How AI is reshaping human skills and thinking

In many areas of work and life, generative AI tools offer such great gains in efficiency—and even creativity—that using them is hard to resist. That rapid uptake has sparked a debate: Will relying on AI weaken our cognitive skills—or can it enhance them?

“There’s clear evidence that people can perform many tasks better with AI. The question is: What happens to our skills if we start relying on it regularly?” said Brooke Macnamara, PhD, an associate professor of psychological sciences at Purdue University who studies how we acquire skills and how they can degrade if we don’t keep them sharp.

While some early research suggests certain skills may weaken if we too-readily offload them to AI (Budzyń, K., et al., The Lancet Gastroenterology & Hepatology, Vol. 10, No. 10, 2025opens in new window), other studies suggest the link is more complex. Passive use can lead to skill decay, but more structured, deliberate use may actually boost critical thinking and creativity.

Lessons on AI, cognition, and human skills

The following prompts (organized by topic) can support psychology educators in discussing how AI use affects human cognition, skill development, and behavior.

1. Cognition and memory
The article describes cognitive offloading as using external tools to reduce mental effort and cites research linking GPS use to a weaker sense of direction over time. Using what you know about memory encoding, retrieval, and the role of effortful processing in learning, explain why offloading a cognitive task to AI might impair performance when that tool is unavailable. Under what conditions might offloading enhance rather than hinder cognitive performance?
Reinforces: memory systems, encoding and retrieval, effortful processing versus automatic processing, and the role of practice in skill acquisition

Psychologists and other researchers are now studying how people interact with AI and identifying design and use patterns that strengthen—rather than erode—core human capabilities. That work is starting to show that the benefits of AI largely depend on how human-AI collaboration is structured.

“The key issue is not whether but how AI is used. The difference lies less in the technology itself and more in the user,” said Jackson G. Lu, PhD, an associate professor of work and organization studies at the Massachusetts Institute of Technology (MIT) who studies generative AI use in the workplace.

Deliberate use of AI is particularly important because unlike many technologies that preceded it, AI can do more than just discrete tasks. These systems also give us the chance to offload cognitive tasks central to who we are as humans, such as critical thinking, judgment, and creativity.

“One of the things that’s seductive but also troubling about generative AI is the ability to offload much more executive functioning than we have in the past,” said Nita Farahany, PhD, JD, a professor of law and philosophy at Duke University who studies the legal, ethical, and social questions raised by emerging technologies. “What does it mean for society if humans are passively receiving information but no longer able to critically interrogate it?”

Most experts agree that while today’s AI systems are highly capable of recombining existing knowledge, they cannot generate true breakthroughs. When entirely original problems arise, such as the emergence of a novel coronavirus, human ingenuity must be sharp enough to solve them, Macnamara said. The challenge now is how to use AI to improve efficiency without eroding the human skills that foster competence, autonomy, and innovation.

We offload cognitive tasks all the time—from storing phone contacts to setting calendar reminders—to free up mental resources for use elsewhere. This process, known as cognitive offloading, refers to using external tools or resources to reduce the mental effort required for a task.

When deciding whether to offload, we weigh factors such as our goals for the task and whether a tool might perform it better than we can. Another driver: If cognitive offloading saves us effort, it becomes particularly appealing.

“There’s a general consensus that people are ‘cognitive misers’—if a task can be done with less cognitive labor, we tend to take that option,” said Evan Risko, PhD, a professor of psychology at the University of Waterloo in Ontario, Canada, who studies cognitive offloading, referencing a 1984 theory by social psychologists Susan Fiske, PhD, and Shelley Taylor, PhD.

Research on memory offloading shows it can improve performance when people have access to the stored information, such as making a grocery list and bringing it to the store, said Lauren Richmond, PhD, an associate professor of cognitive science at Stony Brook University of the State University of New York, who also studies cognitive offloading. But if, for some reason, we unexpectedly don’t have access to the stored information (if we leave the grocery list at home, for example), we tend to recall less than if we had relied solely on memory (Grinschgl, S., et al., Quarterly Journal of Experimental Psychology, Vol. 74, No. 9, 2021opens in new window).

Our history with other technologies shows what can happen to skills when we offload them. For example, research has linked GPS use to a weaker sense of direction when people must navigate on their own, with evidence of decline over time (Miola, L., et al., Journal of Environmental Psychology, Vol. 99, 2024opens in new window; Dahmani, L., & Bohbot, V. D., Scientific Reports, Vol. 10, 2020opens in new window).

“We essentially become passive passengers following directions without processing and synthesizing information,” Macnamara said. “But if a technology is very good and almost always available, maybe that’s a skill we’re willing to lose.”

Early research on AI suggests that large language models (LLMs) have the power to alter cognition much like GPS—encouraging passive consumption rather than active engagement—but across a broader set of cognitive skills.

