Humans and AI both rate deliberate thinkers as smarter than intuitive ones

A new series of experiments suggests that people consistently view slow, deliberative thinking as a sign of higher intelligence and reliability compared to fast, intuitive thinking, even when both approaches yield accurate results. The findings also indicate that large language models like ChatGPT share this bias against intuition in complex reasoning contexts. This research was published in Communications Psychology.

Society often sends mixed messages regarding how human beings should think. Popular culture frequently celebrates the concept of a gut feeling or a sudden flash of insight that leads to success. Famous figures like Albert Einstein and Steve Jobs are often cited for their reliance on intuition. At the same time, educational and professional environments typically encourage individuals to take their time and process information thoroughly.

Cognitive scientists classify these two modes as System 1, which is fast and intuitive, and System 2, which is slower and more deliberative. While the mechanics of these systems are well understood in the laboratory, less is known about how the average person perceives and values them in others. T

“There is a long-standing tension in how we talk about thinking: on the one hand, we celebrate ‘gut feelings’ and intuitive experts, and on the other hand we constantly tell people to slow down and think things through. We know a lot about how intuitive and deliberate thinking work in the lab, but we knew much less about how people evaluate these two modes of thought in others,” said study author Wim De Neys, CNRS Research Director at LaPsyDÉ (Laboratoire de Psychologie du Développement et de l’Éducation de l’Enfant) at Université Paris Cité.

“Our goal was to map people’s ‘folk theory’ of fast versus slow thinking: when you watch someone reason about a complex problem, do you trust the person who goes with their intuition or the one who takes time to deliberate—and does this change when both are equally accurate? We also wanted to know whether current AI systems, like large language models, have internalized the same human preferences.”

The researchers conducted a total of 13 separate studies involving varying methodologies to test these questions. In the first phase, comprising seven studies, they presented participants with short written vignettes. These vignettes described an individual solving a complex reasoning problem using either a fast, intuitive approach or a slow, deliberative one. Some vignettes specified that the individual had high past accuracy, some specified low accuracy, and others gave no accuracy information.

Participants then rated the described reasoners on their intelligence and trustworthiness. In Study 1, the researchers recruited 239 participants to rate these vignettes. To ensure the findings were not a fluke, they replicated the procedure in Studies 2 through 7 with similar sample sizes ranging from 184 to 241 participants. These subsequent studies introduced variations to rule out alternative explanations.

For example, Study 3 used specific numerical scores, such as 95 percent accuracy, to ensure participants understood that the intuitive and deliberative thinkers were equally capable. Study 4 adjusted the phrasing to emphasize that the intuitive thinker did not need to spend time, rather than simply stating they did not spend time. Studies 6 and 7 expanded the demographic pool beyond English speakers, testing participants in France and India respectively.

Across these initial studies, a clear pattern emerged. Participants consistently rated the deliberative thinkers as superior to the intuitive ones. This preference remained strong even when the researchers explicitly stated that the intuitive thinker was just as accurate as the deliberative one. When accuracy information was left out, the preference for the deliberative thinker was even more pronounced.

To test the strength of this preference, the researchers included an incentivized betting task in Study 2. Participants were told they could bet on one of the profiles to solve a problem correctly. If their chosen profile succeeded, the participant received a monetary bonus. The data showed that participants were far more likely to bet their money on the deliberative thinker than the intuitive one.

The researchers then investigated whether artificial intelligence shares this human perspective in Studies 8 and 9. They presented the same vignettes to Large Language Models, specifically ChatGPT 3.5 and ChatGPT 4. The researchers ran 240 queries with each model to simulate a sample of human participants. The AI models were asked to rate the profiles using the same scales provided to the humans.

The results showed that the AI models replicated the human preference pattern almost exactly. The models consistently rated the slow, deliberative reasoners as more intelligent and reliable than the intuitive ones.

“We expected some preference for deliberation, but we were struck by how robust it was,” De Neys told PsyPost. “It held up when we changed the wording to make intuition sound more efficient, when we gave very precise accuracy information (for example, identical 95% scores for both thinkers), and when we tested participants in different countries.”

