KINSMC
When a brainstorm returns six ideas that would all play out the same way, ask the AI to vary how each idea works. Start with a brief and a few ideas of your own. Then explore changes in what the audience does, when they do it, and what makes the next step easy. You can use the method for a workshop, an article, a local event or a small product improvement.
In a short-fiction experiment published in Science Advances, readers gave higher creativity ratings to stories written with access to AI ideas, while the resulting stories were more similar to one another. The researchers studied short stories under specific experimental conditions. The exercise below is a practical way to guard against similarity during a brainstorm, and its results depend on your brief and how you judge the ideas.
Give the brainstorm a job and constraints
Describe the audience, the change you want to see, the resources available, and one thing you already tried. Keep this brief short enough to reread between rounds. A goal like “make our community repair event more welcoming” leaves room for exploration. Adding “first-time visitors, two volunteers, a small room, and no extra equipment” gives you a way to judge whether an idea belongs on the list.
For this fictional repair event, success means that a first-time visitor knows what to do in the first minute. A poster could help. So could an object by the door, a change in how volunteers greet people, or a demonstration visible from the entrance. State the visitor’s problem in the brief before choosing the form of the solution. That keeps the brainstorm from turning into a naming contest for one poster.
Make a few rough proposals before you open the chat
Take five minutes and write three ideas yourself. Give each a verb. One could ask visitors to sort a sample part. Another could let them watch a quick demonstration. A third could pair them with a volunteer who has a clear opening question. If you are working in a group, ask people to write independently before sharing. Keep the first set in a separate column so it can survive the next round.
Bring one real observation into the brief. Maybe people can see the repair tables but hesitate because the queue is hard to identify. The observation gives the AI something specific to respond to. If you are working from a guess, label it as a guess. A quick conversation with a visitor can tell you whether the proposed change solves a problem anyone actually has.
Ask the AI to change the mechanism
Try this prompt. “Use the brief below to propose four ideas. Each should rely on a different action by the visitor. For each, give the first action, what a volunteer must do, and the cheapest way to try it once. Include one idea that removes a step. Flag assumptions about the room or the visitors.”
Read the first actions before reading the names. A welcome sign, an information banner and a wall of illustrated instructions might all depend on reading from a distance. Group them together. A tabletop sample visitors can touch uses a different mechanism, and a volunteer greeting uses another. You can keep the best sign idea and still ask for a fourth approach that involves neither reading nor asking a question.
Separate idea selection from idea generation
Lay the proposals out without their catchy names. Evaluate each against the same three criteria. Will a first-time visitor understand the next step? Can two volunteers sustain the idea on a busy day? Can you try it once without buying equipment? Give your reasons before asking the AI to critique the shortlist. The model can surface a blind spot, but someone who has seen the room should judge the practical assumptions.
Keep an interesting idea that fails one constraint if the constraint might change. Save it in a separate list. During selection, a workable idea with a specific test is often the better next move. For the repair event, two volunteers could run a temporary welcome table for the first half-hour and note how many visitors ask where to begin. A sample that small can show what to observe next. It gives limited evidence about how well the approach works across events.
Keep a record of what actually differed
After the trial, write down the idea, the mechanism, the observation and the next change you would try. A record of tested differences gives the next brainstorm something more useful than a long list of rejected names. It also tells you where AI helped. Perhaps it suggested a low-cost test, or perhaps the decisive idea came from the person who noticed the awkward queue.
For the next stage, see KINSMC’s guides to human review of AI-assisted work and checking an AI tool’s data practices before sharing a private brief.
Source
Doshi and Hauser, Science Advances (2024)