Critical thinking is one example. In a study of 319 knowledge workers led by researchers at Microsoft and Carnegie Mellon University, participants who had more confidence in generative AI also said they engaged in less critical thinking when they used it to complete work tasks. These tasks included developing new ideas, learning about new topics, and getting guidance when making a decision (Lee, H. P., et al., Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems, No. 1121, 2025opens in new window).

“At the moment, most interactions between people and generative AI are transactional. You ask a question; you get an answer; you move on,” said Mutlu Cukurova, PhD, a professor of learning and artificial intelligence at University College London who studies human-AI interaction. “Unfortunately, these types of interactions are more likely to lead to cognitive atrophy than cognitive development.”

In a series of experiments led by Shiri Melumad, PhD, an associate professor of marketing at the Wharton School of the University of Pennsylvania, participants conducted research on a topic, such as how to plant a vegetable garden, then wrote advice for an imagined friend. For their research, writers used either a traditional Google search or an AI tool, such as ChatGPT or Google’s AI overview. Writers using the AI tools spent less time searching and wrote shorter, less detailed summaries. Readers rated the advice from the AI-informed writers as less helpful, less trustworthy, and less likely to be adopted than that of the writers who did their own internet research (PNAS Nexus, Vol. 4, No. 10, 2025opens in new window).

“People learn best when they engage in active exploration, problem-solving, and sensemaking,” Melumad said. “With LLM summaries, this process is completed on our behalf, which transforms learning from an active to a more passive process.”

Using electroencephalography, researchers from MIT examined brain activity during essay writing to further explore these effects. Participants who wrote essays using an LLM had weaker neural connectivity during the task than those who used a search engine or no outside tools. The findings have not yet been peer reviewed (Kosmyna, N., et al., arXiv, 2025opens in new window).

“Research suggests that the use of AI, mediated by cognitive offloading, can lead to a reduction in critical thinking,” said Michael Gerlich, PhD, a professor and head of executive education at the Swiss Business School in Zurich who studies AI and cognitive offloading. “But that’s largely a consequence of how people are using these tools. When used more deliberately, AI can actually enhance critical thinking.”

To test this theory, Gerlich asked 150 participants to write essays on the pros and cons of democracy under three conditions: without AI, using ChatGPT with no guidance, or using ChatGPT with structured prompts designed to encourage deeper thinking. The prompts helped users reflect independently, conduct targeted research, construct and critically review an argument, and revise their work as needed. Essays produced with structured prompting received the highest scores from expert reviewers (Data, Vol. 10, No. 11, 2025opens in new window).

When using an LLM for support, Gerlich recommends flipping the conventional prompting strategy on its head. Rather than providing as much information as possible up front, start by thinking through the problem yourself. When you need additional data, ask pointed questions without revealing details about the broader project. Once you’ve reached an answer on your own, give the AI both the question and your solution and ask for feedback. Gerlich said this approach encourages active engagement and helps mitigate anchoring—where an early answer shapes or biases the thinking that follows—by preventing AI output from steering human reasoning.

In the workplace, research is starting to tease out how, and for whom, strategic use of LLMs could boost cognitive outcomes such as creativity. Lu, of MIT, and his colleagues studied 250 employees of a technology consulting firm in China who were randomly assigned to either receive access to ChatGPT or work without it for one week at a time when ChatGPT was not yet publicly available in China. Separately, employees completed surveys assessing their metacognitive skills, which involve actively monitoring and regulating one’s thinking to complete tasks and achieve goals.

Work done by employees with access to ChatGPT was rated as being more creative by supervisors and external reviewers, particularly when those employees also scored high on metacognitive skills. These employees appeared better able to use AI to generate ideas, switch between perspectives, and retrieve useful information (Journal of Applied Psychology, Vol. 110, No. 12, 2025opens in new window).

“Our findings suggest that employees higher in metacognition are more likely to use generative AI deliberately rather than passively,” said Lu, adding that these nuances can help companies consider when and how humans and AI may collaborate to optimize results.

Research is also exploring whether using AI affects specialized job skills, including experts’ ability to analyze medical images, legal documents, and financial transactions once AI assistance is taken away.

In Poland, researchers in a natural experiment—a real-world study that observes changes already occurring—measured how well physicians could detect polyps on colonoscopies without AI assistance both before and after AI tools were added to their workflow. The polyp-detection rate without AI assistance fell 6 percentage points, from 28.4% to 22.4%, in the 3 months after the AI system was introduced (Budzyń, K., et al., The Lancet Gastroenterology & Hepatology, Vol. 10, No. 10, 2025opens in new window).

With funding from the National Science Foundation, Macnamara is further investigating skill decay in medical settings. Focusing on AI-assisted radiology and laparoscopic (minimally invasive) surgery, she is exploring three questions: When a skill has already been learned, does using an AI assistant lead to skill decay? If clinicians train with AI, do they learn the skill as deeply as those who train without it? And finally, does using AI make clinicians overconfident in their abilities?

