Banning AI Is the Riskiest Thing You Can Do for Your Kids: Two Dads on Teaching Children to Learn in the Age of AI

Who we are: Two dads approaching AI from opposite perspectives

DROdio is a tech CEO building Chief, an AI chief-of-staff platform for founders and executives. He is a dad of three kids ages seven to twelve.
Arthur Levitan is an ordained Taoist master and a former CEO. He built a company to 500 employees and more than $100M in revenue using Daoist principles, and it became a Harvard Business School case study. He was ordained in 2016 by the temple where he studied in China, and he wrote Outlast: 30 Years of Wisdom from Business, Tao, and Life. Arthur has a fourteen year old daughter.

We come at this from opposite ends. DROdio is excited about what AI gives us. Arthur worries about what it takes from us. We agree on more than you'd expect.

The mistake most adults are making

The default response from many parents and schools is fear. Kids will just get the answers, they'll stop thinking, and they'll cheat. So the policy is to keep AI out of the classroom and out of the house.

I think that instinct is understandable and wrong. It protects the adult's sense of safety more than it protects the child's future. We've watched this exact reaction repeat across 2,400 years of technology.

Every generation of adults has done this

Writing (around 370 BC). Socrates warned that writing would "implant forgetfulness" in people's souls. People would rely on external marks instead of their own memory, and they would "seem to know much" while knowing little. We know about that warning only because Plato wrote it down.
The printing press (1470s). Scribes petitioned Genoa to outlaw it. In 1565 Conrad Gessner warned that too much reading was "confusing and harmful."
The loom and the lightbulb. The Luddites smashed looms, and New York lamplighters went on strike against electric light. The Industrial Revolution hurt some workers in the short run, but it ultimately created far more jobs than it destroyed. The workers who suffered most were often the ones unable or unwilling to retrain. Arthur makes the same point about farming. In the US it went from roughly 98% of employment to about 2%, and people found new work.
The calculator (1970s). This is the closest match to today's AI fight. A 1975 New York Times piece describes teachers who "have gone so far as to ban them from their classrooms," set against a University of Maryland mathematics professor who said, "If a teacher finds it necessary to ban calculators, then what he's teaching is trivial." A 1979 federal review noted that nearly 100 studies had been run in just a few years, mostly to ask whether calculators would harm achievement. It concluded the answer "continues to be 'No.'"
- The later evidence agrees. A 1986 meta-analysis of 79 studies found that, at every grade except fourth, calculator use alongside traditional instruction improved the average student's paper-and-pencil skills. It also found that students who used calculators had better attitudes toward math and a better self-concept in it. A 2003 meta-analysis of 54 studies concluded that calculator use "did not hinder the development of mathematical skills."
- The grade-four exception matters, and I'll come back to it.

The workplace computer (1980s). Parents who kept their kids away from computers were making a bet. Alan Krueger's study of 1984 to 1989 data found that workers who used a computer on the job earned roughly 10 to 15 percent more than otherwise similar workers. He also estimated that the spread of computers accounted for between one-third and one-half of the increase in the return to education during those years.[1] The kids who learned the tool early got the premium.
- Arthur lived a version of this. At 13 his father gave him an Apple IIc, and he was the only kid handing in typed homework. His grades went up because the work was better. He says learning to use a computer helped him more than the mental math his father drilled.

The cost of keeping AI from kids is not zero

When parents ban AI, they treat that choice as free. It isn't. These are the costs I see.

1. A real and growing market penalty. PwC's 2026 analysis of more than one billion job ads found that the average wage premium for AI skills is now 62%, up from 57% the year before. Jobs requiring specific AI skills are growing 69% against 9% for the overall job market. This is a consulting firm's analysis of job postings, so treat the exact numbers with caution. The direction matches Krueger's computer premium four decades earlier.

2. The entry-level ladder is changing. In the same PwC data, AI-exposed entry-level roles are seven times more likely to require traditionally senior skills like judgment and leadership. Those roles are growing, while other entry-level roles are shrinking. A kid who can't work with AI and also hasn't built judgment is squeezed from both ends.

3. You give up the most scalable tutor ever built. A World Bank randomized trial in Nigeria gave secondary students a six-week after-school program using GPT-4 as an English tutor, guided by teachers. Students scored 0.23 standard deviations higher in English. The researchers equated the result to 1.5 to 2 years of business-as-usual schooling, and the gains also showed up on the regular third-term exam. Having a patient, personalized tutor available at any hour used to be a luxury. A ban gives it up.

4. You teach the wrong lesson about learning. If kids learn that the powerful tool is something to hide, they use it in secret, with no adult helping them think about it. Banning doesn't make the behavior go away. It removes the guidance.

5. You build fragility, not resilience. Protecting a kid from the world they will actually live in doesn't make them tougher. It makes them less prepared.

The part the "ban it" camp gets right

Here is the evidence that should humble the AI enthusiasts:

A randomized trial of nearly 1,000 high school math students compared three groups: a standard ChatGPT-style tutor, a tutor with guardrails, and no AI. Both AI groups did much better on practice problems. But when AI was taken away for the exam, students who had used the plain version scored 17% worse than students who had never had AI. The guarded tutor, which gave hints rather than answers, largely erased that harm. Students without guardrails used the tool as a "crutch," copying solutions, and they didn't realize their learning had suffered.

A separate trial in an intro programming course (N=275) found something similar. Students using unrestricted ChatGPT scored higher on the exercise but showed no greater learning gains than the control group. The researchers called it a "comfort trap," where students prefer the easy tool even though it doesn't teach them more.

