Can a Fool with AI Beat a Genius? What Research Says About the New Skill Gap

“Even if you are not the smartest person in the room, could AI help you beat the person who is?” It sounds deliberately provocative. But recent research suggests the idea is no longer something we can dismiss as a joke.

Can a Fool with AI Really Beat a Genius?

First, a clarification. I use words like “fool” and “genius” here as shorthand, not as scientific categories or judgments about human worth. The studies discussed below compare things such as prior performance, education, experience, skill level, and business results.

And once you look at those measurements, something interesting appears: AI often helps the weaker performer more than the stronger one.

Conceptual illustration of AI narrowing a performance gap. It is not a recreation of any specific experiment.

At Work, AI Is Already Helping Weaker Performers Catch Up

A major study of 5,172 customer-support agents found that access to a generative-AI assistant increased productivity by an average of 15%.

But the average hides the interesting part. Less experienced and lower-skilled workers gained much more. Workers in the lowest skill quintile improved productivity by about 36%, while the highest-skilled workers saw little measurable productivity gain.

The experience gap also shrank surprisingly fast. Workers with only two months of experience who had AI assistance performed about as well as workers with more than six months of experience who did not have it.

AI was not merely providing information. It was compressing part of the value that normally comes from months of experience.

A separate experiment involving 758 Boston Consulting Group consultants produced a similar pattern. On creative product-development tasks that fell within GPT-4’s capabilities, lower baseline performers improved by 43%, while higher baseline performers improved by 17%.

That does not mean the weaker consultants suddenly became better thinkers in every respect. It means that, for tasks AI handled well, the gap in visible output became much smaller.

Conceptual illustration of AI accelerating workplace productivity and reducing part of the advantage created by experience.

At School, AI Can Make You Look Smarter Without Making You Smarter

Education is where the story becomes more uncomfortable.

In a field experiment involving nearly 1,000 high-school mathematics students in Turkey, researchers tested GPT-4-based tutoring systems. Students using a relatively unrestricted GPT interface performed 48% better on practice problems than students without AI assistance.

So far, that sounds like a spectacular success.

Then the AI was removed.

On the subsequent unassisted exam, students who had used the unrestricted GPT system scored 17% worse than students who had never been given AI access.

They had become better at completing the task while AI was available. That did not necessarily mean they had become better at doing the task themselves.

The study also tested a more carefully designed AI tutor that guided students with teacher-developed hints instead of simply making answers easy to obtain. With those safeguards, the negative learning effect was largely eliminated.

That distinction matters. The same technology can function as a tutor—or as a very sophisticated crutch.

Conceptual illustration of the difference between AI-assisted performance and learning that remains after the AI is removed.

Then Business Owners Turn the Story Upside Down

If AI helps weaker students and workers catch up, you might expect the same thing to happen to business owners.

Not necessarily.

A field experiment with entrepreneurs in Kenya gave business owners access to a GPT-4-powered AI business adviser. On average, the researchers could not establish a clear overall improvement in revenue and profit.

But once the entrepreneurs were divided by their previous business performance, the results became much more interesting.

Higher-performing entrepreneurs may have improved their business results by more than 15%. Lower-performing entrepreneurs, by contrast, did nearly 10% worse after receiving the AI assistant.

The researchers found that this difference did not appear to come mainly from asking different questions or receiving fundamentally different advice. A key difference was which pieces of advice the entrepreneurs chose to implement.

That changes the entire argument.

When the job is “write this email,” AI can produce something close to a finished product. When the job is “should I raise prices, spend money on advertising, hire someone, or launch a new product?” AI can offer options—but a human still has to risk real money on one of them.

At that point, AI may stop reducing differences in judgment and start amplifying them.

Conceptual illustration: AI can provide options, but the business outcome still depends on human judgment and execution.

AI May Be Breaking Our Old Ways of Recognizing “Smart People”

A 2026 randomized experiment involving 1,174 adults aged 25 to 45 makes this shift even clearer.

Without AI, participants with more education outperformed those with less education by 0.548 standard deviations on a workplace-style problem-solving task. With generative AI assistance, that difference fell to 0.139 standard deviations.

In other words, roughly three-quarters of the original education-based performance gap disappeared while AI was available.

Importantly, both groups benefited. The less-educated group was not simply pressing a magic button while the more-educated group failed to understand the technology. Higher-education participants still used the system somewhat more effectively. The underlying human-capital gap had not vanished.

What changed was the output.

For decades, we have treated certain visible abilities as signals of intelligence and competence: writing clearly, summarizing information quickly, producing polished documents, generating ideas, finding information, and sounding knowledgeable.

Generative AI can now supply part of that package to almost anyone.

That means the difference between people may move somewhere less visible.

What should you ask the AI? Which answer looks suspicious? Which option should you reject? Which advice deserves real-world action?

There are already tasks where a weaker performer with AI can beat a stronger performer without it. But once both people have access to AI, the contest changes again.

The most vulnerable person in the AI age may not be the person who knows the least. It may be the person who cannot tell when an intelligent-sounding answer is wrong.

Editor’s Note

Japan has an old saying: “Even fools and scissors are useful if handled the right way.” It is a very old-fashioned line, but AI has given it a strange new meaning. The person once dismissed as the one being “handled” can now grab AI and become the one doing the handling. Maybe the modern version is simply: “Fools and AI are useful if handled right.” The slightly nasty twist is that AI can also make a fool feel like a genius. That part may be harder to laugh about.

References