AI can write the email.
It can summarize the report.
It can explain the difficult paragraph.
It can brainstorm the ideas.
It can solve the problem.
It can tell you what you should consider.
And increasingly, it can do all of that before you have spent much time thinking about the problem yourself.
That convenience raises an uncomfortable question:
If AI keeps doing more of our thinking, what happens to our ability to think for ourselves?
It is tempting to answer that question with one of two extremes.
One side says AI is destroying people's ability to think.
The other says AI is simply another tool, no different from a calculator or search engine, and there is nothing to worry about.
I don't think either answer is sufficient.
The emerging research suggests something more complicated.
AI does not automatically make us worse thinkers.
Used well, it can actually help people compare ideas, investigate problems, question assumptions, and think more deeply.
But when we begin routinely handing over judgment, reasoning, evaluation, and problem solving simply because AI can perform them faster, something changes.
The problem is not merely that AI becomes more capable.
The problem is that we may choose to become less engaged.
Cognitive Dependency Is Already a Public Concern
This concern is not limited to educators or technology critics.
In a large 2026 national survey conducted by Anthropic, cognitive dependency was one of Americans' most frequently identified fears about AI.
People are asking whether constant access to answers will eventually weaken the ability to reach those answers themselves.
Teenagers are raising the same concern.
Recent Pew research found that among teens who expected AI to have a negative effect, overreliance, declining critical thinking, and reduced creativity were among the most common reasons they gave.
The fear is understandable.
For most of human history, if you did not understand something, you had to work through it.
You might read.
Ask someone.
Take notes.
Try something.
Fail.
Think again.
Compare possibilities.
Work through uncertainty.
AI can eliminate much of that friction almost instantly.
And that is both its strength and its danger.
Not All Cognitive Offloading Is Bad
Humans have always used tools to reduce mental workload.
We write things down so we don't have to memorize everything.
We use calculators so we don't perform every calculation manually.
We use calendars so we don't have to remember every appointment.
We use GPS so we don't memorize every road.
This is sometimes called cognitive offloading: moving part of a mental task onto an external tool.
That is not automatically harmful.
In many situations it is extraordinarily useful.
The problem is not that we offload thinking.
The important question is:
What kind of thinking are we offloading?
There is a major difference between asking AI to format information you already understand and asking AI to decide what you should believe.
There is a difference between having AI summarize a hundred-page document and accepting that summary without ever checking the important claims.
There is a difference between using AI to generate possibilities and asking it to choose the right possibility for you.
There is a difference between reducing unnecessary mental labor and surrendering judgment.
Those differences matter.
Efficiency Can Remove Useful Friction
We usually think of friction as a problem.
In technology, good design tries to remove unnecessary steps.
Click less.
Wait less.
Type less.
Think less about routine mechanics.
Most of the time, that is good.
But some forms of friction have value.
Struggling with a question can force you to understand it.
Writing something yourself can expose weaknesses in your reasoning.
Working through competing possibilities can make you recognize what you actually believe.
Trying to explain an idea in your own words can reveal whether you understand it at all.
The easiest answer is not always the answer that produces the deepest understanding.
AI can remove the struggle before the struggle has done its work.
That is where convenience deserves scrutiny.
Research Is Finding Both Outcomes
The evidence so far does not support the simplistic claim that using generative AI automatically destroys critical thinking.
Some research has found that well-structured AI use can support reflection and analysis.
A 2026 meta-analysis examining dozens of empirical studies found an overall positive effect of generative AI on critical thinking among college students.
Other research has found something more specific.
The risk increases when use becomes dependency.
That distinction matters.
Using AI frequently is not necessarily the same thing as depending on AI cognitively.
Dependency begins when we routinely allow the system to perform judgments or reasoning that we no longer meaningfully evaluate ourselves.
One 2026 study examining university students found that focused, productive engagement with AI could accompany stronger critical thinking.
But greater AI dependency was associated with weaker critical thinking.
In other words:
AI use and AI dependence are not the same thing.
That is an important distinction.
The question should not be:
“How often do you use AI?”
A better question is:
“What happens to your own thinking when you use it?”
There Is a Difference Between Assistance and Substitution
Suppose you have to write an argument.
You could ask AI:
“Write an argument explaining why this proposal is wrong.”
Or you could first think through the issue yourself, identify your position and supporting reasons, and then ask:
“Challenge this argument. Identify weak assumptions, missing evidence, and the strongest counterargument.”
