This is a weird time to be a teacher for many reasons — but one big one is this:
You’re supposed to be teaching students to think critically. And yet, more and more, it seems like we can all ask computers to do something that looks a lot like thinking.
In this post, I’m going to share the framework I currently use with teachers for how to make sure you’re still teaching students to think critically in the age of AI.
But first, it probably makes sense to back up and ask: what exactly do we mean by critical thinking?
What Even Is Critical Thinking?
We need a good working definition of the type of critical thinking that teachers want to teach students in the first place.
Then we can talk about what parts of that are under threat from AI — and what parts we might even be able to use AI to help with.
I’m going to start with a definition that was generated by AI, but I promise we’re not stopping there.
I’m using this in kind of the same way that students do when they start an essay with The Oxford English Dictionary definition — not the most original way to start, but it can be a good jumping-off point, because it gives us something that at least feels objective.
Then we can work our way into the fuzzier areas of what critical thinking actually is.
Here is what Anthropic’s Claude gave me as a working definition to discuss with teachers:
Critical thinking is the disciplined habit of examining ideas carefully before accepting or acting on them. Asking why, considering evidence, identifying assumptions, recognizing bias — including one’s own — and evaluating whether conclusions actually follow from the reasoning behind them.
My first reaction is — like my first reaction to a lot of what AI produces — this sounds like a pretty good, workable definition!
That’s often the first step of the emotional arc I go through when I’m asking AI to help me with something. But then I look more closely.
And I can’t help noticing it’s a little light on specifics.
For example, if a teacher I was working with pushed back and said, “What do you mean by teaching students to ask why?” — I would probably immediately start stuttering.
That’s because I’d be trying to elaborate on an idea that really isn’t my idea, Which is kind of the definition of a cliché: it sounds good, but it’s not your idea, and you can’t actually unpack the thinking behind it.
Also: “Asking why” sounds, on the one hand, like exactly what education is supposed to teach you to do from pre-K through grad school.
But it also sounds like exactly the way a student might derail your lesson plan, sending you down a series of off-topic rabbit holes.
It also sounds a little bit like something you’d hear in a professional development session where the presenter sounds confident but cannot answer any follow-up questions.
The same goes for considering evidence and identifying assumptions.
Both good! Both a big part of the point of education! Both a little light on specifics.
And both of these things are things that conspiracy theorists will also tell you they’re doing, so it’s not an exact match for what we want to teach students about critical thinking.
We’re seeing that something can sound good but still not quite be usable, because it’s not exactly saying what we’re trying to say.
The final two examples from the AI definition — recognizing bias, including one’s own and evaluating whether conclusions actually follow from the reasoning behind them — feel closer to what I was looking for.
At the very least, these feel like something that could become a one-day lesson plan for high school students.
That’s a starting point.
Okay — the way I just broke down that AI answer, I’m not the first person to ever think about doing that.
There are certain issues that keep coming up over and over again as educators wrestle with how to teach students to think in the age of AI.
One common recommendation is to question everything AI tells you. I’m going to explain why I don’t think that’s the best approach — but I do want to acknowledge that it’s meant to address some real and important issues.
Problem 1: AI always sounds authoritative, even when it’s wrong.
AI can give you wrong or incomplete answers, and they still sound smart and coherent.
This can take a few different forms.
You’ve probably heard a lot about AI hallucinations.
You may have also heard that AI is trained to be pleasant and agreeable.
It tells us our ideas are good.
We’re so right!
We’ve asked a fascinating historical question!
In some cases, that can lead people down a path of reinforcing delusional thinking.
And all along that path, AI will keep sounding friendly, grammatical, and coherent. It will keep finding you witty and insightful. Your historical questions will still be fascinating.
For our purposes here, let’s treat all of the above as one big problem: With AI, you’re going to get a good-sounding answer even if it’s not a good answer.
Problem 2: AI does the thinking for you.
The most common concern here is that the AI easy button can make people mentally lazy.
But it can also get out in front of a person and make them think that something is their idea when it’s not quite their idea.
This was what happened when I first read the AI definition of critical thinking above.
I would not use “telling students to ask why” as part of my personal definition of critical thinking, but it sure did sound smart when I first read it!
It’s easy to get stuck in a loop trying to justify an AI-generated idea, trying to pull out the thinking that led to it, when there was no thinking that led to it.
It was just words that sounded good.
All this assumes I’m even trying.
What most teachers probably worry about, most of the time, is that students aren’t trying to do that work.
They’re trying to have something to turn in for a grade.
Which means they’re potentially not getting any of the mental exercise that is supposed to make their brains stronger.
So, that’s the second big problem: AI taking over parts of our thinking that we really should be doing ourselves.
Problem 3 is that AI can feel like talking to a human, but it isn’t.
And sometimes you really should be talking to a human.
Why “Question Everything AI Tells You” Is the Wrong Advice
These are three major challenges we face when teaching students to think critically in the age of AI.
But I don’t think the answer is: question everything AI tells you.
