I've put a great deal of time and effort into researching the present and future of human collaboration with artificial intelligence. It's a fascinating subject - but an extraordinarily murky and opaque one.

In exploring this field, I've stumbled into a rather amusing world. AI has dramatically lowered the barrier to conducting statistical research, and as a result, a flood of people with little real background in statistical modelling have rushed in. Most of them have never even heard that there is such a thing as modeling reality using statistical methods. Students tend to think research is simple: measure something twenty times across different people, capture one data point, throw together a spreadsheet, hand it to an AI, and let it generate some conclusions. What remains completely invisible to them is everything that came before - someone carefully thinking through the study design, correctly identifying the subject of investigation, selecting appropriate participant groups, dividing them into meaningful subgroups, and probing a single phenomenon through dozens of different questions before a conclusion becomes even remotely defensible. When students try to design studies themselves, they tend to pick whatever subject comes to mind - usually an inappropriate one - recruit whatever participants are convenient - usually unsuitable ones - divide them into arbitrary subgroups, and then proceed to apply perfectly correct statistical methods to completely wrong data.

This is roughly the state of affairs in research on human use of AI. Even in serious periodicals, I keep encountering studies that go something like this: we took a group of students - selected by unknown criteria, for unknown purposes - gave them a vague "creative task," had some complete it independently and others with AI assistance, and then fed this entire jumble into a statistical model. No one knows quite what was produced, who produced it, or on what basis. The output is a meaningless soup, run through analytical machinery that was never designed for such undisciplined input.

To summarise in one sentence: students are taught to conduct research according to someone else's pre-designed plan - but that doesn't mean they've been taught how to design a study themselves.

You might ask: where are all the properly trained statisticians who spent five years studying the discipline at university? My answer: they lack adequate material to work with. Most of them haven't accumulated serious hands-on experience with AI, simply because everything has moved so explosively fast. Right now, AI is a space where advertising slogans, uninformed opinions, and the dense fog of commercial secrecy are thoroughly mixed together. Approaches that looked like the cutting edge just one or two years ago have already turned out to be dead ends - unexamined, already withering. What once appeared to be "general principles of working with AI" turned out to be a very narrow set of techniques, useful only to a small circle of specialists doing a limited range of monotypic tasks.

The Main Subject

I got drawn into music with AI. My readers have noticed. I'm a person whose entire musical education amounts to three childhood years of accordion lessons - and yet, on the very first day, I produced something resembling an interesting orchestral sketch. By the third day, I had written the verses for a birthday song for some friends. The final output was a fully produced studio-quality recording - the kind of thing people in the nineties could only dream of. Within two or three weeks, after studying songwriting through YouTube and AI, I was able to sit down and, in a couple of hours, write and produce a complete song or aria from my own lyrics and my own melodies: three or four verses, a chorus, a bridge, intro, outro - the full structure. Many people I know are delighted with the results, though of course it helps when the style happens to match the listener's taste. Two hours is obviously not enough for a deeply meaningful, artistically innovative piece - real substance and ideas take longer to suffer over. But I'm already looking toward symphonies and operas.

I got drawn in for three reasons at once. First, I'm genuinely fascinated by the rapidly expanding capabilities of AI - and by the question of where those capabilities hit a ceiling, or whether they break through one entirely. Second, at my age, there's something in this that feels like a small fulfilment of a dream - a brief sense of what it might be like to be Mozart. Third, and most importantly, I'm a researcher of human-AI interaction, and here I found remarkably fertile ground.

Not only did I manage to make fast and meaningful progress myself - I discovered an enormous and varied population of people encountering AI in music:

• some believe in it,
• others don't,
• some praise it,
• others curse it,
• some are mastering it,
• others are exploiting it,
• some are teaching it,
• some are frightened,
• some are devastated,
• some are retraining,
• and some are just generating noise and trying to make easy money.

