In repeated rounds of a partner game, people who first avoided AI partners later came to prefer them, but do those shifts stick?

This explores whether repeated contact with AI changes what people prefer and how they decide, not just their first impressions. The clearest direct evidence in the corpus is about choosing partners. In partner-selection games with 975 participants, people at first avoided agents that were labelled as AI. Over repeated rounds they came to favour them, because the bots behaved generously and predictably, with less variation than human partners Do humans learn to prefer AI partners over time?. What changed was a learned preference, not a fixed attitude. The anti-AI bias was real, but experience wore it down. That suggests preferences about AI respond to how reliable it is in practice, not just to what people believe about it.

A second finding makes this more complicated: early effects don't always last. Long-term studies of the chatbot Mitsuku found that the social pull behind relationship-building faded as the novelty wore off. Results from a single session can't be relied on to predict behaviour over weeks or months Do chatbot relationships lose their appeal as novelty wears off?. Put the two findings side by side and you get a useful distinction. Preferences based on how the AI performs, such as reliability and fair returns, seem to grow with repetition. Preferences based on the novelty of a social experience seem to shrink. Related work points the same way: a single strong cue like a voice creates a sense of social presence more effectively than piling on many weaker cues Do more social cues always make AI feel more present?.

Some preferences change because of who people become around machines, not because the machine has convinced them of anything. People who are likely to cheat prefer reporting to an online form over reporting to a person, because lying to a machine feels less costly Do dishonest people prefer talking to machines?. Here the shift comes from the person choosing the machine, not from the machine changing them. A machine without judgment changes which decisions feel acceptable. Design can also push the other way. Thinking assistants that ask reflection questions alongside giving advice led to better decisions than assistants that only gave answers Do reflection questions help people make better decisions with AI?. So whether repeated use strengthens or weakens someone's own reasoning depends partly on whether the AI hands over conclusions or prompts thought.

The lateral surprise is that the adaptation may run mostly one way. In human conversation, people gradually pick up each other's word choices, which builds rapport. Current conversational AI largely doesn't do this Why don't conversational AI systems mirror their users' word choices?. Even AI agents interacting with each other change their actions when other agents are present, but they don't converge in their language or ideas Do AI agents actually socialize with each other?. If humans adjust to AI while AI doesn't adjust back, the human side is doing the changing. At the scale of a whole society, that asymmetry is the worry behind 'gradual disempowerment': as AI takes over work that once depended on people who cared about the results, systems can slowly drift away from what humans actually want, with no single decision marking the shift Does incremental AI replacement erode human influence over society?.

A candid limit: the corpus has strong evidence on how people's choice of partner shifts and on novelty fading. It has little long-term data on whether everyday decision habits change after months of AI use, such as risk tolerance, trust in one's own judgment, or reliance on AI advice. That gap is worth noticing, because the short-term studies suggest the effects are real but go in different directions depending on design.