More ideas can still mean less variety

When I was shopping for running shoes, I read 7 “TOP 10” articles and every one of them recommended something different. That frustration is how GrasShopper started. Lately I think about the other side of it. If everybody asks the same assistant, everybody may get pointed to the same few pairs of shoes. For shoes that’s probably fine, but for writing, design or music I’m less sure.

Anil Doshi and Oliver Hauser tested this with short stories. Writers who got access to AI-generated ideas wrote stories that were rated better individually, and the AI-assisted stories were also more similar to each other. It’s one experiment with one kind of task, so I wouldn’t generalize it to every use of generative AI. But it shows that the quality of a single result and the variety of the whole collection are two different things, and usually we talk about creativity as if one number covered both.

I don’t want to draw either of the easy conclusions from it. Telling a struggling writer that help is worthless because others have it too would be wrong. Assuming that a tool which helps each person also makes the shared culture richer would be wrong as well.

A text box with an instant answer makes you look at the one person using it. The effect on culture shows up across many people and over a longer time: everybody gets more productive while the references, plots, styles and endings they reach for get narrower.

I see it as a design problem. Compare two writing tools. One gives you a polished opening before you’ve made any decision. The other first asks you for an observation, something you disagree with and something you can’t explain yet, and then helps you look for a structure. Both generate text, but the initiative is in a different place.

So next to the questions I’d ask about any generator, I’d judge creative tools by whether they help people develop their own concerns: - Can I follow an awkward question instead of taking the most recognizable answer? - Can the tool work with local references it hasn’t seen a million times? - Can it challenge my first idea without swapping it for a generic one?

Part of the work is on the user too. Reading outside of what one service recommends, talking to people who’ve lived differently and spending time with real materials and places bring in things the defaults won’t supply. None of it guarantees originality.

I’m not after endless variation, because a thousand small differences can still say the same thing. What I want from creative software is that more people can go after the things that matter to them, including the ones the tool finds inconvenient to express.