May 2026 | Innovation

Thinking Fast, Slow, Artificially: AI and Your Brain

Daniel Kahneman's idea of fast and slow thinking — the distinction between the quick, intuitive judgments people make automatically and the careful, deliberate reasoning engaged when a problem demands real effort — has shaped how economists, managers, and executives understand judgment and choice for decades. In his landmark book and the behavioral science it inspired, Kahneman labeled these System 1 and System 2. But new research from Wharton postdoctoral researcher Steven Shaw and professor Gideon Nave argues that his framework is missing something fundamental today: it was built for a world in which all thinking happens inside the human mind. That world no longer exists.

Shaw and Nave address the gap with a new Tri-System Theory, proposing that engaging with AI has created a third mode of human cognition that influences how the two other modes operate. System 3, they argue, is as influential in today’s decision making as intuition and deliberation. In a paper titled “Thinking — Fast, Slow, and Artificial: How AI is Reshaping Human Reasoning and the Rise of Cognitive Surrender ,” they present three experiments involving more than 1,300 participants and nearly 10,000 individual trials. Their findings carry significant implications for anyone managing people, leading organizations, or making high-stakes decisions in the age of AI.

The Outsourcing Option

People have always sought outside input before making decisions. Deferring to expertise, such as from a doctor, a lawyer, or a financial advisor, is as old as human society, and not necessarily unwise. But as Shaw notes, “An expert is only an expert in one domain, and they’re not going to be available to you all of the time on any topic.” And because experts are human, their advice carries the recognizable marks of a human mind: uncertainty, qualification, and the occasional admission of not having the answer. Those signals prompt the person receiving the advice to evaluate it, question it, and weigh it against their own judgment.

AI changes that entirely. As Shaw describes it, AI is “an expert across all sorts of different domains, and we have access anytime we want it. It speaks with confidence, carries few markers of doubt, and rarely says a subject is outside its area. It is not the advice of a fallible human mind. It is something new.”

And that novelty creates an option that has not existed before: the option to defer thinking entirely, on any question, at any moment. Shaw and Nave call the resulting phenomenon cognitive surrender: the uncritical acceptance of AI’s answers in place of one's own reasoning.

The Signature of Cognitive Surrender

To test when and how cognitive surrender happens, Shaw and Nave created an experiment. Participants in two groups solved reasoning problems from the proverbial Cognitive Reflection Test, which was designed to reveal when people override initial intuition and think more carefully. One group solved the problems on their own. The other had access to an AI chatbot they could consult as often as they liked. But what those in the second group didn’t know was that the chatbot had been programmed to give either the correct answer or a confident-sounding wrong answer.

The results were unambiguous. When participants consulted the AI and it was correct, their accuracy jumped 25 percentage points above baseline. When the AI was wrong, accuracy fell 15 percentage points below the baseline of participants who had no AI access. In other words, the danger wasn't AI. It was deference to AI.

Perhaps more unsettling was that Shaw and Nave found that even when the AI was giving wrong answers roughly half the time, participants’ confidence in their responses went up. Access to AI inflated certainty across the board, regardless of whether that certainty was warranted. The internal signals that would normally prompt deeper deliberation, such as a sense that something doesn’t add up, appear to be suppressed when System 3 is in the picture.

Nave says this is what distinguishes cognitive surrender from the more familiar concept of cognitive offloading. “Using a calculator offloads a specific math task while human reasoning stays in charge. Cognitive surrender is different: it is the moment when AI is not just doing a specialized task but making the decision, and the person adopts that decision as their own without recognizing the transfer has occurred.” Shaw adds, “The locus of control shifts from the person to System 3. That distinction didn’t have a name before this paper.”

Who Is Most at Risk — and What Helps

Not everyone surrenders equally. Shaw and Nave measured several individual differences across their studies, and the patterns are consistent. People who trust AI more are substantially more likely to follow incorrect AI advice and less likely to question it. People who are more analytically inclined, who enjoy thinking problems through and have stronger reasoning ability, are better protected. They are more likely to notice when something is off and push back on a faulty answer.

The implications for organizations are significant. The risk is greatest when the people most eager to use AI are also the least likely to question what it tells them. But there is also a more optimistic finding in the data. When participants were given incentives for accuracy and immediate feedback on each answer, they were far more likely to question and override incorrect AI advice.

“A good AI system is one that helps you when it’s right and doesn't harm you when it's wrong,” Nave says. “Right now, if the AI is wrong, you can end up worse than you were without it. The goal is to find ways to shift people to a place where they benefit from AI when it’s correct and don't lose when it’s wrong.”

That distinction matters because not all cognitive surrender is harmful. In structured, well-defined tasks where AI is simply more accurate than human judgment, deferring to it may be entirely rational. The challenge, for both individuals and organizations, is knowing when deferring to AI is the right call and when the decision requires human judgment.

What It Means for Organizations

The efficiency gains of AI are real, and in many contexts, the case for using it is straightforward. But Tri-System Theory puts a sharper question on the table for leaders: not whether to use AI, but whether the way it is being used is quietly eroding the human capabilities that organizations depend on.

The risk is most acute in domains where judgment, accountability, and critical thinking are not incidental but essential, such as in health care, law, education, and management consulting. Workers who routinely defer to AI without questioning it may find their capacity for independent reasoning gradually eroding through disuse. And unlike a drop in productivity, that erosion is invisible until the moment it matters most: the client meeting where there is no AI to consult, the diagnosis that requires judgment no algorithm can provide, the strategic decision that demands genuine human insight.

The research suggests two broad approaches to addressing this. AI interfaces can be built with features that encourage users to pause and verify before accepting an answer, signals that flag when the AI is uncertain, or prompts that ask users to consider the response before acting on it. And organizations can invest in training people to recognize when they are surrendering and when they should be pushing back.

Tri-System Theory is not a warning against AI. It is a call for organizations to be intentional about how they use it: to understand which decisions benefit from AI's capabilities and which require human judgment that no algorithm can replicate, and to build the conditions, incentives, and habits that preserve the habit of independent thought. In an age of cognitive surrender, that may be the most important skill of all: knowing when to think fast, when to think slow, and when to think for yourself.

---

Share This

Subscribe to the Wharton@Work RSS Feed