# Does Relying on AI Make You Dumber? MIT's Cognitive Debt Study Through an IQ Lens Source: https://iqcat.vercel.app/en/blog/ai-cognitive-debt Published: 2026-09-04 Publisher: NekoIQ (https://iqcat.vercel.app) Tags: 認知能力, 心理学 > In an MIT Media Lab study, people who wrote essays with ChatGPT showed the weakest brain-network connectivity and often could not quote their own text. What cognitive debt means, the limits of a preprint, how it relates to IQ and fluid intelligence, and how to use AI wisely. ## Key takeaways - In a 2025 MIT Media Lab preprint, 54 people wrote essays in three groups, AI, search engine, and no tools; the AI group showed the weakest EEG network connectivity and the tool-free group the strongest. - AI users often could not accurately quote the essay they had just written and reported the lowest sense of ownership of their text. - The study measured short-term brain activity during an essay task, not IQ or a decline in intelligence, and it is a small, not-yet-peer-reviewed preprint. - People who thought for themselves first and then used AI kept their memory and brain engagement, so the order of use is the practical key. Since 2025, the idea that relying on AI makes people dumber has been debated intensely. One trigger was an MIT Media Lab study titled Your Brain on ChatGPT. This article checks what that study actually measured and what it did not, and then asks how the findings should be read from the standpoint of IQ and intelligence. ## The design: three groups, four essay sessions [Kosmyna and colleagues (2025)](https://doi.org/10.48550/arXiv.2506.08872) split 54 participants into three groups: one using a large language model such as ChatGPT, one using a search engine, and one using no tools. Each group wrote essays in three sessions under the same conditions. EEG recorded brain activity during writing, and the essays were assessed with natural language processing, human teachers, and an AI judge. Eighteen participants returned for a fourth session in which the conditions were swapped: former AI users wrote without tools, and former tool-free writers used AI. This crossover is the most interesting part of the study. ## Main results: weaker connectivity and poorer memory in the AI group In the EEG analysis, the tool-free group showed the strongest and most widely distributed network connectivity, the search group an intermediate level, and the AI group the weakest. The more the writer leaned on an external tool, the less the brain was engaged during writing. Behavior differed too. AI users struggled to quote a passage accurately from the essay they had just written, and they reported the lowest sense that the text was their own. The tool-free group was the opposite. In the fourth session, the group that moved from AI to no tools remained at low engagement and did not recover immediately. The group that moved from no tools to AI kept high recall and frontal activation, resembling the search group. The team named this pattern cognitive debt. ## What the study did not measure The key point is that the study did not measure IQ or intelligence. It measured brain activity during one task, essay writing, together with memory and ownership immediately afterward. Whether the reasoning ability captured by intelligence tests declined was never examined. The paper was also a preprint that had not completed peer review when posted, with 54 participants and only 18 in the fourth session. Replication is still needed. Stronger EEG connectivity does not directly mean better thinking either; practiced tasks often become more efficient and show less activity. So the study supports the statement that work handed to AI leaves less memory and ownership behind. The leap to using AI lowers intelligence is not supported. ## What matters from an IQ perspective Intelligence is often split into fluid intelligence, solving new problems, and crystallized intelligence, the accumulated store of knowledge and vocabulary (see [Fluid and crystallized intelligence](https://iqcat.vercel.app/en/blog/fluid-crystallized-intelligence)). Crystallized intelligence grows through thinking things through yourself and retaining them. If content delegated to AI does not stick in memory, that accumulation could thin out over the long run. Fluid intelligence, by contrast, is hard to move much with training in the first place (see [Can you raise your IQ?](https://iqcat.vercel.app/en/blog/how-to-raise-iq)), and there is no evidence that AI use drops it abruptly. In terms of IQ test performance, a period of AI use is unlikely to erase the ability to solve unfamiliar figure problems. The more realistic risk is fewer opportunities to exercise working memory, the ability to hold and manipulate a problem in your head, which underlies many IQ test items (see [Working memory training](https://iqcat.vercel.app/en/blog/working-memory-training)). ## In practice: use AI second The simplest measure that follows directly from the results is to change the order. The group that wrote without tools and then used AI preserved both memory and brain engagement. - Produce your own draft or solution first, and ask AI for revision or counterarguments - Restate AI output in your own words before using it - After reading an AI explanation, close it and check whether you can reproduce it unaided How smartphones and notifications affect thinking is covered separately in [Does your phone make you dumber?](https://iqcat.vercel.app/en/blog/smartphone-brain-rot-iq). The background to reports of AI's own IQ is in [Does AI have an IQ over 130?](https://iqcat.vercel.app/en/blog/ai-iq-score). ## Measure your reasoning without tools You only learn where your reasoning stands by measuring it without tools. The [free NekoIQ IQ test](https://iqcat.vercel.app/en/start) scores 20 questions across spatial reasoning, pattern recognition, logical reasoning, and classification (about 10 minutes) with item response theory and shows an estimated IQ, deviation score, top percentage, and strengths by domain. The numbers are estimates, not the results of a formal psychological assessment. ## Frequently asked questions ### Is it true that using ChatGPT makes you dumber? That goes too far. The MIT study showed weaker brain engagement while writing with AI and poorer memory of what was written. It did not measure a drop in IQ or intelligence, and it is a 54-person preprint that has not completed peer review. ### What is cognitive debt? A term proposed by the research team for the state in which delegating thinking to AI lowers the immediate load but forfeits memory consolidation and critical thinking, with the cost coming due later. It is a borrowing metaphor, not an established medical term. ### How can I keep my thinking sharp while using AI? Write or solve first, then hand the result to AI. In the study, the group that wrote without tools and later used AI kept high recall and brain engagement. Restating AI output in your own words also helps. ## Disclaimer NekoIQ provides estimates for entertainment and self-understanding. It is not a medical, clinical, or official psychological assessment. Percentages are derived from a normal distribution with mean 100 and SD 15.