There is a question I ask myself every time I open my morning playlist and discover that an unfamiliar song has, within thirty seconds, perfectly captured a mood I hadn't yet consciously registered. The question is simple, yet profound: who did this? A person with years of instinct and the scars of failed releases – or a machine that has never lost anything, but has instead calculated everything?
An Old Game with New Cards
The music industry has always been a game of chance. In the mid-20th century, label producers made decisions almost blindly: they listened to demos, trusted their inner voice, and sometimes failed so spectacularly that their mistakes became legends in themselves. It is said that several major labels once turned down bands that would go on to rewrite the history of rock music. The names of those bands are known to all, while the names of those who said “no” are known to almost no one.
This was the era of intuition elevated to an absolute. The producer is a strange creature: half merchant, half shaman. They must simultaneously think about sales and feel the pulse of the street, combining cold calculation with what cannot be explained in words. This profession has existed for over a century, and in all that time, its fundamental mystery has not changed: why does one song touch the heart, while another does not?
Now, a new player has joined the inquiry. The algorithms of streaming platforms – Spotify, Deezer, and their peers – have amassed a volume of data that no producer of the past could have dreamed of. They know the exact moment a listener skips a track. They see at what tempo people keep listening and at what tempo they get distracted. They analyze tonality, rhythm, the structure of verses and choruses, the frequency of chord changes. And based on all this – they predict.
What Exactly Can a Machine Calculate
To understand how serious this competition is, one must have at least a rough idea of what we are dealing with. The algorithms of modern streaming services do not simply act as genre search engines. They build multi-dimensional models of musical preferences – both individual and collective. Every listen, every addition to a playlist, every repeat play is a signal that the machine interprets and integrates into the overall picture.
Spotify once developed a tool called “Loud & Clear” – a system for tracking how revenue is distributed among artists based on their position in recommendation algorithms. But another direction is far more interesting: so-called hit prediction – forecasting a hit even before its release. Several independent research projects are attempting to create models that determine a track's commercial potential from its acoustic characteristics. Some have claimed an accuracy of around 60–70 percent.
Is sixty percent a lot or a little? If you're playing roulette, it's fantastic. If you're a producer building an artist's career over five years, it's catastrophically low. But it is also too high to simply be dismissed.
History Repeating Itself Under a Different Guise
I cannot resist a brief historical digression – not for the sake of academic posturing, but because the picture would be incomplete without it.
In the late 19th century, the science of so-called “musical psychology” was actively developing in Europe. The German psychologist Carl Stumpf and his colleagues sought to understand why some sound combinations are perceived as harmonious and others as dissonant. This was the first serious attempt to explain a person's musical reaction from a scientific standpoint. It seemed then that the mystery was on the verge of being solved.
Over a hundred years have passed. The mystery remains. Only the tools have become incomparably more powerful.
We find ourselves at the same point again: we have a highly complex apparatus that can describe music with incredible accuracy, but still cannot explain why this particular melody at this particular moment sends shivers down your spine. The algorithm is Stumpf's heir, only with incomparably greater computing power and an incomparably larger dataset. The essence of the question is the same.
The Producer as a Historically Conditioned Being
Now, let us think about the producer. Not as an abstraction, but as a specific person with a biography.
A good producer is not someone who guesses. It is someone who shapes the context in which a song acquires meaning. They know that at a given moment, people are tired of a certain sound and are ready for something new. They sense that societal tension has reached a point where a slow ballad will land better than an energetic track. They hear in a budding artist's voice not what is there now, but what it could become in two years.
This knowledge is not stored in databases. It lives in experience – personal, cultural, historical. It is passed on through conversations, through failures, through chance encounters in small clubs where no one yet knows that something important is being born.
An algorithm works differently. It learns from what has already happened. Its prediction is an extrapolation of the past into the future. And therein lies its fundamental limitation: it cannot predict a rupture. It cannot see the moment when everything familiar suddenly stops working and something fundamentally new takes the stage. Because the new has no historical data.
The Paradox of the “Next Bowie”
Allow me to formulate this idea through a specific paradox, which I privately call the “problem of the next Bowie.”
Early in his career, David Bowie did not fit into any existing mold. His music was too strange, his image too provocative, his lyrics too literary for the mass market. If an algorithm trained on the hits of the 1960s had existed at that time, it would have, with high probability, rejected his early recordings as commercially unviable. And it would have been right – by the standards of what already existed. But it would have been catastrophically wrong – by the standards of what was yet to come.
This is precisely where the watershed lies. The algorithm optimizes within a known space. The visionary producer expands that space itself.
