What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
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
Yes, AI can change how we think—but not in one universal direction. It changes the work required to produce an answer, draft, program or analysis: more effort may move from recalling and composing to prompting, supervising, checking and integrating. That shift can free attention for exploration, or it can remove the practice that builds understanding. The outcome depends on the task, the user and the design of the interaction.
The most useful question is therefore not whether AI makes people smarter or less intelligent. It is whether a particular AI use requires you to reason, retrieve and explain—or lets you approve a plausible answer without doing those things yourself.
What the AIhub discussion actually concludes
AIhub’s Coffee Corner discussion published on August 13, 2026 brings together Joydeep Biswas of the University of Texas at Austin, Sanmay Das of Virginia Tech, Rina Dechter of the University of California, Irvine, Sabine Hauert of the University of Bristol, Michael Littman of Brown University and Marija Slavkovik of the University of Bergen.
It is an expert conversation, not a controlled experiment or a systematic review. The participants describe what they see in programming, research work, writing and data analysis, but those observations should not be read as measurements of how an entire population has changed. Their disagreement is itself important: some see AI as reducing drudgery and making deeper exploration possible; others worry that rapid generation and approval will weaken foundational skills and sustained thought.
#1 Best Overall
Hauert captures the conditional nature of the issue: “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.” Dechter asks the sharper diagnostic question: “Since we’re using AI as an assistant now, the question is: to what extent is AI doing the thinking?”
The cognitive shift: from making an answer to managing a process
Generation becomes cheaper
A chatbot can produce a first draft, code fragment, explanation or data transformation in seconds. That lowers the cost of trying an idea and makes one-off tools practical for people who could not have built them from scratch. In the AIhub conversation, participants describe quicker coding, custom tools for particular investigations and work with structured data as useful forms of augmentation.
Cheap generation can increase the number of ideas you test. Instead of abandoning a question because writing a script or formatting a table would take an afternoon, you can explore several approaches and compare their results. This is a real change in the economics of thinking: more attention can go to choosing questions, designing tests and interpreting outcomes.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Evaluation becomes more important
Generated work is not self-validating. The human role moves toward stating constraints, checking sources, testing edge cases, detecting unsupported assumptions and deciding whether the result answers the original question. A fast answer that is never examined is not equivalent to a carefully reasoned answer, even if both look polished.
This is why AI assistance can feel mentally easier while still demanding high-level judgment. A user may write fewer lines of code but spend more time deciding whether the code is safe, complete and appropriate. A writer may edit more than they drafted. A manager may compare several generated scenarios but still carry responsibility for the decision.
Rank #2
Iteration can become either exploration or superficial approval
Rapid cycles are valuable when each cycle has a hypothesis, a test and a reason for accepting or rejecting the result. They become shallow when the user asks for another version until one sounds convincing. The interface does not determine which pattern occurs; prompts, time pressure, incentives and the user’s knowledge all matter.
What controlled evidence says about learning
Two education studies point in different directions, which is exactly why “AI helps” or “AI harms” is too broad a conclusion.
| Evidence | What was compared | Result | How far the result applies |
|---|---|---|---|
| Bastani and coauthors, Proceedings of the National Academy of Sciences, 2025 | A randomized high-school mathematics trial in Turkey: standard GPT-4 chat, a teacher-informed safeguarded GPT Tutor, or no generative AI | Standard chatbot access improved assisted practice, but students later performed worse on an unassisted exam than controls. The safeguarded tutor group performed similarly to the control group on that exam. | Applies to the studied school, mathematics tasks, interfaces and testing design. The abstract record does not provide a single participant count to quote. |
| Pardos and Bhandari, PLOS ONE, 2024 | ChatGPT-generated help versus help written by human tutors across four mathematics topic areas; 274 participants | The evaluated conditions produced equivalent learning gains. | This does not establish long-term retention, transfer to every subject or identical results with every chatbot and prompt. |
The Turkish trial separates two outcomes that are often confused. Students can perform better while an AI assistant is present and still know less when the assistant is removed. Assisted performance measures what the human and tool can accomplish together; an unassisted exam measures what the learner can retrieve and apply independently. Those are related but not interchangeable.
The safeguarded tutor result also shows that “AI tutoring” is not one intervention. A system that gives an answer immediately can remove the need to attempt a problem. A system that asks the learner to try, offers a hint, requests an explanation and checks understanding can preserve more of the learning process.
What a workplace survey can—and cannot—tell you
A Microsoft Research and Carnegie Mellon University paper presented at CHI 2025 surveyed 319 knowledge workers about generative-AI use, perceived cognitive effort and confidence. Its findings describe self-reported work practices and perceptions. They are not a causal demonstration that AI universally degrades critical thinking, nor do they measure a population-wide decline in intelligence. Read the paper for the exact survey measures and analysis: The Impact of Generative AI on Critical Thinking.
The practical implication is narrower and more useful: when people delegate a task, they may report investing less effort in producing it and more effort in reviewing the output. Whether that review is rigorous depends on expertise, available time and the ability to notice a plausible-looking mistake.
Recommended Free Tools
Answer-giving and guided practice are different tools
When choosing how to use an AI system for learning or skill development, compare the interaction rather than the marketing label.
| Interaction pattern | What the user does | Likely short-term benefit | Main risk |
|---|---|---|---|
| Answer-giving | Requests a solution, accepts it and moves on | Speed and immediate task completion | Less retrieval and fewer opportunities to discover misconceptions; assisted success may not transfer to unaided work |
| Scaffolded guidance | Attempts first, receives hints or questions, explains steps and checks the result | Feedback while preserving practice | Requires more time and a system that can maintain an appropriate level of difficulty |
| Evaluation partnership | Generates alternatives, tests them against criteria and documents why one is preferred | Broader exploration and clearer comparison | Automation bias if the user treats fluent output as evidence |
For a learner, a good default is to ask for a hint, counterexample or diagnostic question before asking for a complete solution. For professional work, ask the system to state assumptions, identify failure cases and propose tests. These prompts do not guarantee correctness, but they make the human’s reasoning visible and easier to inspect.
