Can My AI teach me how to play trombone?

blog@dws.team
September 11, 2025
3 months ago
Can My AI teach me how to play trombone?

AI can teach us a lot of things, but conveying tacit knowledge is not one of them.

AI can teach us a lot of things, but conveying tacit knowledge is not one of them.

At the back of my office there's a music school for brass instruments.

Imagine a courtyard that has been taken in by a dozen competing lords, who have erected walls in some uneasy truce. That’s the rear view of my office. Or rather, the rear sound. The high buildings surrounding that courtyard make for a perfect sound garden.

The music school must be located somewhere in the high buildings. They've many pupils, but it’s the beginners who have the most impact on me. In hot weather I like to open the back door, but when lessons start I close it lest the sounds have a disruptive effect on my mental health. Drive me crazy.

Poor teacher. The beginnings of mastering trumpet, trombone or tuba all too often sound like someone dying.

Apparently, the thing to learn first is embouchure. Literally, “how to get your mouth around it”.

Got me thinking. If My AI is so smart, can it teach me to play trombone?

Not that trombone interests me all that much. No talent at all: after a lifetime of practise I can barely strum along to the house of the rising sun.

The pupils at the music school are undoubtedly more endowed. But they must first master the “getting sound out”: embouchure.

Their teacher is human. Actually, we never met, but I think it’s a reasonable assumption.

Being human, there is a mouth, and that mouth has spent decades practising embouchure. Trial and error. Might be able to tell you the moment the mouth first got the proper sound out of a brass instrument. But couldn’t tell you how they did it.

That kind of knowledge is tacit, gained through lived experience. Can’t be translated into language.

My AI is a very sophisticated language machine. Decades of trial and error went into the development.

Early researchers built what we call “expert systems”, a set of rules that worked within a very narrow scope. Any effort to expand the scope failed, continuous failures led to what people in the field call the “AI Winter”: the collapsing of interest, and of funding, in AI.

Even when the field picked up again, optimism was hit down time and time again. Trial and error. But now we have this general purpose language machine, that can string tokens into words, words into sentences. It can tell you what embouchure is like nothing we have ever built before. But you don’t ask it to show you how, because you know it can’t.

Many AI breakthroughs relied on the tacit knowledge of researchers. Intuitive leaps, unspoken insights, experiential wisdom. Engineers often "just know" which parameters to tweak, based on feel. Based on tacit knowledge.

Tacit knowledge is learned through experience. It played a big role in the development of My AI, but My AI itself can’t muster tacit knowledge.

So no, My AI CAN’T teach me how to play trombone. No mouth. Missing the “bouche” of embouchure, missing all the other bits and pieces as well, it can’t teach me anything about being human.