iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
There is no good evidence here that everyday AI use lowers general intelligence or causes permanent brain damage. But some studies raise a narrower concern: when people let AI do the thinking, they may get help with the task in front of them without getting better at doing similar tasks alone. Other research finds benefits when users actively guide the tool—or gains in immediate learning that do not necessarily mean deeper engagement. The important question is not simply whether you use AI, but whether it replaces or supports your own reasoning.
What does “dumber” mean in these studies?
“Dumber” sounds like a change in general intelligence. The studies most often cited in this debate do not directly measure that. They look at narrower outcomes: performance on a particular task, brain activity during a task, self-reported effort, or whether learning transfers to a later task without AI.
Those outcomes are not interchangeable. An AI tool can help you get a question right now without improving your ability to answer a different question later. A brain-signal measure can describe engagement during one activity, but it is not an IQ score. And a survey can reveal patterns in what people report without showing that AI caused lasting skill loss.
So the defensible concern is about practice and transfer: if you routinely outsource the steps that build a skill, you may get fewer chances to practise those steps. Whether that happens depends on the task and how you use the tool.
#1 Best Overall
What did the MIT essay-writing study actually find?
MIT Media Lab’s June 10, 2025 summary of “Your Brain on ChatGPT” describes a small experiment in which participants wrote essays using an LLM, a search engine, or no tool. The first three sessions included 54 participants; 18 completed a fourth session. The researchers examined EEG connectivity, essays, and participants’ reports of ownership and recall. The summary reports the strongest and most widely distributed connectivity in the brain-only group, moderate engagement in the search group, and the weakest connectivity in the LLM group. It also reports lower alpha and beta connectivity for participants who moved from LLM use to writing without a tool in the fourth session.
These are results from a particular writing task and a small sample—not a direct finding that AI reduces general intelligence. EEG connectivity is a measure of brain activity patterns, not a verdict on a person’s intelligence or long-term cognitive health. The paper is identified as a preprint on arXiv; the MIT Media Lab publication page gives the study’s summary and date.
Rank #2
Does AI assistance improve independent performance later?
A separate MIT Media Lab study examined whether people could carry learning from an AI-assisted task into a later task without AI. In a month-long study, 67 participants repeatedly judged news items with chatbot assistance and later judged unseen items without it. During AI-assisted interaction, accuracy rose by about 21 percentage points on average. The overview says participants did not improve their unassisted performance on new items over the study, and that performance immediately after AI use declined over time. Interactions in which the chatbot supplied answers or facts directly were associated with worse later independent performance.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallThis is a useful distinction between doing better with help and learning to do better alone. It does not show that every kind of AI assistance harms transfer: the task involved judging news items, and the overview describes a particular study rather than all AI use. The MIT Media Lab project overview summarizes the design and findings.
What does workplace research say about critical thinking?
A Microsoft Research study presented at CHI 2025 surveyed 319 knowledge workers who used generative AI at work at least weekly and collected 936 first-hand examples. Its summary reports that greater confidence in AI was associated with less reported critical-thinking effort, while greater self-confidence was associated with more. Participants also described work shifting toward verifying information, integrating AI responses, and managing tasks.
This was a survey of self-reports and associations, not a longitudinal test of cognitive ability. It cannot establish that AI caused lasting decline. It does, however, suggest that confidence may shape whether people scrutinize a generated answer or accept it with less effort. See the Microsoft Research study summary for the study details.
Can AI help people think more creatively or learn?
Creativity depended on how employees used the tool
A randomized field experiment summarized by MIT Sloan involved 250 employees at a technology consulting firm in China. Employees with access to ChatGPT received higher creativity ratings from supervisors and external evaluators only when they showed strong metacognitive strategies: planning their approach, monitoring how it was going, and revising their strategy when needed. The result argues against a simple claim that AI always makes users worse thinkers; access alone was not the condition linked to higher ratings.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Jackson G. Lu, the study’s lead author and a Sloan School Career Development Associate Professor of Work and Organization Studies, put the point this way: “Metacognition — thinking about your thinking — is the missing link between simply using AI and using it well.” MIT Sloan’s June 23, 2025 summary says these strategies can be taught and describes the experiment.