When AI can deepen thinking
Exploring more alternatives
AI can generate multiple approaches to a design, outline competing explanations or create small simulations that would otherwise be too time-consuming. The benefit comes from comparing the alternatives and deciding what evidence would distinguish them, not from accepting the first output.
Building small, disposable tools
In the AIhub discussion, participants describe custom research tools and structured-data work as areas where assistance is useful. A domain expert can describe the required input, output and constraints, then inspect and test the generated script. The tool removes repetitive implementation work while leaving the question, validation plan and interpretation with the expert.
Making feedback available during practice
The PLOS ONE study’s equivalent gains for ChatGPT-generated and human tutor-authored help suggest that, in the evaluated mathematics settings, generated explanations can support learning when they function as help rather than a substitute for all effort. The result is encouraging but bounded: it does not prove durable retention or guarantee quality in another subject.
When AI can weaken thinking and skill
Skipping foundational practice
Foundational abilities are built through repeated retrieval and problem solving, not only through exposure to a correct explanation. If a learner always asks for finished code, prose or calculations, they may lose the fluency needed to recognize a bad answer or work when the tool is unavailable. Biswas makes this point about programming: “But I also think learning how to program on a whiteboard, which is the way I did it, has certain benefits.”
Confusing polish with understanding
A coherent paragraph or elegant proof can conceal a wrong premise. Fluency also creates confidence, and confidence can reduce the checking that would reveal an error. Ask yourself whether you could reproduce the key reasoning, explain why the answer is valid and adapt it to a new case without copying the same text.
Letting the tool choose the question
Generated options can narrow attention to what is easy to phrase or measure. Deeper work often begins by defining a problem, deciding what counts as evidence and noticing what is missing. If the system supplies the framing as well as the answer, the user may never examine those choices.
Windows 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 reinstallOutdated 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 matchA practical protocol for using AI without outsourcing your thinking
- Make an unaided first attempt. Write a thesis, solve the first problem, sketch the algorithm or list the facts you already know. This creates a baseline for learning and exposes what you do not understand.
- State the goal and constraints. Tell the system the audience, allowed methods, definitions, format, safety limits and what a successful result must include. Vague requests encourage generic answers.
- Request guidance before completion. Ask for a hint, a Socratic question, a worked example with a different set of numbers or a list of likely misconceptions. Delay the full solution until you have made a genuine attempt.
- Explain the output in your own words. After reading an answer, close the chat and reconstruct the reasoning. If you cannot, you received information but have not demonstrated understanding.
- Verify independently. Recalculate important figures, run code on edge cases, check primary sources and compare the answer with course notes, documentation or domain rules. Treat citations and confident wording as claims to inspect, not proof.
- Test transfer. Solve a new problem, modify the code for a changed requirement or explain the idea to someone unfamiliar with it. Transfer is a stronger check than repeating the assisted example.
- Keep a decision record for consequential work. Note which parts came from the system, which checks you performed, what uncertainty remains and why you chose the final option. This makes review possible later.
How to tell whether AI is doing the thinking
- Can you state the problem and success criteria without showing the generated response?
- Could you identify a plausible error in the output and describe a test for it?
- Did you make a prediction or attempt before seeing the answer?
- Can you solve a nearby problem, explain the result aloud or adapt the method to a new constraint?
- Are you spending more time evaluating evidence and assumptions than merely choosing the most fluent wording?
- When the tool is unavailable, can you still perform the essential parts of the task?
If the answers are mostly no, the system is probably replacing practice rather than augmenting it. That does not make every use wrong; it means the task is being optimized for immediate completion instead of independent capability.
Best Value
What this means for common uses
Programming
Use AI to generate boilerplate, explain unfamiliar APIs or propose tests, but preserve manual practice with core data structures, debugging and system design. Review dependencies, permissions, error handling and performance instead of treating compilable code as correct.
Writing
Let a system suggest outlines or point out unclear passages after you have decided the argument. Keep ownership of the claim, evidence and voice. Compare an AI edit with your original and make each change for a stated reason.
Mathematics and other study
Prefer hints, incremental feedback and questions that expose a misconception. Periodically work without assistance under exam-like conditions. The Turkish trial shows why assisted practice and unaided performance must be monitored separately.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Repair Windows errors before they cause bigger problems3Fix the driver behind crashes, sound loss and screen glitchesResearch and analysis
Use AI to transform data or generate candidate hypotheses, then inspect the source data, definitions and assumptions yourself. A custom tool is valuable only when its outputs are tested against known cases and its limitations are recorded.
Workplace decisions
Ask for competing options and failure modes, not just a recommendation. Assign a person with relevant expertise to verify high-impact claims, and retain an audit trail when errors would affect customers, safety, finances or legal obligations.
What remains unresolved
The long-term direction is still an open question. The AIhub participants disagree about whether easier generation will release people for deeper thought or encourage rapid, shallow iteration and approval. Existing evidence covers specific classrooms, mathematics tasks and self-reported workplace behavior; it does not settle how years of routine AI use will affect memory, expertise, creativity or judgment across every field.
The defensible conclusion is conditional. AI changes the distribution of mental work. It can expand exploration and provide useful tutoring when the interaction preserves attempts, explanations and checks. It can also remove exactly the practice that produces independent skill when it supplies answers with no requirement to reason. Measure both immediate performance with the tool and unaided performance without it, and design the workflow around the capability you want to keep.
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.
