Over a year ago I played an AI song to my grandad, who is a cellist. He did not realise that nobody had played it.
What I find most interesting about AI music is that it has become good, and that to me reveals a lot about creativity itself. A compelling song used to imply that someone had learned an instrument, absorbed thousands of influences, and worked on it until it was right. The same was true elsewhere. A photograph implied that a photographer had stood where the camera stood. A piece of writing implied that someone had wrestled thought into words. An illustration implied a trained eye and hand. A recorded voice implied that somebody had spoken.
Generative AI can increasingly produce the artefact without any of that, which suggests the work behind it mattered mainly for what it taught. Musicians spend years listening until rhythm and harmony become instinct, and writers and painters do the same with language and form. They take in examples and absorb the patterns until they can produce work of their own, which is roughly how a machine learns too. Culture has always worked this way, and music has been especially open about it. Whole genres were built on a handful of drum breaks lifted from old records, and much of what reaches the charts is assembled by teams working from proven chord progressions and sounds anyone can buy. AI shows how much creative skill can be learned.
That unsettles me, because we tend to read a skilled work as evidence of something deeper and more human than skill, and AI lets us prise the two apart. It can produce the signal of virtuosity without years of practice, and the signal of grief without having grieved. Much of what we feel in a song, then, is built by the listener from patterns that could have come from anywhere.
But people absorb more than culture. They fall in love, lose people, and change their minds, and that ends up in what they make. Human creativity is also the world passing through a person and becoming culture, and that is where a point of view comes from. A model has taken in everything at once. It can write in any style but has none of its own, and left alone it drifts towards the average of what it was trained on. Turning 50 Cent's "Many Men" into a 1960s soul record is nowhere near that average. Someone wanted to hear that, used a machine to make it, and the result still made me feel.
So a large part of creativity, the craft, was likely always learnable. That makes AI less of a threat than it first appears. Its real competition is work that had craft and nothing to say. For most of history, having something to say required years spent acquiring the means to say it, and many people who had the first never got the second. If a machine can now lend anyone the craft, what remains scarce is knowing what you want to make, and why. My grandad spent a lifetime on the cello, and a computer produced the same sound. When he found out, he was excited.