When Kellogg professor Brian Uzzi wanted to challenge how his students thought about artificial intelligence, he started by giving them a simple test called the Divergent Aptitude Test (DAT), which measures general creativity. Test-takers have four minutes to come up with a list of ten words that are as different as possible from one another. They then ask a chatbot to do the same.
The test is hard. Every new word you add to your list can be simultaneously far away from one word but close to another. Most of the students expected the machine to outperform them. But a funny thing happened: it didn’t. The class average matched the bot’s, almost exactly. And a few students came up with lists of words that were far more original than the chatbot.
Since the DAT is widely administered, Uzzi had a sample of thousands of human test scores and hundreds of thousands of bot test scores to review alongside his class. He found that bots, while potentially having access to tens of thousands of words, tend to populate their ten-word lists from a narrow corpus of about 850 words.
“Human language has around 50,000 words, and that’s where all the separate human perspectives can be so powerful,” Uzzi says. “The machine plays the greatest hits, so to speak, over and over again—and it misses out on those gems.”
With this simple experiment, Uzzi showed the students how AI tends to produce average ideas, while people can come up with more unexpected ones. The same principle applies to creative work in business, Uzzi says. When people rely on AI for quick answers, the bot’s lack of creativity can narrow their own thinking.
To Uzzi, this isn’t an argument against using AI. Instead, it’s a reminder that there are more-valuable ways to use the technology than letting it brainstorm for you.
“To get the most out of a bot, don’t ask it for answers,” the professor says. “Ask how to approach a problem. You want advice on how to think, not what to think.”
Uzzi offers advice on how to collaborate with—not delegate to—AI on creative tasks.
How AI dulls creativity
When conducting the experiment with his class, Uzzi noticed another stubborn pattern among students: even when they knew the bot wasn’t more creative than they were, some still preferred its answers. The reason was simple: speed.
“They get sucked in by the efficiency,” Uzzi says. “Someone in class will say, ‘The bot’s score is no better than mine, but I get it in 10 seconds instead of several minutes.’ To them, that feels like a good trade-off.”
That logic, he warns, can be a trap—the quick return of responses can silently dull creative thinking. Uzzi says that once people get an answer from the AI, they tend to defer to it, even when it is not terribly creative. In the experiment, students spent about four minutes on the task of generating their own disparate words. But when they are given an opportunity to use the bot and then to edit the bot’s list of words, most stop after a minute and make only a few changes.
“They get anchored to what the bot gives them,” he says. “It gives them mediocre answers and generic answers that lack differentiation from others. When those two kinds of answers happen in combination, innovators unintentionally overlook the opportunity to add their own experiences and insights into the creative mix with the bot, which is the sweet spot that can make human and bot creativity most profound.”
Slow down to get broader ideas—even from AI
Uzzi noticed that once students start deferring to the technology, efficiency rather than discovery becomes the goal. And isn’t a brainstorm about being creative and original, rather than being finished first?
“Speed is important,” Uzzi tells his students, “but differentiation is ultimately the most important thing, because if you’re fast and someone else could copy you just as fast, then whatever advantage you had, you’ve lost.”
True creativity, he adds, depends on the process of how people generate and develop ideas, not how quickly they reach the finish line. And that process begins not just with accumulating the best answers, but with developing the right questions.
Uzzi’s experiment showed that instead of asking the bot for an answer—“give me 10 words that are as different as possible”—asking the bot how you should approach the task will give you the best outcome.
When framed that way, he says, the AI can outline a process—a heuristic—that will lead to a wider array of words. The bot’s heuristic is to approach the DAT test in two steps. The bot suggests starting by coming up with ten broad categories of words, such as science, business, or art. After the ten categories are chosen, it suggests choosing one word within each category. In this method, the anchoring that occurs when humans follow the bot is removed and the tendency for innovators to get mentally trapped is lessened. In other words, once a test-taker thinks of a word—say “cat”—their mind tends to reach for related words like “dog” or “lion,” which inhibits their creative thinking. Following the two-step method, people’s creativity score “just skyrockets,” Uzzi says.
Build alliances with AI
The realization that AI works better at processes than products led Uzzi to develop a framework for identifying and leaning into the different roles the technology can play—whether it’s automation, answers, or alliances.
“Automation” means using AI to copy a process that humans already do—think churning out a large number of simple email responses. “Answers” involves using AI to track down a specific answer that already exists—similarly to how web searches functioned in the past. Finally, “alliances” entails working with AI in areas that draw on human creativity—such as Uzzi’s “ten words” experiment with his students.
“It’s a pyramid,” Uzzi says. “The base is automation—which includes most of the applications which speak to scaling and efficiency. As you move up the pyramid, you get fewer applications—but those you get are likely to be bigger in impact. It’s the level that addresses coming up with new solutions and adapting to environmental changes.”
At the top of the pyramid are the kinds of work that rely on creativity and imagination—things like strategy and innovation. “That top part of the pyramid—that’s going to be most important.”
This framework can help organizations think about how to apply AI for business decisions. A director, for instance, might ask AI how to think about where to open a new factory. The system would outline the key factors to consider, and the director could then add their own knowledge before asking the bot for feedback.
Uzzi calls this kind of back-and-forth “the best kind of collaborative AI work.”
Remember the creative power of teams
To demonstrate the power of human collaboration, Uzzi asked his students to repeat the same creativity test—listing ten words as different from one another as possible—but this time in teams, without AI. Their scores on average rose above their individual scores and the individual bot scores. But when they added the bot as a “team member,” their creativity scores dropped again when they relied on the bot for what to think rather than how to think about a problem. Teams that asked a bot how to think did better than individuals and better than bots on the creativity test.
Uzzi’s findings point to a simple truth: creativity remains a fundamentally human endeavor. It depends on networks of people who challenge and build on one another’s ideas. AI can assist in that process, but it cannot replace it. The most effective teams, Uzzi suggests, will use AI to expand the way they think—not to shortcut it.
“Creativity is a very human activity,” Uzzi says. “It comes about mostly through collaboration. We all stand on the shoulders of other people. If it’s just you and the bot—or if you rely on the bot for answers—you’ll get an answer fast, but others can get the same answer just as fast. If you want a competitive advantage that builds on what makes you special, don’t ask a bot what to think, ask a bot how to think.”