Yes—but not in one uniform way. Generative AI can move mental work from recalling, drafting and solving toward prompting, supervising, checking and integrating. That shift may free time for exploration, or it may remove the practice needed to build knowledge. The outcome depends on the task, the user’s effort and whether the system gives an answer or guides the person to reason.
What “AI changes the way we think” really means
The useful question is not whether an AI system magically raises or lowers human intelligence. It is which parts of thinking the person still performs. When a model writes a first draft, proposes code or explains a math problem, it changes the sequence of work. The user may spend less effort producing an initial answer and more effort specifying a goal, judging alternatives, correcting errors and fitting the result into a larger plan.
As an Amazon Associate I earn from qualifying purchases.
That is a change in cognitive workflow, not proof of a permanent change in ability. The same tool can support learning in one setting and weaken practice in another. A student who asks for a solution and copies it gets a different experience from a student who attempts the problem, requests a hint, explains the reasoning and then checks the result. A programmer who accepts generated code without testing delegates understanding; one who uses a model to generate small experiments may explore more possibilities than before.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWindows 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 reinstallAn AIhub Coffee Corner discussion published August 13, 2026 brings together six researchers—Joydeep Biswas, Sanmay Das, Rina Dechter, Sabine Hauert, Michael Littman and Marija Slavkovik—to examine that tension. It is an expert conversation, not a systematic review or a controlled experiment, so its examples and opinions should not be treated as population-wide measurements.
#1 Best Overall
What the AIhub participants agree—and disagree—about
| Theme | Potential benefit | Potential cost or uncertainty |
|---|---|---|
| Producing work | Faster drafts, code and one-off tools can lower the cost of trying an idea. | Less time doing foundational writing or programming may mean less practice. |
| Research and data | Custom tools and help with structured data can make analysis accessible. | Convenience can hide unsupported assumptions or errors that the user does not inspect. |
| Iteration | Rapid alternatives may encourage experimentation and broader exploration. | Rapid approval of plausible outputs may replace slow, deeper reasoning. |
| Learning | Hints, explanations and feedback can provide additional practice. | Answer-first use can improve assisted performance while reducing unaided recall or transfer. |
The participants do not reach a single verdict. Sabine Hauert describes a possibility that AI can make some people think more, not less: “And so that made me think, this whole spiel about it replacing thinking, I think it does that for many things, but for some people, it does the opposite, where they end up thinking much more.” That is an interpretation from the discussion, not a measured effect.
Rina Dechter frames the central boundary: “Since we’re using AI as an assistant now, the question is: to what extent is AI doing the thinking?” The answer changes with how much of the problem formulation, evidence assessment and final judgment remains with the human.
Programming illustrates the trade-off. Joydeep Biswas notes, “But I also think learning how to program on a whiteboard, which is the way I did it, has certain benefits.” A model can accelerate implementation, but mentally tracing an algorithm, designing it from constraints and debugging without assistance can build skills that instant code generation does not automatically provide.
What controlled and survey evidence says
A high-school mathematics trial: assisted practice is not the same as learning
A randomized controlled trial in a Turkish high school compared three conditions for mathematics learning: access to GPT-4 through a standard chat interface, a teacher-informed safeguarded GPT Tutor, and no generative AI. On a later unassisted exam, students who had used the standard chatbot performed worse than the no-AI control group. Students using the safeguarded tutor performed similarly to the control group. The result is reported by Bastani and coauthors in the Proceedings of the National Academy of Sciences (2025): https://doi.org/10.1073/pnas.2422633122.
The finding is specific to that study’s high-school mathematics setting and interface designs. It does not show that every chatbot harms learning. It does show why “students completed more work with AI” and “students learned the material for an unaided test” must be measured separately. A system that supplies a complete answer can raise short-term task performance while reducing the learner’s opportunity to retrieve, plan and verify.
A 274-person study: generated help can match human-authored help
Pardos and Bhandari studied 274 participants across four mathematics topic areas, comparing ChatGPT-generated help with help written by human tutors. The evaluated conditions produced equivalent learning gains, according to the 2024 PLOS ONE paper: https://doi.org/10.1371/journal.pone.0304013.
This result complicates any claim that AI tutoring is inherently harmful. It also has clear boundaries: the study does not establish long-term retention, performance in every subject or the effect of every conversational interface. “Equivalent learning gains” means equivalent under the tasks and assessments used, not proof that a model can replace a teacher in all circumstances.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →A 319-worker survey: reported effort is not a causal decline in critical thinking
Microsoft Research and Carnegie Mellon University surveyed 319 knowledge workers about generative-AI use, perceived cognitive effort and confidence. Their CHI 2025 paper reports self-described work practices and perceptions: The Impact of Generative AI on Critical Thinking.
Survey responses can reveal when people feel they are exerting less effort or relying on a system, but they cannot by themselves prove that AI universally degrades critical thinking. There is no single population-wide effect estimate in this evidence. The practical lesson is to treat confidence as something to test: a fluent answer and a feeling that it is correct are not substitutes for independent checking.
The mechanisms that can change a person’s thinking
Cognitive offloading
External tools have always carried memory and calculation burdens. Generative AI expands the range of tasks that can be offloaded: outlining, translating, summarizing, coding, proposing hypotheses and transforming data. Offloading can free working memory for strategy and synthesis. It can also reduce retrieval practice. If the user never reconstructs the reasoning, the capability may remain in the tool rather than in the person.
From creation to evaluation
AI outputs are usually plausible before they are reliable. Users therefore spend more time deciding whether a response is accurate, relevant, complete, safe and properly sourced. Evaluation is genuine higher-order work, but it requires domain knowledge. A novice may be least able to notice a confident error precisely when the tool is most tempting.
