AI Speeds Up Work, but the Brain May Shrink: De-Skilling Looms Over Chatbot Users

The convenience offered by chatbots such as ChatGPT and Claude carries a rarely noticed side effect: users' ability to think, write, and remember is slowly eroded. In the economics literature this phenomenon is called de-skilling, and a number of researchers now warn that the instant productivity of AI may go hand in hand with a long-term decline in cognitive capacity.

Susette Brooks, a marketing director in Philadelphia who holds an MFA in nonfiction writing, says she felt the impact firsthand. At first she used AI to draft emails so the work would be lighter. But not long after, she realized her writing and reasoning abilities had weakened too. "I felt my writing and thinking skills decline right then," she said, noting difficulty remembering information and stringing ideas together.

The mechanism behind these symptoms is known as cognitive offloading—delegating the burden of thinking to an external tool. We have long used calculators and calendars for similar purposes. What sets this apart is that ChatGPT and Claude now take over far more complex tasks: writing papers, building presentations, even writing program code. The scale of cognitive delegation like this is unprecedented.

Daniel Willingham, a professor of psychology at the University of Virginia, explains that the human brain is in fact designed to avoid the effort of thinking. "Your mind is basically built to save you from having to think," he said. "Thinking is what you do when you're in trouble." Because of that, even though many people feel AI is harmful, the temptation to take the fastest route remains hard to resist.

The impact is starting to show in classrooms. Longitudinal research by the University of California Irvine found students spent less time and scored better on text-based questions after ChatGPT appeared. But when tested offline without AI access, their performance actually dropped. A similar pattern emerged in an experiment at Middlebury College: students who used AI to deepen their understanding (augmentation) performed better than those who let AI do everything (automation).

"People may find it hard to notice their skills are declining because their performance with AI stays high. They don't realize how low their independent ability is when the tool isn't available." — Brooke Macnamara, professor of psychology at Purdue University

Interestingly, AI is most useful when applied to fields one has already mastered. A working paper by MIT economist David Autor and colleagues—some of whom work at Google, which funded the study—showed 133 patent lawyers saw performance gains when using AI. But that advantage disappeared for junior lawyers once the tool was withdrawn, while senior lawyers remained better than peers who had never used AI.

Anjali Singh, a researcher on human-AI interaction at the University of Texas at Austin, highlights the loss of moments for reflection. "We no longer get the chance to ponder," she said. The old way of accessing information—reading a book or searching Google—took place more slowly and gave the brain room to process material. Now, answers arrive in a sea of text that is instead confusing.

For the Indonesian context, similar concerns are relevant. Adoption of generative AI among office workers, students, and MSME players in the country is growing rapidly, while frameworks for digital literacy and academic ethics are not yet fully established. Without disciplined usage guidelines, the risk of de-skilling could be greater, especially for young people still building basic competencies in school and on campus.

Experts suggest several practical steps. First, ask yourself whether a task really deserves to be handed to AI. Second, make use of special modes such as Study Mode in ChatGPT, Claude for Education, or Gemini for Students that act as a tutor, not an instant answer provider. Third, turn the interaction into a collaboration: ask for five draft options, compare them yourself, then decide. "Just comparing alone already stimulates the critical thinking process," said Advait Sarkar, an AI researcher at the University of Cambridge.

The most honest test is to occasionally set AI aside and rely on your own brain—writing a day's emails without help, looking up facts through a conventional search engine, or writing code manually. If the results are disappointing, that is a signal the skill needs to be retrained. The question now is no longer whether AI will keep being used, but how consciously we choose which tasks to delegate—and how willing we are to hold back so the brain keeps working.