These studies point to a new challenge for organizations—not just whether to adopt AI but how to use it strategically alongside human strengths, Lu said. Part of the answer involves deciding which tasks to offload. Cukurova suggests performing a “task analysis” for each employee, separating tasks that simply need to be completed (such as drafting routine legal memos) from those that involve essential learning (such as analyzing case law to build a legal argument). Psychologists and learning scientists can help guide these decisions by identifying the cognitive processes involved.

An employee’s prior knowledge and experience also matter. Experts are generally better equipped to critically evaluate AI outputs, whereas novices may struggle to do so. For example, an experienced lawyer using AI to find novel outputs can better spot flawed reasoning than a first-year law student can, Farahany said. Like an intern, AI can support work, but it cannot replace the judgment needed to catch mistakes, edit weak ideas, or recognize when something has gone wrong. In practice, however, organizations may not always know which employees have enough experience to use AI independently and which still need to build foundational skills.

“The problem is that it’s not always easy for employers to distinguish between novice workers and experienced ones,” Cukurova said.

Without clear guidance, employees with high workloads and those who fear losing their jobs may also turn to AI for support or to gain a competitive advantage, said Mindy Shoss, PhD, a professor of psychology at the University of Central Florida who studies AI use through an industrial and organizational lens. Organizations tend to be most successful when they set clear norms and guardrails for AI use, she said.

“When incentives are vague, you can end up with a chaotic situation where people are sending AI-generated content back and forth, but no one is quite sure how it’s being used or evaluated,” Shoss said, describing what some have called “AI slop.”

Employees across many workplaces are experimenting with AI in real time. At Microsoft, David Evans, PhD, a social psychologist who conducts market research, regularly uses LLMs for literature reviews and to help publicize and disseminate findings. He says AI is leading to upskilling just as often as de-skilling—he replaces a skill he uses less with another one he wants to develop.

“It comes down to how we use the extra time that AI gives us,” Evans said. “We might do more of the same, we might do less, but we might also follow our curiosity to learn and upskill in ways that weren’t possible before.”

AI can support upskilling, but only when it is used deliberately, Lu said. Rather than passively offloading, workers can use AI to streamline tasks such as searching for, organizing, and digesting information. That frees up cognitive resources, which can then be redirected toward higher-order thinking.

Organizations also help determine whether AI supports upskilling or simply accelerates output, Shoss said. Workers are more likely to embrace AI as a tool for growth when employers provide opportunities for learning and development, which workers often view as a sign of organizational support (Rothwell, J., “The American Upskilling Study Shows Workers Want Skills Training,” Amazon News, Sept. 9, 2021opens in new window). They are also more likely to build new skills when given the time, psychological safety, and incentives to experiment with AI in thoughtful ways.

“AI does not inherently lead people to build new skills,” Shoss said. “If people are overloaded or feel pressure to simply produce more, AI will be used to speed up existing work. Organizations therefore need to create time and incentives for employees to explore, learn, and build on their existing strengths.”

Still, Farahany said the choice whether to de-skill or upskill ultimately lies with the individual. She chooses to offload certain tasks—such as using AI to sort and prioritize emails—but devotes the extra time to learning new skills, such as taking a sewing class with her daughter.

“What happens when we start to go into passive mode of receiving rather than generating is that our overall competence starts to erode,” Farahany said.

As individuals, we bear responsibility for keeping our cognitive skills sharp. Users should be mindful of which tasks they offload and how much of their learning becomes passive, Farahany said. But it’s also important for leaders to consider how current incentive structures shape choices around AI use.

“In most organizations, there’s a powerful incentive for employees to produce output as quickly as possible,” Farahany said. “You may want to foster your critical thinking skills, but you’re penalized for doing so.”

If the focus is on efficiency rather than quality, AI can produce results that are fast and acceptable—but mediocre, Gerlich said. Over time, that is likely to erode both company culture and the bottom line.

On a broader scale, the stakes extend beyond individual workers and organizations. If people continue to outsource core cognitive tasks to AI, experts worry it could gradually weaken the human capacity for creativity and innovation.

“Imagine you have a candle. If you tell AI to improve it, you’ll get a candle that shines brighter and longer, costs less, and probably has the best design,” Gerlich said. “But AI will not make the leap from candle to light bulb. That is out-of-the-box thinking—a type of creative thinking that AI, which is based on pattern recognition, cannot do.”

Maintaining those human capabilities may ultimately depend less on the technology itself than on the structures and incentives shaping how it is used.

“The only way we preserve human competence over time is by rethinking our system of incentives so that they not only support short-term output but longer-term flourishing as well,” Farahany said.