Recommended Free Tools
Lower-cost experimentation
Prompting can make a small script, data transformation or prototype cheap enough to attempt. That lowers the barrier to exploring alternatives and can produce more cycles of hypothesis and feedback. The benefit appears only when the user compares alternatives and learns from the outcomes, rather than selecting the first fluent result.
Changed incentives for foundational practice
When a tool performs an activity quickly, people may practice it less. Writing a paragraph, solving an equation by hand or tracing code can feel inefficient when an answer is one prompt away. Yet these activities build mental models that later make checking possible. A sensible workflow preserves enough unassisted practice to maintain that independent baseline.
Answer-giving AI versus guided practice
The strongest distinction in the evidence is interaction design, not a simplistic “AI versus no AI” split.
| Interaction | What the system does | Likely cognitive pattern | Safer use |
|---|---|---|---|
| Answer-giving | Returns a complete solution, draft or code immediately. | High convenience; risk of shallow processing and weak unaided transfer. | Use after an independent attempt; explain and verify every critical step. |
| Scaffolded or tutor-like | Asks the learner to attempt, supplies hints, probes misconceptions and checks understanding. | Preserves retrieval and explanation while providing feedback. | Set a no-final-answer rule until the learner has shown work. |
| Critique and comparison | Reviews a human-produced answer or presents alternatives with reasons to test. | Shifts effort toward evaluation and judgment. | Require evidence, counterarguments and a human decision. |
The Turkish trial’s different outcomes for a standard chatbot and a safeguarded tutor are consistent with this distinction. They do not prove that one interface will work identically for every learner, but they show that guardrails and prompts can change the educational result.
Free tools Windows power users keep installed
One-click scans. No signup required.
How to use AI without giving away the thinking
For students
- Attempt first. Write a solution, outline or hypothesis before opening the tool. Mark the exact step where you are stuck.
- Request a hint, not an answer. Ask for one question or a small clue, and prohibit a complete solution until you respond.
- Explain back. After receiving help, state the reasoning in your own words and ask the system to identify a specific gap.
- Close the tool and retrieve. Solve a parallel problem or reproduce the method without assistance.
- Keep an error log. Record what you believed, what was wrong and how you detected it; this turns correction into learning.
For programmers
- Define inputs, outputs, constraints and tests before requesting code.
- Ask for alternatives and trade-offs, not just an implementation.
- Read every generated line, run tests that can fail, and inspect edge cases and security assumptions.
- Rewrite or explain the key algorithm without the tool so you can maintain it later.
For writers and analysts
- Draft the claim, audience and evidence structure yourself.
- Use AI for variant outlines, counterarguments or transformations, while keeping source selection and factual claims under human control.
- Check each important assertion against the underlying source; a polished paragraph is not evidence.
- Compare the final text with your original purpose and remove additions that you cannot defend.
A practical test: is AI improving your thinking?
Track outcomes beyond speed. For a task or course, compare AI-assisted and unaided work on measures that matter:
Best Value
- Psychology-Themed Design: These decorative bookends feature a psychology-inspired design, making them a unique and thoughtful addition to any home or office.
- High-Quality Construction: Crafted with durability in mind, these bookends are made from high-quality materials, ensuring they remain sturdy and reliable for years to come.
- Versatile Size: Measuring 6x6.6x1.2 inches, these bookends are the perfect size to support a variety of books, from small paperbacks to larger hardcovers.
- Ideal Gift for Psychology Enthusiasts: Whether it's for a psychologist, therapist, or anyone with an interest in psychology, these bookends make a thoughtful and practical gift.
- Dedicated Customer Service: We are committed to providing excellent customer service. If you have any questions or need assistance, our dedicated team is here to help.
- Independent performance: Can you solve a similar problem without the tool?
- Transfer: Can you apply the idea in a new context rather than repeat the original format?
- Explanation: Can you justify each important step in your own words?
- Error detection: Do you catch plausible but incorrect outputs before they reach someone else?
- Retention: Can you still retrieve the method after a delay?
- Process quality: Did the tool help you test more hypotheses, or did it merely shorten the route to an answer?
If speed rises while unaided performance, explanation or transfer falls, the tool is probably replacing practice. If those independent measures hold steady or improve while you explore more alternatives, the tool is functioning more like augmentation.
Common failure modes and fixes
“It sounds right, so I accepted it.”
Fix: Ask for assumptions, evidence and a falsifiable check. Verify important facts in authoritative sources and test generated code or calculations independently.
“I used AI for every easy step.”
Fix: Reserve recurring blocks for unaided retrieval and production. Use the tool after the first attempt or for feedback, not as the default first move.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problems“The tutor gave too much away.”
Fix: Set a staged protocol: question, hint, learner attempt, targeted feedback, then solution only if necessary. Require the learner to solve a fresh example afterward.
“I became faster but cannot explain the result.”
Fix: Add an explanation checkpoint before delivery. If you cannot describe the method, dependencies and limitations, you are not ready to rely on the output.
What remains unresolved
The available evidence does not support a blanket claim that AI makes people smarter, less intelligent or universally worse at critical thinking. The AIhub discussion records a live disagreement: AI may free some people to think at a higher level while encouraging others to iterate rapidly and approve shallow answers. The studies point in the same conditional direction. Standard answer access can undermine unaided mathematics performance in the Turkish trial, while a safeguarded tutor did not; generated help matched human tutor help in the 274-person PLOS ONE study; and the 319-worker survey describes perceived effort rather than a causal cognitive decline.
The durable question is therefore behavioral and measurable: what thinking did the person do before, during and after the AI interaction? Design the interaction to preserve attempts, explanations, checking and later unaided performance, and AI is more likely to extend human capability than quietly replace it.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
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.
