AI and Thinking Tools: A Partner in the Brainstorm, Not a Replacement

In 1987 Michael Diehl and Wolfgang Stroebe gathered the earlier studies of brainstorming and found that groups exchanging ideas face to face usually produce fewer ideas than the same people working separately. That was so in 18 of 22 studies. This paradox helps us see where AI belongs among thinking tools.

Four tools with known origins

Brainstorming. The term became widely known through the advertising man Alex Osborn's Applied Imagination (1953). Its rules are four: go for quantity, withhold criticism, welcome wild ideas, and combine and improve ideas.

The Six Thinking Hats. In Edward de Bono's Six Thinking Hats (1985), the thinker takes on six roles in sequence, not all at once. White is for facts, red for feelings, black for criticism, yellow for the positive, green for the new, and blue for managing the session.

TRIZ. The theory of inventive problem solving was begun by the Soviet Genrich Altshuller in 1946, when he worked in the inventions inspection department of the Soviet Navy's Caspian Sea flotilla. With colleagues he drew recurring patterns from hundreds of thousands of inventions and arrived at forty inventive principles.

SCAMPER. Devised by Bob Eberle in 1971 from Osborn's checklist of questions: substitute, combine, adapt, modify (magnify or minify), put to other uses, eliminate, and reverse or rearrange.

There is no magic in these tools. They are organised questions that stop the mind halting at the first idea.

What experiment says about brainstorming

Why do groups lose out? Diehl and Stroebe tested three causes: free riding (relying on others), fear of evaluation, and "production blocking", meaning that only one person can speak at a time. They concluded that production blocking explained most of the loss. In a key experiment, those made to wait for their turn produced less than half as many ideas as those who could speak freely.

The practical moral, before any language model enters the picture: let everyone write their ideas alone, then pool them.

And what does the language model add?

The model resembles a participant who does not wait for a turn, does not feel shy and does not tire. The researchers Doshi and Hauser measured its effect in an experiment published in Science Advances in 2024: participants wrote short stories, and some received ideas from a language model. Stories by those who received ideas were rated more creative, better written and more enjoyable, with a larger effect for less creative writers. But those stories were more similar to one another than stories written without help. The authors described this as a social dilemma: the individual gains while overall diversity shrinks.

Practical suggestions

These are inferences from the two studies, not the result of a direct experiment:

• Write first, then ask. Record your ideas before seeing the model's, so that it does not narrow your horizon.
• Make it the black hat. Ask it to list what might fail in your idea or to argue against it; this is usually more useful than asking for praise.
• Draw out questions, not answers. Ask it for SCAMPER questions about your project, then answer them yourself.
• Do not accept the first reply. Repeat the request from different angles; sameness of answers is the risk the experiment observed.

After generation, one question remains that cannot be delegated: is this idea beneficial? Is it consistent with our values? Does it carry a hidden harm? That judgement is borne and answered for by a human being; a machine is not called to account. And in this there is proper reliance on God: we take every means, and ask God for success.