AI help can make your idea better while making everyone's ideas sound more alike — what decides which one happens?

This explores the gap between how good one AI-assisted idea is and how varied a whole group's ideas are. They can move in opposite directions at once. The clearest case is a 293-writer experiment. Giving writers AI-generated story ideas raised evaluator ratings of their individual stories, most of all for less creative writers. But the stories became more similar to one another across the group Does AI assistance help individual writers or hurt collective creativity?. The authors describe this as a social dilemma. Each writer is rationally better off using the AI, but the group as a whole loses variety. Judging ideas one at a time can't show you that loss, because you only see it when you compare ideas side by side.

The surprising part is that the sameness seems to depend on what job the AI does, not just on whether people use it. A preregistered study found that AI which *generated* ideas reduced collective diversity for every kind of writer. AI that *refined* ideas people already had left diversity intact Does AI assistance homogenize or preserve creative diversity?. The same study found that non-native English speakers added more diversity than native speakers, and that AI ideation was what erased this advantage. So the variety lives in the people. AI damages it when it replaces their starting point and leaves it alone when it polishes what they bring. This calls for some care with the evidence. One widely cited study concluded that AI narrows ideas, yet it measured only how many ideas people produced and how detailed they were, never how varied they were Does AI assistance actually narrow the diversity of ideas?. Claims about AI and homogenization are only as strong as the diversity measure behind them.

Why would AI ideas converge in the first place? One reason is that the models themselves converge. An analysis of more than 70 models answering 26,000 open-ended prompts found an "Artificial Hivemind": different models, built by different companies, independently produce strikingly similar answers because they share training data and alignment methods Do different AI models actually produce diverse outputs?. So switching chatbots, or combining several, doesn't buy as much diversity as you might expect. A related argument from the philosophy side is that AI scales up *claims* without scaling up the *points of view* behind them. A thousand fluent AI articles may represent roughly one perspective Does AI generate diverse claims or diverse perspectives?. Quality and diversity separate here because fluency is cheap and a real point of view is not.

Work on multi-agent systems approaches the same tension from the other side. Teams of AI agents with different thinking styles can beat a single agent at ideation, but only when each member has real domain expertise. Diverse teams without expertise do worse than one competent agent, because the variety becomes noise rather than insight Does cognitive diversity alone improve multi-agent ideation quality?. Even inside one model, structuring reasoning as a dialogue between distinct voices produces more varied problem-solving than a single monologue Can dialogue format help models reason more diversely?. Taken together, the work suggests that diversity isn't a free byproduct of more ideas or more models. It comes from distinct, competent perspectives kept separate long enough to matter. For a human group using AI, that means letting people form their own ideas first and bringing the AI in afterwards to sharpen them, not the other way around.