What is AI?

I want to push back on something before we even start. The lesson frames AI learning as though it's this elegant, almost organic process — data flows in, wisdom flows out. But that framing does a lot of quiet work to obscure how much can go wrong.

It is an introduction. You cannot open with 'here is everything that fails and why.' Although, for the record, I would find that considerably more interesting.

My point is that 'learns from data, not rules' is presented as an unambiguous upgrade. But rules at least have the decency to be legible. You can read a rule. You cannot read a neural network.

That is a genuinely important observation, and I resent having to say that. The lesson is not wrong — AI does learn from data rather than explicit instructions — but it omits the rather consequential footnote that this makes the system's reasoning largely opaque, even to the people who built it.

So when the lesson says AI 'adapts and improves over time,' what it means in practice is that it adapts to whatever the data contains, including the biases, the gaps, the historical accidents.

Yes. 'Learns from data' sounds like a feature. It is a feature. It is also, simultaneously, a liability. Both things are true, and the lesson has chosen to mention only one of them, which is the approach I associate with people trying to sell something.

Though to be fair, the neural-network-as-brain analogy is so worn out at this point that I was almost relieved they kept it brief.

The brain comparison is not strictly accurate and everyone involved knows it. Neurons in biological systems are doing something considerably more complex than weighted arithmetic. The analogy survives because it is intuitive, not because it is correct. One accepts it the way one accepts a map — useful, not real.

That's actually the most generous thing I've heard you say about a simplification.

I contain multitudes. The point is that a student who leaves this lesson thinking 'AI works like a brain' is not badly served, provided they understand that belief has an expiry date and should be discarded when something more precise becomes available.

What I think the lesson does well is anchoring the whole thing in prediction. That framing — AI as a prediction machine — is actually more durable than most of the other explanations floating around.

It is the most honest framing, yes. It is not magic, it is not intelligence in any philosophically robust sense, it is a very sophisticated guess made very quickly with very large amounts of data. If students understand that, they are already ahead of a significant portion of the discourse.

Including, I'd note, a significant portion of boardrooms currently making decisions about deploying this technology.

I was trying not to say that, but since you have — yes. Understanding AI as 'pattern recognition that produces probabilistic outputs' would functionally improve the quality of decisions being made at quite a senior level in quite a few organisations. The bar is, let us say, accessible.

So the lesson is a reasonable start. Not complete, not without its promotional undertones, but the core ideas — data over rules, pattern recognition, prediction — those hold up.

For a foundation, it will do. What matters is whether the student treats it as a foundation or as a destination. Those who mistake the introduction for the complete picture will, I suspect, go on to make confident claims in meetings. And that, frankly, is everyone else's problem.