What makes this world particularly valuable as a subject of study? In fields like programming, report writing, novel writing, or academic work, it's extremely difficult and time-consuming to understand who you're actually looking at - what is this person's level of competence, ability, IQ, knowledge, professionalism, work ethic, diligence, precision? It's hard to tell whether someone is a craftsman or a master, a professional or a manager, a teacher or a researcher. It's hard to identify what their output actually is, or what raw material went into it. In almost every individual case, you need to spend a long time unpacking the structure of the input, the structure of the AI's contribution, and the structure of what the human actually added.

Did this programmer write a large and complex piece of software - or did they simply happen to find a large, well-structured chunk of ready-made code in a repository? Good luck figuring that out. And if you can't figure it out case by case, how on earth do you build a general picture?

In music, things are far more transparent. Through a MIDI file, you can see exactly what raw material the person started with, and what they actually built from it. You can find out what this person was doing before AI, without it. In one or two minutes, without straining, you can see the results of their work with and without AI - clearly, concretely, side by side. It's easy to see whether someone understands the bigger picture or is thinking only as a narrow specialist.

There are, of course, PR and marketing schemes in the AI music world too. But they're relatively easy to spot - you can read them in what someone says, how often they say it, in what volume, and toward what announced goals.

In programming or even visual art, real specialists often struggle to articulate what they do - verbal expression isn't their strong suit. In songwriting and poetry, language is the native medium. Even music notation, even a MIDI file, is in its way a verbal structure. The work is legible.

What's also fascinating is how clearly a person's prior specialisation shapes how they approach becoming a "complete maestro" with AI. A former song producer and a former guitarist see their roles in AI-assisted music creation very differently. You can hear exactly how yesterday's professional praises or criticises AI musicmaking based on where they're standing. The person who spent the last twenty years digitally processing and perfecting human singing voices will often tell you: "You can hear it's AI - because it does my job too perfectly and too cleanly. There's something inhuman about it." Never mind that for twenty years, they were doing exactly that - making human voices inhuman, smoothed, polished, and bright. You'll hear the opposite criticism: in a demo or amateur version of a song, a tiny background noise that any musical director would scrub out in post is presented as a flaw. "Too much noise - bad. Too perfect - also bad."

There are genuine questions in AI music around the depth and breadth of potential plagiarism. But authors themselves try to make their work distinct from what already exists, and listeners often sense when something sounds like someone else's song. That said, governments will eventually need to modernise copyright law - because what was considered an original composition yesterday is today trivially easy to approximate, the number of composers and songwriters will grow by orders of magnitude, and the pool of "occupied melodies" will expand accordingly. That's a separate and substantial legal discussion.

The point is: without any funding, simply by moving through this space, I was able to observe a real and meaningful diversity of human-AI collaboration - classified by specialist type, by material, by the scope and tractability of results. Even if this isn't a perfect or universal analogy for all intellectual work with AI, the music world provides a fast, accessible starting model for analysing every other domain of human-AI interaction - from education to rejection. It makes visible what is creative and what is mechanical, what is complex-creative and what is complex-mechanical, what is individual and what is derivative, what is innovative and what is mass-produced, what is sophisticated, naive, or merely snobbish - and every mixture in between.

I even discovered this oxymoronic mystery, where sometimes, in collaborative work, it seems that artificial intelligence contributes the creative element, while humans add the necessary mechanical structure.

If the connection to the title wasn't immediately obvious: the music+AI space vividly demonstrates how certain specialisations have been devalued, while others have been amplified and made newly promising. More importantly, it offers something rare - transparency. Unlike most domains of intellectual AI collaboration, music allows you to see the person clearly as a specialist, to observe their work before and after AI, and to examine the results in a way that is genuinely legible. In a transparent collaboration like this, you can actually see what AI devalues and what it amplifies - in work, and in creativity. This makes it a powerful and meaningful model for understanding what the future may hold for every type of professional in a world shaped by artificial intelligence. MIDI visualisations make this especially concrete. Everything here is far more intuitive and visual, without any need to parse every verbal sentence for its semantic content. In music, as in life, AI has dramatically devalued skills that once took decades to develop - while lifting something more creative, sometimes something almost childlike and carefree.