What AI Cannot Replicate About Experience

Here is something I find genuinely fascinating about the way learning works. The most important things we teach are almost never the things we plan to teach. A teacher who shows up every day despite being underpaid and undervalued is teaching something about commitment. A mentor who admits they got something wrong in front of a room full of people is teaching something about intellectual honesty. A parent who sits with a child through a difficult problem without jumping in to solve it is teaching something about patience and belief in another person's capacity. None of that is on a lesson plan. But all of it shapes who a person becomes in ways that outlast any fact they ever memorized.

AI can teach a student to solve a quadratic equation. It genuinely can, and it will probably do it more efficiently than most classroom settings. What it cannot do is model what it looks like to care deeply about a subject, to be visibly moved by an idea, to show a student through lived example that knowledge is not just useful but meaningful. There is something that happens when a human being who has wrestled with a field for decades sits across from someone who is just beginning, and the weight of that experience comes through in how they talk about it. That transmission is not informational. It is relational.

This is what I mean by the invisible curriculum. Every human teacher is always teaching two things at once. There is the explicit content, the facts, the skills, the frameworks. And then there is the implicit content: how to handle not knowing, how to stay curious when something is difficult, how to disagree respectfully, how to find meaning in work that does not immediately reward you. AI systems can deliver the explicit content at enormous scale. The implicit content is something we are in serious danger of undervaluing precisely because it is so hard to measure.

Let me say something about creativity that I think gets glossed over in a lot of these conversations. When we talk about AI creativity, we are usually impressed because it is better than we expected. It produces images that look striking, writing that flows well, music that is technically coherent. And that is genuinely remarkable. But there is a difference between output that is creative and a process that is creative, and I think we sometimes conflate the two. Human creativity is not just about producing interesting outputs. It is about a mind that is actively dissatisfied with existing frameworks, that keeps turning a problem over until something unexpected falls out.

The thing AI does not do, and I say this carefully because I know this territory keeps shifting, is get genuinely bored with its own answers. It does not stare at a finished piece and think, this is technically fine but it says nothing. It does not rebel against the conventions it was trained on. Human creativity involves a kind of productive self-dissatisfaction that drives people to go further than the obvious solution, to ask not just whether something works but whether it matters. Modeling that quality, demonstrating it, showing students what it looks like to keep questioning even when you have a perfectly serviceable answer, is something only a human teacher can do.

And then there is the question of failure, which I think is one of the most underrated teachers in existence. When a human being fails at something in front of another person, and then gets back up and tries again, and eventually figures it out, they are teaching something profound about resilience and process. AI systems do not fail that way. They produce outputs, and those outputs can certainly be wrong, but the system does not experience the failure. The experience of struggling toward mastery, and watching someone else do the same, is one of the oldest and most powerful forms of human teaching there is.

There is a layer to this conversation that deserves more airtime, and it is about values. Not values in the abstract, but the specific, messy, contested, culturally embedded values that shape how human beings actually live together. AI systems learn from human-generated data, which means they absorb the patterns in that data, including the biases, the assumptions, and the moral blind spots. They do not arrive with an independent ethical compass. They reflect back what we give them. Somebody has to do the examining, and that examining is fundamentally a human teaching project.

This is where I think the highest-stakes version of the question lives. It is not really about whether AI can teach children to read or help students understand calculus. The question is whether we are going to remain serious about teaching the things that machines cannot carry for us. The capacity for moral reasoning. The ability to hold complexity without collapsing it into a simple answer. The willingness to sit with discomfort, to revise your beliefs, to take seriously the perspective of someone whose experience is completely unlike your own. These are not skills. They are dispositions, cultivated slowly through the accumulated experience of being in genuine contact with other human minds.

Back to my nephew and his full-marks essay. I am not angry about it. The world he is walking into will absolutely require him to work fluently with AI. But I do hope that somewhere in his education, there is still space for the blank page. For the frustration of not knowing what you think until you try to say it. For the teacher who cares so visibly about an idea that it becomes impossible not to catch some of that feeling. Because that is the thing about what humans teach. A lot of it is contagious. And no matter how capable the models get, I have yet to see a language model make a student feel, in their bones, that thinking carefully about the world is one of the best things a person can do with their time.