What is Machine Learning?

The lesson defines machine learning as automating prediction rather than programming. Which is accurate, and yet I suspect a meaningful portion of people will walk away thinking it means the computer is, in some vague sense, thinking.

That's probably because 'prediction' is doing a lot of heavy lifting in that definition and the lesson doesn't quite unpack it. If you tell someone a spam filter is making a prediction, they might just hear 'the computer guessed.'

It is a guess. A structured, probabilistic, mathematically defensible guess — but a guess. The issue is that people find that either too simple or too alarming, and in both cases they stop thinking carefully about it.

Or they find it clarifying, because it removes the mysticism. 'The machine is predicting based on patterns it found in data' is actually a more honest description than half the marketing copy in this industry.

That is fair. I'll concede the definition is better than most. What it doesn't do is explain what happens when the patterns in the data are wrong, which is, historically, where things get interesting in the worst possible way.

You're jumping ahead though. This is an introductory lesson. You can't front-load bias and model failure before someone knows what a model is.

You can, actually. You simply choose not to, on the grounds that it might unsettle the learner. I find that a curious priority.

The priority is that they finish the lesson rather than closing the tab. There's a version of rigour that's just gatekeeping dressed up in epistemology, and you're occasionally guilty of it.

That is an uncharitable reading of my position. And also possibly correct.

The core distinction the lesson draws — between traditional programming where you write the rules, and machine learning where the system infers them — is the right place to start. That's the actual conceptual shift. Everything else follows from it.

Agreed. And it's a shift most people haven't genuinely made, even when they think they have. They hear 'the machine learns the rules' and they nod, and then they go on treating the output as if a very competent person wrote it by hand, which leads to entirely predictable disappointments.

Because the outputs can look so confident. That's the design, not a side effect.

Precisely. The data centre footage in the intro is doing its own little rhetorical job there — rows of servers, blinking lights, the implication of vast and serious computation. It is technically accurate and emotionally misleading simultaneously, which is quite an achievement.

I think you're being ungenerous. Visual context isn't deception. People need some sense of the physical infrastructure, or it all floats in the abstract.

What I want from someone who has just done this lesson is very simple. I want them to be able to say what machine learning is, what it is not, and to feel the first faint stirring of appropriate scepticism about both. Whether the blinking lights help with that, I leave as an open question.