In a creative meeting, everyone first asks AI for several options, then brings the most appealing one to the table. The copy may be more complete, yet the ideas may also resemble one another more closely than before. A team cannot assume that better individual proposals automatically mean a larger field of possibilities.
In an online experiment with 293 writers, Doshi and Hauser allowed some participants to receive either one or up to five story ideas from GPT-4. AI assistance improved the average novelty and usefulness of individual stories, with especially strong benefits for less creative participants; at the same time, the AI-assisted stories were more similar to one another. The study tested eight-sentence short stories, so it cannot stand in for every kind of workplace creativity. It does, however, reveal how individual gains and collective difference can move in opposite directions.
Once AI enters a team's creative process, difference can no longer be treated as a by-product of personal inspiration. It has to become part of how collaboration is designed. Instead of having everyone begin with the same prompt, a team can preserve independent judgements before AI organises them, then explore separately around different users, constraints and failure conditions, and finally compare the assumptions that are genuinely distinct. AI can improve the finish of each proposal; the team remains responsible for ensuring that the available futures do not collapse into one.
https://doi.org/10.1126/sciadv.adn5290
Discover more from Geoffrey Chen
Subscribe to get the latest posts sent to your email.