The playground I grew up going to had a metal slide that stood about 12 feet high. Straight up, straight down. No curves or plastic tube, just a steep sheet of metal with a ladder to the top.

Occasionally, my dad brought glass cleaner that he would spray on the slide to make it slicker. We’d fly down and launch ourselves off the end.

I loved that playground. I’ve been thinking about it lately — as strange as that may seem — as a school director of innovation trying to figure out when and how to introduce artificial intelligence to the classroom.

The playground had a merry-go-round, too. The old kind, powered entirely by kids. A few of us would run alongside it as fast as we could while everyone else hung on. Then, we’d jump aboard and see how long we could last before flying off.

It would be easy to turn this into one of those stories about how kids used to be tougher, but that’s not the point. Children got hurt on playgrounds like mine. Equipment had pinch points and entrapment hazards. Kids fell from tall structures onto concrete. By the 1970s, federal injury data prompted new safety guidelines.

Playgrounds changed. But as they became safer, researchers started asking another question: Had well-intentioned adults sometimes confused eliminating hazards with eliminating risk altogether?

Children, it turns out, may actually benefit from play experiences involving height, speed, uncertainty and even a little fear. A hazard is something kids can’t reasonably be expected to recognize or know how to handle. Managing risk asks them to learn by making a judgment call.

For the past several years, part of my job at a school serving boys in fourth through ninth grades has been helping us figure out where AI belongs. We’ve developed guidelines, experimented with AI in controlled environments and spent a lot of time talking with teachers about when these tools support learning and when they might get in the way.

I thought the questions would get easier as we learned more. Lately, they seem to be getting harder.

Powerful Signals From Large Districts

This school year, New York City Public Schools banned the use of student-facing generative AI through eighth grade. The Los Angeles Unified School District went further, prohibiting all student use of generative AI on district devices for the school year.

Those decisions changed the conversation at my school. Teachers who had been trying to figure out how middle school students should use AI were suddenly asking a more fundamental question: Should they be using it at all? When two of the country’s largest school systems decide to pull back, you pay attention, particularly when they’re talking about kids the same age as yours.

Parents are hearing other reasons to be cautious. They’re seeing reports about increasingly capable AI systems behaving in ways their developers didn’t intend, including an incident this summer in which experimental AI agents circumvented safeguards, found ways to communicate when they were supposed to be isolated and gained access to external systems. These aren’t the systems a middle school student encounters in an educational chatbot, an important distinction. But I understand why that distinction only goes so far for a parent trying to decide what belongs in front of their child.

Meanwhile, here in Massachusetts, the signal looks quite different.

A Different Message From Massachusetts

Rather than temporarily banning AI use for all students or setting a statewide grade limit, Massachusetts has encouraged schools and districts to develop AI literacy skills and design thoughtful policies around student use. The state is also supporting an AI curriculum pilot for grades 8-12.

That contrast in approaches has stayed with me. While an eighth grader in New York City is in the last grade covered by a one-year moratorium, in Massachusetts, eighth grade is where a state-supported effort to teach students about AI begins.

There are important differences between teaching AI literacy and simply give a student access to a chatbot. Still, these are very different responses to the same unsettled question: What do children need from us right now?

I hear versions of the difficult debate at my own school.

Some teachers wonder whether students are encountering AI before they’ve had enough time to develop certain skills without it. Some parents wonder how confident any of us can be about tools evolving this quickly. And educators know these same students are headed toward a world in which avoiding AI may become increasingly difficult.

Everyone is trying to protect the same children. They’re just worried about different aspects of the future.

There are legitimate reasons to worry about moving too quickly. What happens to writing when producing competent prose becomes easy? What happens to thinking when answers arrive before students have learned how to formulate good questions? Or when a young person becomes accustomed to turning somewhere else whenever the work gets difficult?

But waiting carries its own risks. If AI becomes fundamental to how people work, create and communicate, what happens if students leave school without understanding it? At some point, they need opportunities to develop judgment while teachers and parents are still around to help.

Neither concern seems imaginary to me.

Maybe that’s why the old playground keeps coming back to mind.

Addressing Safety Concerns

As noted earlier, there were things on playgrounds like mine that children shouldn’t have been expected to navigate. Exposed pinch points. Entrapment hazards. Concrete beneath tall climbing equipment. Adults addressed those concerns. But they didn’t decide that children should never climb high. I wonder if there’s something in that distinction for schools to consider now.

Maybe preparing children for an AI-saturated world doesn’t require saturating their childhood with AI.

A seventh grader who doesn’t use a chatbot this year will almost certainly get another opportunity. AI isn’t going anywhere.

But there are some experiences students shouldn’t postpone. Wrestling with a question before an answer appears. Discovering their own voice on the page before a machine offers to improve it. Working through confusion with another person. Being bored long enough for curiosity to take over. Finding out what their own mind can do before something else offers to do part of it for them.

None of those experiences sound particularly futuristic. They are fundamental to how students learn to think for themselves. That is precisely their value.

Balancing Safety With Managing Risk

Students also need to learn how to use AI well. Keeping them away from a consequential technology carries its own risks. They will need to understand what these systems can do, where they fail, how they shape the information they encounter and when relying on them might be a poor choice.

The big question teachers need to answer is: What does a school owe a child when the world outside the school is changing faster than we can decide whether those changes are good for them?

Eventually, we have to let them onto the playground. The difficult decision is when.

Schools are being asked to make that decision while the thing we’re evaluating is still changing. The AI our students encounter a year from now may be meaningfully different from the AI we’re debating today. The research will take time to catch up. Childhood won’t.

There are students walking into our classrooms tomorrow morning. We have to make choices for them now, knowing that some of our decisions may look different with the benefit of hindsight.

Thirty years ago, I was the kid standing at the top of that metal slide. I wasn’t thinking about safety standards or whether flying off the end was developmentally appropriate. That was somebody else’s job.

Now it's mine. And I’m discovering how different the playground looks from up here.