But – and here I must be honest – such producers are a rare few. The majority operate within the same templates as any algorithm, only less accurately. They too look at the charts, they too are guided by what has already worked, they too are afraid to take risks. For them, the algorithm is not a competitor, but simply a more efficient colleague.
When Data Begins to Shape Reality
There is another aspect that concerns me far more than the question of predictive accuracy. It is the question of the direction of influence.
Initially, the algorithm was created to reflect listeners' preferences. But over time, something significant happened: the algorithm began to shape those preferences. When a platform with hundreds of millions of users recommends a certain type of music, that type of music begins to be heard everywhere. Artists who want to get on algorithmically promoted playlists begin to tailor their work to the parameters that the machine deems successful.
We get a feedback loop: the algorithm recommends what worked well before → listeners hear it more often → the algorithm confirms that it works → artists create more of the same. The circle closes, and the space of the possible imperceptibly narrows.
This is not a conspiracy or malicious intent. It is simply the logic of optimization taken to its extreme. Medieval scribes did not seek to destroy the diversity of manuscripts either – they simply copied what was in demand. The result was similar: massive circulation for some texts and the complete disappearance of others.
What the Numbers Themselves Think
I promised myself not to overuse statistics, but a few observations are nevertheless necessary – not to argue with numbers, but to feel the scale of what is happening.
Studies conducted in various periods over the last decade have identified one consistent trend: the average length of intros in popular tracks has been roughly halved compared to what was common twenty or thirty years ago. This is a direct consequence of algorithmic pressure: platforms define a “track skip” within the first thirty seconds, and artists adapt – the intro is removed or compressed to a minimum, and the hook appears immediately.
This is not necessarily a bad thing from the perspective of an individual listener. But it is a serious change in musical grammar, one that occurred not by the will of composers or producers, but under the pressure of an algorithmic stimulus. A kind of evolution under external environmental pressure – only here, the environment is code, not nature.
Tool or Oracle?
Let us return to the original question: can an algorithm predict a hit better than a producer?
In my view, the question itself is slightly imprecise. It is like asking if a weather radar can predict the weather better than an experienced sailor. In the short term, over a limited horizon – probably, yes. But the sailor understands the sea differently: they feel what the radar does not register and make decisions under conditions of uncertainty that an algorithm simply cannot handle.
The algorithm is a brilliant tool for working within the known. A producer with true instinct is a navigator at the border of the known and the unknown. Their areas of expertise do not overlap – they complement each other.
The problem is not that the algorithm exists. The problem is when it begins to be seen as an oracle, not a tool. When a label refuses to sign an artist because “the data doesn't support it” – that is not progress. It is a surrender of imagination to the convenience of numbers.
The Beauty of the Predictable and the Value of the Unpredictable
In the end, there are two types of pleasure from music, which exist in parallel and never cancel each other out.
The first is the pleasure of recognition. You hear a track that perfectly matches your mood, your genre expectations, a familiar sound. It is pleasant, it is comfortable, it works. The algorithm is better at creating this pleasure than most people.
The second is the pleasure of discovery. You hear something unexpected, something that knocks you out of your usual rut, something that makes you stop and listen again. It is not what you wanted – it is what you did not know you wanted. The algorithm is significantly worse at creating this pleasure, because by its very nature, it contradicts the logic of optimization.
The history of music is the history of the second type of pleasure. Jazz, rock, punk, hip-hop, electronic music – each of these phenomena was an unpredictable breakthrough that no algorithm of its time could have foreseen. And each of them emerged thanks to people who heard something that did not yet exist.
This is not nostalgia for the “good old days.” It is simply an observation about how cultural evolution works. It does not move along optimal trajectories, but through random leaps, mutations, and mistakes that suddenly turn out to be discoveries.
Instead of a Conclusion: The Question That Remains
I began with the question of who compiled my morning playlist. And now, after all these reflections, I am not sure that the answer is more important than the question itself.
Because behind this question lies another, deeper one: what do we want from music? If we want comfort and predictable pleasure, the algorithm will perform brilliantly, and soon, even better. If we want music to sometimes do something to us that we did not ask for, then we need people who can hear what does not yet exist.
Both desires are legitimate. Both have always existed. The Stoic philosopher Marcus Aurelius wrote about the difference between what is needed and what is wanted. Music, perhaps, is the one area where this distinction ceases to matter: here, the necessary and the desired can coincide, or they can diverge so far that one could spend a lifetime trying to catch up.
The algorithm knows what you want today. The best producer knows what you will want tomorrow – even if you yourself do not suspect it yet. And the greatest among them knew what you would want in ten years. That is why their records are still not gathering dust on a shelf.
Everything new is old, but with a filter. The question is who adjusts this filter: a person who hears the future, or a machine that remembers the past.
