What is cognitive surrender? It is the term two Wharton researchers use for what happens when we take an answer from artificial intelligence and simply adopt it, without checking whether it makes sense. The AI’s output replaces both our first instinct and our own slower reasoning, so the answer we give is no longer really ours.
Most of us have done a version of this. You ask a chatbot a quick question, the reply sounds fluent and confident, and you move on. That works well when the answer is right. The problem is what happens when it is wrong, because the same habit that saves time also stops us noticing the error.
The idea comes from a 2026 working paper by Steven Shaw and Gideon Nave at the Wharton School of the University of Pennsylvania. It is a preprint, which means it has not yet been through peer review, so its findings should be treated as promising rather than settled.
From two systems to three
Much of modern decision science rests on the dual-process view of thinking, made famous by Daniel Kahneman in his 2011 book Thinking, Fast and Slow. System 1 is fast, automatic, and intuitive. System 2 is slower, effortful, and deliberate. Many everyday mistakes happen when a quick intuition goes unchecked by careful thought.
Shaw and Nave argue that this picture now needs a third part. In their Tri-System Theory of human reasoning, System 3 is artificial cognition that operates outside the brain, such as a generative AI assistant. It can support our own thinking, but it can also take its place entirely. Cognitive surrender is what they call that second route: adopting AI outputs with minimal scrutiny and overriding both intuition and deliberation.
This is different from ordinary cognitive offloading. Psychologists Evan Risko and Sam Gilbert describe cognitive offloading in Trends in Cognitive Sciences as using physical action, such as writing a note or setting a reminder, to reduce the mental demands of a task. Offloading a shopping list to your phone is sensible. Handing over the judgement itself is the step that worries Shaw and Nave.
What the experiments found
The researchers ran three preregistered experiments using an adapted version of the Cognitive Reflection Test, a set of puzzles with an intuitive but wrong answer and a less obvious correct one. Across 1,372 participants and 9,593 trials, people could choose whether to consult an AI assistant. Without telling participants, the researchers secretly controlled whether the AI gave accurate or faulty answers.
- People used the AI a lot: Participants chose to consult it on more than half of the trials.
- Accuracy followed the AI: In the first study, compared with people who had no AI access, accuracy rose by 25 percentage points when the AI was right and fell by 15 percentage points when it was wrong.
- Confidence went up regardless: Using the AI made people more confident in their answers, even after it had led them into errors.
- Pressure did not fix it: Time pressure in the second study and payment and feedback for each correct answer in the third changed overall performance, but faulty AI still pulled accuracy down.
The combination of lower accuracy and higher confidence is the most striking part. A wrong answer that you feel unsure about invites a second look. A wrong answer that you feel sure about does not, and that is how a small error can travel a long way.
Who is most prone to cognitive surrender
Not everyone surrendered to the same degree. Across the studies, the people who showed the strongest pattern had higher trust in AI, lower need for cognition (a measure of how much someone enjoys effortful thinking), and lower fluid intelligence, the ability to reason through new problems.
That does not mean cognitive surrender is only a problem for some people. Trust in AI is easy to build because chatbots are often right and always sound fluent, and anyone who is tired, rushed, or out of their depth leans more heavily on whatever help is nearest. Psychreg has looked at a related bias, the Gell-Mann amnesia effect and why we trust AI when we cannot check it, which helps explain why confident answers in unfamiliar areas are so persuasive.
How to keep your own judgement switched on
The study does not say that using AI is a mistake. When the AI was accurate, it helped people, including under time pressure. The aim is to use System 3 to support your thinking rather than replace it.
- Answer first, then ask: Write down your own answer or best guess before you open the chatbot, so you have something to compare against.
- Ask how it got there: Request the reasoning or sources, and check whether the steps actually lead to the conclusion.
- Notice the confidence boost: If you feel more certain simply because the AI agreed with you, treat that as a cue to double-check.
- Match checking to stakes: A dinner recipe needs little scrutiny, but health, money, legal, or work decisions deserve a second source.
- Keep practising the skill: Regularly doing some tasks without AI keeps your own reasoning in shape, a concern also raised in research on whether using AI makes students think less.
What this means for the rest of us
The authors note that their experiments took place in controlled settings, so real life may look different. Even so, the core finding is easy to recognise: when a confident answer is one click away, it is tempting to stop thinking. Cognitive surrender is not laziness or stupidity. It is a predictable response to a tool that is usually right, and the best protection is a simple habit of checking before you accept.
FAQs
- What is cognitive surrender in simple terms? It means accepting an AI’s answer with little or no checking, so the AI’s output replaces both your instinct and your own reasoning.
- Is cognitive surrender the same as cognitive offloading? Not quite. Offloading uses tools to reduce mental effort, such as writing a list. Surrender hands over the judgement itself.
- Has the research been peer reviewed? Not yet. The Shaw and Nave paper is a preprint posted on SSRN in 2026, so its findings may change.
- Does this mean I should stop using AI? No. The same study found AI improved accuracy when it was right. The key is to form your own view first and check answers that matter.