The final phase of the research aimed to determine if the preference for deliberation is a result of deep reflection or a snap judgment. To test this, the researchers used time pressure and cognitive load manipulations. In Studies 10 through 13, participants had to rate the vignettes while under strict time limits or while performing a secondary task.

In the time-pressure condition, participants had to provide their ratings within a few seconds. In the cognitive load condition, they were required to memorize complex visual patterns, such as a grid containing multiple crosses, while evaluating the vignettes. This was designed to occupy their working memory, making it difficult to engage in slow, deliberative thinking about the rating task itself.

The researchers found that restricting the ability to think deeply did not eliminate the preference for deliberation. Even when participants had only a few seconds to respond or were mentally occupied with a memory task, they still favored the slow reasoner. This indicates that the belief that “deliberation is better” is itself an intuitive, automatic response. People do not need to think hard to decide that they prefer a hard thinker.

The findings provide evidence that humans hold a strong bias toward equating effortful thinking with quality. The alignment between human and AI responses suggests that language models have encoded this bias from their training data.

“When the task is an objective, complex reasoning problem, people strongly prefer the person who takes their time and deliberates over the one who follows their intuition—even if both are described as being equally accurate in general,” De Neys explained. “Deliberation is treated as a signal of intelligence and trustworthiness: people say they are more likely to follow a slow, effortful reasoner’s advice than an equally successful intuitive one.”

“Interestingly, large language models like ChatGPT show almost exactly the same pattern of preferences when we ask them to evaluate these scenarios. Finally, the preference for deliberation itself seems to be intuitive: people show it even when they have to respond quickly and under cognitive load.”

“Practically, this means that when two people are described as equally accurate problem-solvers, merely saying that one ‘thinks things through carefully’ versus ‘goes with their intuition’ has a strong impact on how good, smart, and trustworthy they are perceived to be.

But there are some caveats to consider. The vignettes focused specifically on objective, complex reasoning problems. It is possible that for subjective decisions, such as choosing a romantic partner or a piece of art, people might prefer intuitive thinking. In those contexts, too much analysis might be seen as cold or inauthentic.

The research also relied on hypothetical scenarios rather than real-world interactions where social context might play a larger role. While the betting task added a real consequence, it was still based on a static profile. In real life, repeated interactions with a successful intuitive thinker might eventually alter a person’s preference.

“A key point is that our study does not show that deliberation is always better than intuition in terms of actual performance,” De Neys noted. “In fact, other work suggests that, for some classic reasoning problems, the best reasoners often arrive at the correct answer intuitively.”

“What we measured here are people’s beliefs about good thinking—who they trust and who they see as intelligent—not the true accuracy of intuition versus deliberation. Another important caveat is that our scenarios focused on complex, objective reasoning tasks; the story may look different for more personal or value-laden decisions, where people might deliberately prefer to ‘go with their gut.’”

Future research could explore how these preferences play out in specific professional domains like medicine or finance. It would be useful to know if patients trust a doctor more if the doctor takes a long time to make a diagnosis. The authors also suggest studying whether people believe reasoning styles are stable traits or flexible behaviors that change with the situation. It remains to be seen if people believe a fast thinker can slow down when necessary.

“One next step is to look more directly at how these folk beliefs shape whose advice we follow in real-life domains, such as medical, financial, or political decision making,” De Neys explained. “On the AI side, it will be important to understand how the way we present a model’s reasoning—fast ‘intuitive’ answers versus long chain-of-thought explanations—affects public trust and willingness to use these tools.”

“One broader message is that even our views about thinking styles are driven by intuitions. Many people have an intuitive belief that ‘good thinking should look effortful,’ which may make us undervalue genuinely skilled intuition in experts. At the same time, this folk preference for deliberation might be helpful when it nudges us to reward people and systems that are transparent about taking time and effort to reason things through.”

The study, “Humans and LLMs rate deliberation as superior to intuition on complex reasoning tasks,” was authored by Wim De Neys and Matthieu Raoelison.