Our brains are wired to find mental shortcuts because we don’t have time to second-guess everything we take in. We’re constantly assessing whether we trust the person giving us information, looking for outward signs of whether something is legit, or just matching new information against what we already know.
Those are all mental shortcuts. AI is just a new tool in the toolbox.
Telling students to question everything AI gives them is not realistic advice. And because it’s not realistic, it’s not a good way to teach students to think critically.
The reason people use AI as a mental shortcut is because it does make a good mental shortcut.
Does it sometimes make sense to check the output of AI? Absolutely. But instead of telling students not to use AI as a shortcut — which they probably won’t listen to — what I would say instead is:
Using AI as a shortcut can be great. It can be a huge benefit. But it also has risks. You should know the risks, and then decide if those risks are worth it for what you’re about to use AI for.
Students should understand that AI can sound authoritative while giving them wrong or incomplete answers. They should understand that there are parts of their thinking they should not outsource. And they should understand that there are times when it is better to talk to a human.
But there are also times when, after considering all of these issues, AI makes a fantastic shortcut. Part of teaching critical thinking is knowing when those times are.
Here’s the framework I currently use with teachers to talk to students about critical thinking in the age of AI.
Part of critical thinking is knowing what is the best source of information for what you need in the moment. There are four major sources students should know about, and they should develop the ability to identify which one is most likely to serve them best.
1. AI. Yes, this is a legitimate source — and sometimes the best one.
2. Another human being. Someone accessible to you, someone you can ask and get a direct answer from.
3. A non-human, non-AI source. A book, Google, an encyclopedia, your textbook. I’m oversimplifying by lumping these together, but they occupy a similar category for our purposes.
4. Yourself. Things you just know. This is actually a combination of two things.
- There are gut feelings and body reactions: if something feels a little off about an interaction, that’s a data point you should pay attention to, and you can’t get that from AI, another person, or a textbook.
- The other part is things you’ve learned so well that they’re part of your basic knowledge base.
This brings us back to the AI definition of critical thinking, where one element was recognizing bias, including your own bias — because part of the job of education is to train people to have better gut feelings.
Or at least gut feelings that are based on some type of objective knowledge.
If you know basic math and someone tells you two plus two is five, you already know that’s not true. The same goes for knowing the scientific method, knowing a basic outline of history.
You need enough general knowledge to provide a scaffold for new information coming at you, because that helps you spot misinformation and disinformation.
That’s a big part of what school is for.
When AI Is Actually the Right Choice
The framework above lets us acknowledge that sometimes AI is your best source of information.
One of those times is when you genuinely need a shortcut — the quickest possible way to balance out a gap in knowledge.
Another is when you want a human-like answer without a human interaction.
Because yes, sometimes you should be talking to a human. But there are times when a human touch is not a plus — it’s a drawback. We get self-conscious in front of other humans. Nobody likes to feel like the one person who doesn’t get it, and AI is not going to make you feel that way. You don’t have to be self-conscious that someone wants to know why you need to know this, or why you don’t already know it. AI isn’t going to give you an attitude like, “Why do you care so much?” Think about how helpful that can be for a teenager struggling with material in front of a class of other teenagers, or someone with a pressing question that’s just embarrassing to ask in public.
AI can help people fill in their knowledge gaps without all the baggage of interacting with another human. And this is especially important because not every human-to-human interaction is meant to be educational.
In many cases, humans are looking to take advantage of a gap in knowledge.
Here’s an example from my own life. I’m on a plan where my dealership does basic maintenance on my car for free, and they’re all nice people. That being said, I have walked out of that dealership with some incredibly high bills after my regularly scheduled free maintenance. Why? Because they do a check on your car and give you a list of recommended treatments, and it is really hard to know what to say yes to — what’s necessary, what’s not a big deal, what’s kind of a scam.
Now, if I had a cousin who was a mechanic, I would call that cousin and say, “Do I really need to have this part removed, cleaned, and put back in? Or is this just something mechanics say because they assume you don’t know better?” I don’t have a cousin who’s a mechanic. But I have gotten in the habit of checking with AI every time I get that list of recommended repairs before I approve any of them.
Some of those repairs have seemed necessary. Some could go either way. But for some of them, I’ll get an answer like, “This is usually not needed unless you have a very high mileage vehicle” or “If your car is not making this particular sound, you actually don’t need to worry about this.” In those cases, it has saved me a lot of money — because it made it much easier for me to say no without feeling like I was putting my family in danger.
That’s one example of using AI to quickly balance out a knowledge gap that could leave you at a disadvantage. But it’s also worth admitting that sometimes we just are too lazy to learn something unless we can learn it quickly.
In recent weeks, I have used AI to look up whether German chocolate cake is really from Germany, what kangaroo diseases can be transmitted to humans, and what that spinning thing is on top of a Waymo car.
Now I know all three of those things — and I never would have bothered to look them up otherwise.
Knowing When Not to Use AI
That said, there are absolutely times when AI is not where you want to turn.
There are times when you should double-check its outputs.
And there are times when you should double-check any piece of information, from any source.
Checking information from more than one source is a basic research skill — and teaching research skills in the age of AI is a topic for another post.