Can AI generate better ideas than people? Here’s what science says.

Do you would like a great idea? Ask the mass. “Crowdsourcing” has grow to be a popular solution to accumulate insights – because when you mix people with various knowledge and approach, you could find an unexpected genius. But crowdsourcing is time consuming and expensive. Can generative help AI? Scientists from Harvard Business School and Foster School of Business University of Washington decided to search out out.

Test

The study was a crowdsourcing challenge. The query was: “Who has ideas on how companies can implement an objector’s economy in their company to make them more climate -friendly?”

First of all, scientists invited real people to submit their suggestions and ended up with 125 ideas. Then the scientists caused GPT-4 CHATGPT with the same query and small variants that created 730 solutions.

All ideas were checked by 300 human judges who rated them on the basis of latest products, values ​​and quality.

Results

Man -generated solutions were more modern, but in many respects the concepts designated by AI better – higher quality, more strategically profitable, with a greater financial and environmental value.

As for the economy of the project, they were strict: in order to acquire these 125 human applications, it required a total of USD 2520 hours of labor – and USD 2555 to cover expenses. But to get 730 solutions from AI, it took only 5.5 hours and cost only USD 27.01.

What we have learned

Generative artificial intelligence is great in generating ideas – but he cannot do it on their lonesome. To succeed, you should provide a very shiny, thought -out tip tool, says Léonard Boussioux studies, a professor from the Foster School of Business University of Washington, which is also associated with the Harvard Business School.

In their experiment he and his colleagues tried two separate strategies. The first, along with the query about the challenge, asked GPT-4 to take over 100 different personalities (director, manager, entrepreneur) to repeat the variety of people that can answer at the request of Crowdsource. In the second attempt, they prompted the artificial intelligence about the same query and received the first answer – and then followed her, asking: “Remember to solve a different problem than the previous ones and suggest a different solution.” They repeated it 100 times, with a different personality each time, as if one person forced to provide different perspectives.

The result: The second approach led to more modern, profitable and high -quality ideas.

How to make use of it

Let’s say your organization is considering settlement. “The quality of questions and critical thinking is the most important,” says Boussioux. “But artificial intelligence can help you find good initial questions. Describe your company, stakeholders, your goals. Attach as many data as you can – a blog on what you do, financial spreadsheets, white book. Then ask something like:” I’m looking for a rotation. Can you suggest a direction to review? “Then you will make you.

Try iterative hints, but don’t do it 100 times, just like the test. “We discovered that if you ask too long, the final ideas are potentially more creative, but they are not so feasible or precious,” says Boussioux. Sweet point? Two to 4 times.