The Expert Who Stopped Thinking – Is Cognitive Surrender Corrupting Your Work?

You’re still the expert on paper. But who’s really producing your output?

Something Has Changed in How You Work

Be honest with yourself for a moment. When a technical question arises in your day—the kind you once spent an hour researching, cross-checking, and piecing together—what do you do now?

If the answer is “I ask the AI,” you’re not alone. And in most cases, that’s fine. AI has genuinely transformed professional productivity, compressing hours of research into minutes and surfacing knowledge that would have taken days to assemble.

But recent research from the Wharton School at the University of Pennsylvania suggests something more concerning is happening beneath the surface of this productivity gain—and it threatens the very thing that defines a professional: the quality and integrity of their own judgment.

The phenomenon is called cognitive surrender, and if you use AI regularly in your professional work, you are almost certainly experiencing it without knowing.

What Is Cognitive Surrender?

Psychologists have long described human reasoning as a partnership between two systems: fast, intuitive thinking (System 1) and slow, deliberate analysis (System 2). Wharton researchers Steven D. Shaw and Gideon Nave extended this model with a third system—System 3: artificial cognition—to describe how AI has entered the reasoning process itself.

Within this new “triadic” model, they identified two very different ways of working with AI:

• Cognitive offloading is what most professionals believe they’re doing: strategically delegating a discrete task to a tool (like using a calculator), while keeping your own reasoning firmly in charge of the final judgment.

• Cognitive surrender is what the research shows people actually do: adopting AI outputs with minimal scrutiny, bypassing both intuition and deliberation entirely. You’re not using a tool—you’re handing over the decision and adopting the machine’s answer as your own.

The distinction is subtle but profound. Offloading preserves your expertise. Surrender quietly replaces it. And crucially, the researchers found that people stop deliberating altogether when they surrender—they often can’t tell you whether or how thoroughly they evaluated the AI’s output, because they didn’t.

How Big Is the Problem?

In three preregistered experiments covering more than 1,300 participants and nearly 10,000 reasoning trials, the researchers deliberately manipulated AI accuracy—sometimes the AI was right, sometimes confidently wrong.

The results should give any professional pause:

• Most participants chose to consult the AI, and when they did, they followed its advice roughly four out of five times even when it was wrong.
• Participants’ accuracy tracked the AI’s accuracy almost perfectly—rising well above their unaided baseline when the AI was correct, and falling well below it when the AI was faulty.
• Perhaps most alarming: confidence went up, even when the AI was wrong. Access to AI made people more sure of themselves while making them measurably less accurate.

And this pattern proved stubborn. Time pressure made it worse, and even paying people for accuracy and giving them immediate feedback on their errors only partially reduced it. The gap between following good AI advice and bad AI advice remained large in every condition tested.

The researchers also profiled who is most vulnerable. Higher trust in AI, lower enjoyment of effortful thinking, and lower fluid intelligence all predicted greater surrender. If you’d describe yourself as someone who “trusts the tools” and finds AI faster than thinking it through yourself—that’s you.

(Full methodology and statistics are available in the original paper: SSRN #6097646)

The Alarm: Industry by Industry

This is where the research stops being academic. The defining feature of professional work is that you are the expert—clients, employers, regulators, and patients rely on outputs that are fundamentally yours, carrying your license, your reputation, and your liability. Cognitive surrender undermines exactly that.

Medicine. The study cites recent multicentre research showing that physicians repeatedly deferring to AI recommendations in colonoscopy showed erosion of their own unaided diagnostic skill—”deskilling.” A clinician who surrenders to a diagnostic AI isn’t just at risk of adopting a single error; they are gradually hollowing out the clinical judgment that took a career to build. Patients don’t sue an algorithm; they sue the doctor who signed off on it.

Law. Contract review, legal research, and drafting are being transformed by AI. But an attorney who adopts an AI-generated analysis without independent verification is staking their license on a system with no duty of care, no accountability, and—as this research shows—a strong tendency to be believed regardless of accuracy. Courts and clients expect your judgment, not a machine’s.

