Recommended Free Tools
There is no good evidence here that ordinary AI use lowers general intelligence or causes permanent brain damage. The more specific concern is real: if you let AI do the thinking you need to practise, you may finish a task without getting better at doing it yourself. Other studies find benefits, too—depending on the task and how people use the tool.
What these studies measured—and what they did not
“Dumber” is a broad claim about a person’s intelligence. The studies below instead measure narrower things: performance on particular tasks, brain activity during a writing exercise, self-reported effort at work, or whether learning transfers to a later task without AI. Those outcomes are not interchangeable.
![Claude AI for Beginners Bible: [5 in 1] The Ultimate Guide to Automate Your Work, Save Hours Every Week, and Use AI for Real-World Results](https://m.media-amazon.com/images/I/415+fSJacsL._SL160_.jpg)
As an Amazon Associate I earn from qualifying purchases.
A chatbot can help someone produce a better answer right now without teaching them how to produce that answer unaided later. To assess whether AI supports learning or replaces practice, it matters whether the user can explain, recall, or apply the result after the tool is gone.
Free tools Windows power users keep installed
One-click scans. No signup required.
How the evidence compares
| Study | Participants and setting | Evidence type | Finding reported by the source |
|---|---|---|---|
| MIT essay-writing study | 54 participants in the first three sessions; 18 completed session four. Essay-writing task. | Small experiment; the paper is an arXiv preprint. | Brain-only participants showed the strongest and most distributed connectivity; LLM users the weakest. The study also analyzed essays, recall, and self-reported ownership. |
| Microsoft Research knowledge-worker study | 319 knowledge workers who used generative AI at work at least weekly; 936 first-hand examples. | Survey of self-reports and associations. | Greater confidence in AI was associated with less reported critical-thinking effort; greater self-confidence was associated with more. |
| MIT Media Lab misinformation study | 67 participants in a month-long study judging news items with and without a chatbot. | Study involving repeated judgments and later unassisted judgments of unseen items. | Accuracy during AI-assisted interaction rose about 21 percentage points on average; unassisted performance on new items did not improve over the study. |
| MIT Sloan workplace creativity experiment | 250 employees at a technology consulting firm in China. | Randomized workplace field experiment. | Employees with ChatGPT access received higher creativity ratings only when they showed stronger metacognitive strategies. |
| MIT Media Lab learning-mode comparison | 50 participants learning nuclear safety protocols. | Comparison of unrestricted chatbot, Socratic hint-based chatbot, and adaptive tutor modes. | The unrestricted chatbot produced higher immediate learning gains; the adaptive mode showed higher EEG engagement. |
Why the MIT “brain on ChatGPT” result is easy to overread
The MIT Media Lab publication page, dated June 10, 2025, describes an essay-writing study comparing an LLM, a search engine, and a brain-only condition. Its reported EEG differences are about connectivity during that task, not a direct measure of general intelligence. The page reports lower alpha and beta connectivity for the LLM-to-brain group in the fourth session. The study also considered what participants wrote, remembered, and felt ownership of.
The sample was small, and the paper is identified on arXiv as a preprint. Its findings can raise questions about how answer-first assistance affects engagement in essay writing; they cannot establish that AI generally makes users less intelligent, or that any effect is permanent.
Getting help is not the same as learning
The MIT Media Lab misinformation study makes the distinction especially clear. Participants made judgments with chatbot assistance, then later judged unseen news items without it. The overview reports that performance during AI use improved, but that unassisted performance on new items did not improve across the study; it also says 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.
That result is specific to this misinformation task and study design. It does not show that every AI-assisted activity weakens a skill. It does show why an accurate answer produced with help is not, by itself, evidence that the user learned to reason better.
Why workplace reports do not prove skill loss
In the Microsoft Research survey, higher confidence in generative AI predicted less reported critical-thinking effort, while higher confidence in one’s own abilities predicted more. Participants described their work shifting toward verifying information, integrating AI responses, and managing the task.
These are reported associations from knowledge workers, not a longitudinal test showing that AI caused lasting cognitive decline. They suggest that confidence and the kind of work people delegate may shape how much effort they invest; they do not tell us that every user, job, or use of AI has the same effect.
AI can support creativity or learning, depending on the interaction
Creativity depends on how people work with the tool
In MIT Sloan’s June 23, 2025 summary of a randomized field experiment at a technology consulting firm in China, access to ChatGPT improved supervisor and external evaluator creativity ratings only for employees who demonstrated strong metacognitive strategies: planning, monitoring their process, and changing strategies when needed. The summary says those strategies can be taught. As Jackson G. Lu, the study’s lead author, put it: “Metacognition — thinking about your thinking — is the missing link between simply using AI and using it well.”
This is a conditional result, not evidence that simply turning on AI makes everyone more creative. The difference is whether a person steers the work—setting a direction, judging suggestions, and revising—or accepts the first output.
Immediate gains and active engagement can diverge
In MIT Media Lab’s comparison of three ways to learn nuclear safety protocols, the unrestricted chatbot gave participants the strongest immediate learning gains, while the adaptive tutor mode showed higher EEG engagement. The overview says unrestricted-chatbot users often retrieved direct answers; Socratic users initially reasoned through hints but tended to disengage later. The project summary does not establish which approach produces better long-term retention, so immediate scores and engagement should not be treated as the same outcome.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Use AI without skipping the thinking you want to keep
These studies point to a practical test: is the tool helping you inspect and improve your reasoning, or replacing the practice that would build it? Before accepting an answer, try this sequence:
- Form a first attempt. Write down your answer, plan, or explanation before asking AI, even if it is incomplete.
- Ask for scrutiny, not just a solution. Request a counterargument, a hint, or questions that would help you find a gap in your reasoning.
- Verify important claims. Check the relevant sources and evidence rather than treating fluent output as proof.
- Revise and explain. Use the tool’s response to improve your work, then explain the result in your own words.
- Try again unaided. For a skill you want to learn, see whether you can solve a similar problem without the tool.
If you can defend the answer and reproduce the reasoning later, AI may have served as a scaffold. If you can only repeat what it gave you, you may have completed the task without practising the underlying skill.
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.
