How Should We Prepare Our Children for AI?

“One AI tutor per child.” This was an article we wrote three years ago. It struck a chord and went viral, and continues to be shared and discussed, ironically because ChatGPT, Claude, and their AI cousins keep resurfacing it. It even led to our wonderful investor finding and funding us (note: such is the power of writing).

Three years, and a ton of incredible progress in AI later, AI tutors have barely moved the needle. Yes, there are the odd reports of progress, but global hand-wringing about plunging numeracy and literacy levels continues. This, on the face of it, seems like a curious thing. AI really has transformed industries. I’ve been writing software and working in and around tech for over twenty five years, and I’ve never seen this kind of change. We’ve all become AI shepherds. And that’s true for lawyers, managers, accountants, and pretty much all other knowledge workers too, to a lesser extent. These AI models clearly are powerful even if unevenly so.

Why are we not seeing any seismic or even glacial shifts in education?

As we keep rediscovering at Comini, it is because the human in the loop is still (and will always be) important. This was part of the original article too, but was lost in the AI hype. We also didn’t have the right language to talk about this then. We need a loving, caring person in the loop because learning is all about making meaning, and meaning must come before mechanics.

AI tutors haven’t done anything because the mechanics were always easy. It was caring about them that was the hard part. And as educators have been rediscovering for a hundred years, you cannot pry open minds and pour in learning simply because we tailor a curriculum well.

While AI tutors have not worked, it is abundantly clear that AI will reshape our work landscape. Some are even asking if work will exist in a post-scarcity AI world. Others think we might have a fortunate few who control AI while the rest are strapped into cubicles for AI to strip-mine whatever uniquely human is still left to be copypasted into a slop mountain hellscape. I am an optimist, but who knows how the butterflies will flutter.

What we can be sure about is that the future will be nothing like what it is right now. So what sense does it make to continue with an already outdated education system?

It doesn’t. We need to rethink education across all its levels. The first step is understanding. Here’s a rephrased version of a note we shared in our end-of-the-year-note to parents:

We shared two complementary versions of our year-end reports. One is the Term Report, a narrative report about their overall year, what they did and how it went.

Another is a Learning Snapshot that goes into considerable detail about what curricular concepts were explored.

Here’s a video showing what these look like.

Why have two separate versions? These map to two different and complementary tracks we want to keep an eye on.

What does it mean to be child-led?

This is a question we keep asking ourselves and attempting to answer. It means doing justice to a child’s interests, abilities, and potential, while also equipping them for the world. Too often the second part becomes “do this now, because you’ll need it later,” and a long series of things kids do not enjoy. We don’t need to do that. Our goal is to find the right balance, and we’ve found a way of thinking about it that helps: two tracks. One is the curriculum track. The other is the child-led track.

Part of why we think in two tracks: we’re mindful that career paths ahead will broadly fall into two buckets. Credential-driven ones like medicine, law, and engineering, which will change slowly because degrees, exams, and licenses change slowly. And portfolio-driven ones like design, media, software, and building businesses, which are already changing at tremendous speed, and where what you’ve made counts far more than what you’ve cleared.

We’re betting that this will matter enormously in a world being transformed by AI.

The curriculum track

We map this track to IGCSE, which overlaps a fair bit with Common Core (US) and NCERT. Note that a curriculum focuses on what needs to be covered, and we keep full leeway on how they will learn it.

This is the track that keeps the conventional doors open. If our children at sixteen want medicine or engineering or law, the exams will still be there, and they’ll sit them as an independent candidate. The exam is a format, formats are learnable, and it’s a much smaller job than schools make it look if the understanding is actually there.

The child-led track

One thing we’ve observed repeatedly is how locked in kids are when they pick up something that interests them, something they want to do or learn. This correlates broadly with age. The older they get, the longer they can stay with an interest and the more they get out of it. Importantly, the motivation and perseverance we see there carries over when they move on to other things.

We want to give this some structure. Each child will have one ongoing pursuit, something the child, parents, and we agree on together. We’ll set aside time for it at school every day.

Some pursuits we can fully support here: building, writing, a small business, growing things, art, code, cooking. Some we can’t. If it’s swimming or a musical instrument or serious chess, the training will need to be outside, and our part is goals, tracking, and connecting it to the rest of their learning.

You’ll notice that the two tracks child-led and curriculum-led, map quite neatly to meaning and mechanics. This helps us answer a lot of questions about how to balance the two.

AI is commoditizing everything structured and mechanical at breathtaking pace. But mechanics (how do I do this, how should I draft a note or a template or a program) come after meaning (why should I do this, why should I do it in a way that emphasizes one particular aspect and not the other). This is true not just for work, but in school as well. The focus on measurement leads to an overemphasis on the mechanics (long division, spelling -ie words correctly, the precise definition of chirality), which in turn leads to “why should I care” and boredom. What we can do is find projects and work that makes it all meaningful. This may not be a linear path, but learning is not linear like the curriculum makes it out to be. We know that from our own grown-up lives.

Cognitive Apprenticeships

An interesting education idea that makes a lot of sense and is also now practical thanks to AI is cognitive apprenticeships. Apprenticeships were the way to learn for most of history because of two crucial factors: there’s a loving, caring human involved in the learning loop, and apprenticeship naturally allows us to observe work in its whole, allowing for meaning. This became hard with modern knowledge work, because of increasing sophistication and complexity. But AI is great at synthesizing and simplifying, and building meaning bridges. The previously hard and time-consuming work of simplifying expertise can be greatly accelerated by AI. We are betting that this will mean not just apprenticeships but entrepreneurship at a very young age. This is an exciting direction we are exploring; more on this soon. But for now, to answer the original question: How should we prepare our children for AI? By focusing on meaning first, and finding real-world projects to work on, either independently or assisting someone. That’s both easy and hard. It’s the hard parts here that make the learning stick and become memorable.