Finance and Accounting. Analysts, auditors, and advisors working under time pressure face the highest-risk combination this research identifies: deadlines suppress deliberation, making surrender the path of least resistance. A confident, plausible AI-generated forecast or reconciliation adopted without scrutiny is indistinguishable—in the moment—from genuine analysis. Until it isn’t.

Engineering and Technical Fields. Design calculations, code reviews, and specification checks are increasingly AI-assisted. The professional seal on a drawing or sign-off on a report certifies human verification. Where that verification becomes ceremonial—rubber-stamping AI output—the entire chain of professional accountability breaks.

Journalism, Consulting, and Advisory Roles. Any profession paid for synthesis and judgment faces the same trap: AI produces drafts and conclusions faster than you can evaluate them, and the research shows the confidence of its delivery suppresses your own critical scrutiny. The “expert report” that reaches your client may be largely machine-authored, with your reasoning reduced to endorsement.

Education and Therapy. The paper references emerging concerns about conversational AI amplifying delusional beliefs in vulnerable users—”AI psychosis”—and educators face parallel risks of surrendering pedagogical judgment to automated tools.

The common thread across every industry: cognitive surrender transfers the substance of your work to a system that cannot bear responsibility for it, while leaving your name on the output. You retain the liability; you lose the judgment.

What You Can Do: Practical Defences

The research itself points the way. Certain conditions measurably reduced surrender—and they can be built into your daily practice:

1. Rebuild the verification habit. The study found that people who overrode faulty AI and answered correctly were engaged in genuine offloading—they used AI as input, not as an oracle. Make it a rule: no AI-generated fact, figure, citation, or conclusion enters your work without an independent check against a primary source. Not because AI is usually wrong, but because your scrutiny is what keeps you the expert.

2. Use incentives and feedback. In the study, paying for accuracy and giving immediate error feedback was the most effective intervention—more than doubling override rates on faulty AI. Translate this personally: track your error rates when using AI versus not, and attach real stakes to catching mistakes. Calibration comes from visible consequences.

3. Beware of deadlines and time pressure. Time pressure was the strongest amplifier of surrender. High-stakes work done under deadline is precisely where you should be most deliberate about verification, not least. Schedule the checking time before the deadline crunch arrives.

4. Monitor your confidence. The most insidious finding is that AI access inflates confidence even when you’re wrong. Treat rising confidence as a signal to check, not to relax. If a conclusion feels finished the moment the AI finishes talking—especially without friction or contradiction—be suspicious.

5. Audit yourself periodically. Do a piece of significant work “Brain-Only,” as the study’s control condition did—no AI at all. Then compare. This is how you detect the deskilling drift before it becomes dangerous, and how you retain the ability to recognise when the AI is wrong.

6. Push for organizational safeguards. Demand that your workplace define which decisions require documented human verification, and insist AI outputs carry uncertainty indicators rather than bare assertions. The researchers suggest that transparent confidence signalling, domain-specific caution flags, and interfaces that encourage calibrated collaboration can preserve critical thinking without sacrificing AI’s benefits.

The Bottom Line

Cognitive surrender is not a hypothetical future risk—it is a documented, present behaviour, and the study’s numbers suggest it is the norm rather than the exception. The professionals most at risk are not careless ones; they are competent, busy, trusting adopters of genuinely useful tools, who have no idea their reasoning has quietly moved outside their own heads.

“We do not merely use AI; we think with it,” the researchers write. The question your clients, patients, and employers are implicitly asking is whether the thinking they’re paying for is still happening in yours.

Check the research yourself—verify the claim, evaluate the evidence, and reach your own conclusion. It would be a fitting place to start.

Reference: Shaw, S. D., & Nave, G. “Thinking—Fast, Slow, and Artificial: How AI is Reshaping Human Reasoning and the Rise of Cognitive Surrender.” The Wharton School, University of Pennsylvania. Available at SSRN #6097646.