MIT Media Lab research reveals that excessive use of ChatGPT leads to cognitive debt accumulation - How will we protect our thinking?

Introduction – The Shadow of Convenience, The Stifling of Thought

Generative AI is changing the way we speak, write, and think even at this moment. In particular, language models like ChatGPT seem to surpass humans in the ability to structure information and refine sentences. For those who struggle with writing, it is a liberation, and for industries that prioritize information processing speed, it is an innovation. However, convenience does not always guarantee benefits. A recent study published by MIT Media Lab experimentally shows how dependence on AI can undermine human cognitive abilities in the long term. This essay presents practical directions to maintain AI's usefulness while keeping control over our thinking.

Warning from MIT Media Lab Research - Automation of Thought and Cognitive Debt

In 2025, the MIT Media Lab conducted a four-month essay writing experiment involving 54 college students and young adults. Participants were divided into three groups: a ChatGPT usage group, a Google search usage group, and a "brain only" group composing solely using their minds without tools. During the essay writing, participants wore EEG devices to analyze brain activity and connectivity in real-time, while the creativity, originality, and self-consistency of the written pieces were assessed by both human judges and AI. The results showed that the brain activity of the ChatGPT usage group was the lowest among the three groups. These participants exhibited lower activation in the frontal lobe, responsible for judgment, memory, and creativity, leading to low scores in memory, critical thinking, and creativity. Some participants even failed to remember sentences they wrote or did not recognize the sentences as their own. In contrast, the "brain only" group demonstrated the strongest brain connectivity and produced creative and logical writing through spontaneous thought processes. Particularly interesting was that when these individuals later used ChatGPT, their brain activation actually increased. This suggests that users who trained their thinking skills before using tools could utilize AI tools productively. The MIT researchers explained this concept as "Cognitive Debt." When AI is used in a manner that replaces thinking rather than assisting it, humans gradually lose their abilities to judge, remember, and question.

The structure in which generative AI paralyzes thinking

The way generative AI hinders thinking is simple. It 'provides' sentences, and users only need to 'select.' The problem arises right here. Language is a tool for thought and thought itself. The process of writing is essentially the process of organizing and systematizing thoughts. However, when a complete sentence is provided, humans do not internalize the logic, reasoning, or context contained within it. Users do not 'understand' the sentence they merely 'use' it. This is a significant omission in thinking. As it happens repeatedly, the brain 'learns' this. Initially, AI was a 'support tool,' but before long, it takes the initiative in thinking. Users stop questioning, stop verifying, and no longer create unique expressions. They accept only the results and abandon the process. At this point, the brain's executive functions, creative circuits, and memory circuits gradually deteriorate. Ultimately, the habit of outsourcing thinking is precisely what cognitive debt is.

Four Practical Strategies to Reduce AI Dependence and Strengthen Critical Thinking

AI is not a subject of prohibition. Rather, its usefulness is undeniable. The important aspect is the 'method' of use. Below are strategies for utilizing AI that reduce dependence and enhance critical thinking. 1. Limit AI to being a 'draft generator.' 2. Reinterpret instead of 'copying' sentences. 3. Train to identify AI's errors. 4. Consciously set aside time for 'thinking without AI.' The sentences suggested by AI are not final products. Rather, they should be used as a starting point to organize your thoughts or experiment with various perspectives. It is advisable to first construct the core argument, logical flow, and references of your writing without AI, and then request a draft from AI based on that. Sentences generated by AI must be translated into your own language, and the underlying premises and logic must be examined. For example, questions like, "Is this argument compelling?" and "Can I support what this sentence is saying?" should be asked. AI-generated text is just material, not the completion of thought. Actively searching for AI's errors, distortions, and biases is an excellent way to train critical thinking. Since generative AI often presents plausible fictions, the distinction between fact and opinion blurs if users do not verify it. It is necessary to develop the habit of treating AI as an object of criticism rather than a 'watcher.' Before seeking AI's assistance, let us secure at least 5 minutes to think about the topic solely in our minds. It is important to cultivate the habit of first recalling and structuring the composition of the text, core questions, examples, etc. AI is a tool to organize these thoughts, not a substitute for thinking.

Conclusion

The biggest message from the MIT research is this: 'When AI is used as a substitute rather than a means, human thinking ceases.' Generative AI has excellent language abilities, but it does not doubt, question, or remember. Thinking is solely a privilege and responsibility of humans. We cannot entrust our thinking to AI or transfer that responsibility. Therefore, the real challenge in the era of generative AI is not 'how do we use it better,' but 'how do we maintain control over our thinking?' Using AI effectively means being able to think without relying on it. Truly smart humans use AI to ask deeper questions, critique more sharply, and think more creatively. In such cases, AI becomes a tool, and humans remain the subjects of thought.