Best Value
Immediate learning gains and engagement did not move together
In another MIT Media Lab project, 50 participants learned nuclear safety protocols using one of three modes: an unrestricted conversational chatbot, a Socratic chatbot that offered hints without final answers, or an adaptive tutor guided by cognitive engagement. The project overview reports higher immediate learning gains in the unrestricted-chatbot condition, but higher EEG engagement in the adaptive mode. Participants often used the unrestricted chatbot to retrieve direct answers; Socratic participants initially reasoned through hints and then tended to disengage.
The findings show why an immediate test result and a measure of engagement should not be treated as the same outcome. The overview is a project summary, so it supports these reported comparisons but not stronger claims about long-term retention or which approach is best for every learner. See the MIT Media Lab overview.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How to use AI without handing over the practice
When the goal is to learn, improve judgment, or develop an idea, use AI in a way that leaves you responsible for the reasoning. A practical sequence is:
- Try first. Write down your answer, approach, or draft before asking for a solution. Even a rough attempt gives you something to compare and revise.
- Ask for support, not just an output. Request a hint, a counterargument, an explanation of a step, or questions that expose gaps in your reasoning. For a task where you need to practise retrieval or judgment, avoid asking the tool to decide for you.
- Check the result. Verify factual claims against reliable sources and inspect the reasoning. Treat fluent wording as presentation, not proof.
- Revise deliberately. Decide what to keep, correct, or reject, and be able to explain why. That keeps the final judgment with you rather than the tool.
- Test yourself without assistance. Close the chat and explain the idea, solve a fresh problem, or make a new judgment on your own. If you cannot, the tool may have completed the task without teaching you the skill.
For routine work where speed matters more than skill-building, direct AI assistance may be a sensible trade-off. For learning, writing, analysis, or consequential decisions, the extra effort of forming an initial view and checking the answer is part of the work—not friction to eliminate.
What the evidence supports—and what it does not
| Study | Who and what | Reported result | What it can tell you |
|---|---|---|---|
| MIT essay-writing study, summarized June 10, 2025 | 54 participants in the first three sessions; 18 in the fourth. LLM, search-engine, and no-tool essay writing. | Brain-only participants had strongest and most distributed connectivity in the summary; LLM users had weakest. Lower alpha and beta connectivity was reported in the LLM-to-brain fourth-session group. | A small, task-specific study of EEG connectivity and essay work; not a measure of general intelligence or permanent damage. The paper is labeled a preprint on arXiv. |
| Microsoft Research CHI 2025 survey | 319 weekly workplace GenAI users; 936 first-hand examples. | Higher confidence in GenAI was associated with less reported critical-thinking effort; higher self-confidence with more. | Self-reported associations, not proof that AI caused cognitive decline. |
| MIT Media Lab misinformation study | 67 participants in a month-long study of assisted and later unassisted news judgments. | Accuracy during AI-assisted interaction rose about 21 percentage points on average; the overview reports no improvement on new unassisted items over the study. | Assistance during this task did not establish independent transfer; it does not settle outcomes for other tasks. |
| MIT Sloan workplace creativity experiment, summarized June 23, 2025 | 250 employees at a technology consulting firm in China; randomized workplace field experiment. | Higher creativity ratings with ChatGPT access appeared only among employees showing strong metacognitive strategies. | Evidence that outcomes can depend on how people engage with AI, not access alone. |
| MIT Media Lab learning-mode comparison | 50 participants learning nuclear safety protocols with unrestricted, Socratic, or adaptive tutoring. | The overview reports higher immediate gains for unrestricted chatbot use and higher EEG engagement in the adaptive mode. | Immediate test gains and EEG engagement diverged; the project overview does not establish long-term retention. |
Across these studies, “AI effect” is too broad a label. The tasks, participants, time horizons, and evidence types differ. Some findings concern assisted accuracy, others self-reported effort, EEG activity, creativity ratings, or performance without AI. Taken together, they justify a practical distinction—not a verdict that AI makes people globally less intelligent.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
