn social norms better than humans?). Yet it can't take part in the community processes that create and revise those norms (Can AI predict social norms better than humans?). So the AI knows our scripts very well from the outside, while we are building a new script for it. The corpus has a lot on how the AI script forms and what triggers it. It has less on whether the two scripts leak into each other, for example whether bluntness learned with chatbots carries over to how people treat other humans. That question is still open.