Both involve AI.
But cognitively they are very different.
The first can replace much of your reasoning.
The second can pressure-test your reasoning.
One substitutes.
The other assists.
That distinction appears repeatedly throughout The Necessary Evil series.
AI is often most useful when it extends human capability rather than quietly replacing the human activity that gives the work meaning.
Do the First Thinking Yourself
One simple practice can protect a surprising amount of human judgment:
Think before you prompt.
Before asking AI for an answer, ask yourself:
- What do I already know?
- What do I think is happening?
- What am I unsure about?
- What would a reasonable answer look like?
- What decision am I actually trying to make?
You do not need to spend an hour thinking before every simple prompt.
The amount of effort should match the importance of the task.
But for meaningful questions, forming an initial position before seeing AI's answer helps preserve something important:
your own intellectual starting point.
Otherwise AI's first response can become the frame through which you understand the entire problem.
Do Not Let Fluency Become Proof
AI has another cognitive advantage over us:
It can make an answer sound finished before the thinking actually is.
A polished response creates psychological pressure.
It looks organized.
It sounds informed.
It presents conclusions confidently.
That can make us less likely to challenge it.
This is why verification is not only about catching hallucinations.
Verification is also a thinking exercise.
Ask:
- Why do I believe this?
- What evidence supports it?
- What evidence would contradict it?
- What assumptions is the answer making?
- Are there other explanations?
- Is the conclusion stronger than the evidence allows?
Those questions force the human mind back into the process.
Ask AI to Make You Think
One of the best ways to avoid cognitive passivity is to change what you ask AI to do.
Instead of always asking for answers, sometimes ask for resistance.
Try:
“Don't solve this for me yet. Ask me questions that will help me work through it.”
Or:
“Here is my reasoning. Find weaknesses in it.”
Or:
“Give me three plausible explanations without choosing one. I will evaluate them.”
Or:
“What evidence would I need before reaching a conclusion?”
Or:
“Argue the strongest case against my current position.”
Now AI becomes something closer to an intellectual sparring partner.
The machine still assists.
But you remain cognitively responsible.
Keep Some Skills Without AI
There is also value in occasionally doing important tasks without assistance.
Not because AI use is inherently wrong.
Because unused abilities weaken.
If AI writes every first draft, occasionally write one yourself.
If AI solves every technical problem, sometimes troubleshoot before asking it.
If AI summarizes everything you read, occasionally read the entire source and make your own notes.
If AI plans every decision, deliberately reason through some decisions first.
You don't have to reject the tool.
You need to know that the capability still exists without the tool.
That is especially important for skills connected to your profession, responsibilities, relationships, and decision-making.
The Real Risk Is Not Becoming Less Intelligent
I do not think the most useful way to frame this issue is:
“AI is making people stupid.”
That is too simplistic, and the research does not justify it.
The deeper concern is cognitive surrender.
A person may remain perfectly intelligent while gradually becoming less willing to exercise that intelligence.
AI can make it easier to skip:
- uncertainty,
- research,
- reflection,
- comparison,
- struggle,
- judgment,
- and verification.
If we repeatedly skip those things, the danger is not that the machine somehow steals our intelligence.
The danger is that we stop practicing parts of it.
Convenience Should Serve Human Capability
The purpose of AI should not be to see how little humans can think.
It should be to help humans accomplish more while retaining the capabilities and judgment that make meaningful work possible.
There are many things AI should save us from doing unnecessarily.
Reformatting repetitive information.
Searching large bodies of material.
Generating routine drafts.
Organizing notes.
Finding patterns.
Comparing alternatives.
Reducing administrative burden.
Those efficiencies can free people to spend more time on higher-level thinking.
But that only happens if we actually use the saved time for higher-level thinking.
If every cognitive efficiency merely becomes an opportunity to offload the next level of judgment too, eventually very little remains with the human except approval.
And even approval becomes meaningless if the person no longer understands what they are approving.
A Five-Question Thinking Check
Before accepting significant AI-assisted work, ask yourself:
- Did I understand the problem before asking AI?
- Can I explain the answer in my own words?
- Did I evaluate the reasoning instead of merely reading it?
- Could I recognize if the answer were wrong?
- Am I still making the final judgment?
If the answer to several of those questions is no, the problem may not be that AI is doing too much.
The problem may be that you are doing too little.