This echoes the grade-four exception in the calculator research. Sustained use at that age hindered basic skills in average students, even though the tool helped at every other level. The tool is rarely the issue. The age and the way it's used decide the outcome.

Another trial points the other way. In a randomized experiment with 334 university students, unrestricted access to an AI tutor outperformed restricted access by 0.21 standard deviations. Restricted access produced intense bursts of prompting that disrupted learning flow. The benefits were largest for students with stronger self-regulation skills.

Put together, the evidence says:

Unguarded, answer-on-demand AI can harm learning, and kids often can't tell.
Well-designed, hint-first, guided AI helps, and in some settings helps a lot.
Blanket restriction doesn't solve the problem and can backfire.
Self-regulation is what separates the kids who benefit from the kids who coast.

So the real fight is not AI versus no AI. It's guided use versus unguided use, and teaching kids to learn is what makes guided use work.

Teaching kids how to learn in this world

This is what we think parents and educators should do.

1. Teach the loop, not the tool

Have them attempt the problem first. Then ask for a hint, not the answer. Check the output against what they know. Then explain it back in their own words without the screen. That loop is the skill. If a child can't do the last step, the AI did the learning for them.

2. Run the "tool goes down" test

This is Arthur's test. The question isn't which tool the child uses but who is using it. When the tool disappears, does the child break down or step up and figure it out? The work may be worse without it, but it should be adequate.

3. Teach them to drive stick

If self-driving cars become safer than human drivers, I want my kids to use them most of the time. I also want them to know how to drive manually. Arthur plans to teach his daughter stick shift, and I'm teaching mine too. His worry is specific: if she hits self-driving every time she's in trouble, she never learns how to get out of trouble. The skill we're teaching is the judgment of when to use the tool and when to take over. The US Navy brought back celestial navigation after ship collisions for the same reason. A baseline of skills protects you when the tool fails.

4. Protect pattern recognition on purpose

Arthur's firmest point is that pattern recognition is a fundamentally human trait. Reading faces, tone, and rooms is how we bond with people and sense danger. If kids outsource it, they lose something we may not get back. He asks his daughter to analyze the people around her instead of looking at her phone. I agree, with a nuance. AI is excellent at finding creative connections humans miss, but you need your own pattern recognition first so you can tell what you're looking at.

5. Don't outsource self-understanding

Arthur sees adults asking ChatGPT to describe their own psychological state. It often says something beautiful that misses the point, and it means they never learn their own internal signals. Kids should use AI to learn about the world. They shouldn't use it to find out who they are.

6. Stop filling their heads with knowledge

This is where we converged after some pushback. I'm trying to decide what is worth filling my daughter's head with and what can be handed to AI so she can focus higher. Arthur's counterpoint is that some things, like sensing the world around you, can't be offloaded. We agreed that the goal is skills, not stored facts. The skills that kept coming up were staying aware, staying focused, and being resilient.

Arthur frames it as knowledge versus wisdom. Knowledge is information anyone can acquire. Wisdom is knowledge integrated into your life as a practical tool. AI makes knowledge always accessible. Schools still focus on memorization and rarely teach integration.

7. Teach them to ask, not just answer

Arthur's strongest claim is that teaching kids to create questions matters more than teaching them to find answers. Wisdom is asking the right question at the right moment. His method is simple. What do you believe? Where did that belief come from? Did you make it up, or hear it somewhere? A chatbot will confidently answer anything. The kid who can interrogate the answer wins.

8. Don't interfere with who they are

This is the Taoist part, and it matters most to Arthur. He doesn't teach his daughter in the usual sense. He waits for her to ask, then guides her thinking rather than imposing his own. Most parents unconsciously try to self-actualize through their kids and remake them in their own image. He calls non-interference "the hardest thing to do because you think you know best." You think you're course-correcting, but you're copying and pasting your life onto them. He's also clear that this isn't a sink-or-swim model. Not interfering is different from abandoning.

That principle applies to AI, too. If your fear of the technology is really your own discomfort, your child pays for it.

What to do Monday morning

For parents:

Let your kids use AI, and sit with them while they do. Ask them to show you how they used it.
Set a house rule: attempt first, ask for hints, check the output, explain it back.
Make "where did you hear that?" a regular question.
Build in deliberate no-tool practice: reading people, navigating without GPS, solving a problem on paper.
Pay attention to age. Younger kids building foundational skills need more guardrails than teenagers.

For educators:

Stop designing assignments an answer machine can finish, and don't rely on bans you can't enforce.
Use hint-first tools, or set up prompts that guide rather than answer. The research suggests this is where the benefit is.
Grade the process: the attempts, the reasoning, the explanation. Include some assessment with no tools, so you can see what the student can actually do.
Teach AI literacy as its own skill, including hallucinations, bias, and over-reliance. The Nigeria program started by teaching exactly that.

What not to do:

Don't ban AI out of fear.
Don't hand it over with no guidance.
Don't let kids offload self-understanding or social perception.
Don't rescue them from every difficulty. The more a child avoids difficulty, the less resilience they build.

One remaining tension between me & Arthur

We didn't fully resolve one point. I believe kids gain more by offloading some rote skills and moving up the stack. Arthur thinks we may be giving away something fundamental that we can't get back. Our shared answer is to keep both views in tension. Build the baseline first, then upscale it, and keep checking which side of the line a given skill sits on.

What we both reject is the idea that the safe choice is to wait. The kids will live in a world shaped by these tools whether or not their parents are comfortable with it. The best thing we can give them is the ability to learn, question, and stay themselves while using the most powerful tool in history.

Where do you draw the line between "let the tool do it" and "make them learn it the hard way"? We'd like to hear how other parents and educators are handling it.