Does Using AI Lead to ‘Cognitive Surrender’?

In early September, New York City told nearly 600,000 public school students – every child from 2-K through eighth grade – that they may not use generative AI for the coming school year. Announcing the moratorium, Mayor Zohran Mamdani argued that children need to build relationships with their teachers and work through hard problems on their own. Few would quarrel with that goal. But behind the policy sits a fear that has spread through classrooms, boardrooms and newsrooms this year: that when people use AI, they stop thinking.

That fear now has a name. In January, Wharton researchers Steven Shaw and Gideon Nave introduced the term “cognitive surrender” to describe people adopting an AI’s answer as their own with little scrutiny. Their experiments, involving 1,372 participants, produced a headline-ready finding: when a chatbot confidently gave a wrong answer, people accepted it about 80% of the time. They also felt more confident than those who had reasoned without AI.

The research is careful, and its data deserve to be taken seriously. The conclusion many people have drawn from it, however, does not. “Surrender” implies something inevitable, even built into the technology. Read the findings closely and they say something quite different.

Consider the other half of the results. When the AI was right, participants’ accuracy rose 25 percentage points above that of people working alone. When it was wrong, accuracy fell 15 points below. The same tool, in other words, produced large gains or large losses depending entirely on whether its output was checked. The researchers also found that surrender was most pronounced among people with high trust in AI and a low appetite for effortful thinking.

Just as important is that participants were not given any guidance on how to use the tool. They simply had access to an optional AI chatbot, which may have led many to treat it like a Google search: ask a question, get an answer, accept it. Tellingly, when the researchers added a modest nudge in a later experiment, rewarding accuracy and giving feedback after each answer, participants overrode faulty AI advice more often and their accuracy improved. Surrender did not disappear, but it shrank.

That is not a portrait of a technology that erodes the mind. It is a portrait of a tool being used badly, as an oracle to be consulted once and obeyed. Hand someone the keys to a car without a single driving lesson and you will get accidents. No one argues from this that cars cause “navigational surrender.”

Nave has framed the real question well: not whether we will rely on AI, but how much of our thinking we are willing to give away. We agree. Our quarrel is with the word “surrender,” which answers that question before anyone has had the chance to learn a better way. Our book, Creativity in the Age of AI: Toolkits for the Modern Mind, was written to teach that better way. It rests on the simple conviction that everyone can enhance their creativity.

The book’s goal is to help any individual or organization do so through a personalized portfolio of 12 sets of approaches, from challenging your mental models and drawing analogies from unrelated fields to experimentation and iteration. Each of these approaches can be turbocharged by AI, but only when AI is used correctly: not as a search engine but as a smart research assistant. The first means typing a question and accepting whatever comes back. The second means interrogating the answer, asking why, requesting alternatives, challenging assumptions, testing whether a conclusion holds under different conditions.

Used this way, AI does not replace judgment; it sharpens it. The result is not cognitive surrender but cognitive enhancement. Every reply becomes a prompt for the human to think harder, not less. One of us (Mukul) has been using AI to write and compose songs, going through more than 40 versions before arriving at the final form. Similarly, when he used AI to help write a play, each act went through multiple revisions and more than a year’s worth of effort before it accurately reflected his mental vision.

Every creative act involves both inspiration and perspiration. The closer a task sits to the inspiration end of the spectrum, the more firmly it belongs in human hands. The closer it sits to the perspiration end, the more it can be handed to AI, which often does it faster and better. This division of labor lets anyone make the best use of both their own mind and AI. Musicians working with AI song tools routinely generate dozens, sometimes more than 100, versions of a track before choosing one. That is not surrender; it is curation. And curation is a deeply creative act.

The process is the point

The same error of judging the output while ignoring the process shows up in the fast-growing market for AI detectors. These tools scan a finished piece of writing and pronounce it human or machine. They are prone to false positives; sometimes articles written years before generative AI existed have been flagged as machine-made. More fundamentally, they rest on the assumption that any AI involvement is a form of cheating.

But a writer who drafts an essay, asks AI to edit it, argues with the suggestions, rewrites, and repeats the cycle a dozen times has not outsourced her thinking. She has extended it. A detector cannot see those painstaking steps. The same muddled thinking afflicts much of the public debate about cognitive surrender. What matters is not whether AI touched the work, but whether a human was actively directing, questioning and deciding throughout.

This is why New York’s moratorium, however well-intentioned, points in the wrong direction. If the Wharton study shows anything, it is that the gap between good and bad outcomes is a gap in skill. And skills are learned. Keeping 600,000 children away from AI at school will not keep them away from it at home or in the workplaces they will one day enter. It will simply ensure that they meet it without guidance or preparation.

The city deserves credit for planning AI literacy courses for high school students and for convening a coalition to study the moratorium’s effects. We would urge that coalition to go further: run proper experiments, with control groups, comparing students kept away from AI with students taught to use it critically, and measure not only test scores but creativity, persistence and engagement. Our bet is that the second group will think more, not less.

The history of technology is littered with warnings that new tools would make us dumber. Writing, Socrates feared, would destroy memory. Calculators were supposed to end numeracy. Cameras, it was believed, would make painting obsolete. Each time, the tools rewarded those who learned to use them well and penalized those who did not.

AI is more powerful than any of its predecessors, which makes the stakes higher but doesn’t change the conclusion. The choice before educators, managers and policymakers is not between surrender and abstinence. It is between leaving people to stumble into passive dependence and teaching them to become active partners with a remarkable new collaborator.

Cognitive surrender is real, but only as a description of what untrained users do. It is not destiny. The remedy is not to fear the tool, but to master it.

Editor’s note: Deborah Yao is a co-author of “Creativity in the Age of AI: Toolkits for the Modern Mind.“