During the last years, a lot of effort has been made in computer sciences to try to apprehend music as it has been done for language. This endeavor has seen significant strides, with researchers attempting to break down the complexities of music into analyzable and computable elements. Indeed, most of these attempts at replicating human musical creativity are now made through the looking-glass of extreme mathematisation and formalisation of the musical language. This approach generally reduces music to a series of equations and algorithms, where the nuances and subtleties of artistic expression risk being lost in translation. This is often done without daring the first delve into the decades, if not centuries, of artistic research and reflexion that has been made on the musical language, preferring to barely imitate the misunderstood surface of Western tonal music. As a matter of fact, there is a tendency to overlook the deep, rich history of music theory and practice, choosing instead to replicate a superficial layer of what is often mistakenly perceived as universal musical standards, particularly those rooted in Western tonal traditions.
The fact that some people perceive music mainly (if not only) through the lens of mathematics, statistics, and models does not mean that music is merely a direct application of probabilistic processes. Music, at its core, is an emotional and cultural phenomenon, deeply intertwined with human experience. In my view, this perspective only indicates that such individuals may have an overly strong tendency to analyze the world around them through an ethnocentric logical-scientific framework. This inclination often leads to a narrow understanding of music, where its essence as a form of human expression is reduced to quantifiable metrics, stripping away the layers of cultural, historical, sensorial, emotional, and philosophical significance that music embodies.
Connections between music and the arts in the West are not a novelty. As an example, the Pythagoreans revered music as the language of the Universe, seeing it as a translation of mathematical principles. The Pythagorean approach to music, while mathematical, was deeply philosophical and spiritual, treating music as a bridge between the human and the divine. Yet, even then, this view of music stemmed from a bias toward a perhaps overly reverent view of mathematics. The mathematical approach to composition in that era, however, was fundamentally different from what we see today; it was used to lay the foundation and construct musical theory, not as a tool for creating artistic works. Centuries later, this notion of looking to music through the mathematical prism reappeared, by mixing with the modern era's inheritance of the human-machine concept, which emerged in the western 19th century. The idea that the human brain works like a complex numerical machine, and therefore that a numerical machine could replicate the behaviour of a brain, is the fundamental hypothesis beneath all the research that is being conducted in computer sciences for composition. As we live in a Western society in which this scientific only approach plays the role of an axiom, it is likely that even our most personal perceptions has been reshaped by this way of thinking, altering our relationship to music and art in general. In contrast to the Pythagorean era, by assuming that a machine can imitate the human mind's productions, today’s application of mathematics to music now seeks to generate music directly, bypassing the interpretive and creative processes that are central to artistic expression.
With the rise of Generative Artificial Intelligence and the statistical hypotheses it involves, the notion of creation as a logical-mathematical process has been strongly reinforced in Western artistic consumption habits, or at least within scientific research addressing artistic questions. AI-driven music generation tools have become increasingly popular, not just in experimental settings but in mainstream music production as well. However, Western musical language, which has been greatly enriched since the late 19th century by observing and studying non-Western cultural productions, and by the realization that artistic absolutes do not exist - especially after the two World Wars - has now been significantly impoverished by the algorithmic and statistical need to adhere to rules that can be encoded on a machine. This trend threatens to reverse the progress made in understanding and appreciating the diversity of musical expressions, reducing them to templates that fit within pre-defined, machine-readable formats, thereby narrowing the scope of what is considered Music.
The advent of Machine Learning revived the 19th-century concept of the artist-genius, but distorted it by considering such individuals as highly performant supercomputers. This reinterpretation equates human creativity with computational power, suggesting that creativity is merely a product of processing speed and data capacity. This flawed reasoning has led to the conclusion that supercomputers could be artist-geniuses, and by extension, artists. The idea that an artist's genius could be replicated by a machine overlooks the deeply human aspects of creativity-intuition, emotion, and cultural context. This reinterpretation was only made possible by researchers establishing mathematical criteria to evaluate artistic quality. These criteria often prioritize technical precision and statistical novelty over emotional depth and cultural relevance. By comparing algorithms to artists based on these mathematical criteria (with few alternative evaluation methods), the scientific community has become complacent in the belief that algorithms can increasingly approximate the output of human artists. This reasoning is tautological: if both artists and algorithms are assessed using criteria that are not particularly valid for determining artistic quality (if such a concept even exists-an assumption rarely questioned in scientific publications), it is unsurprising that algorithms perform comparably to their human counterparts. Evaluating algorithmic success by its popularity with the public is, in my opinion, an unreliable measure, given how these algorithms are heavily utilized in mainstream music production, skewing listeners' consumption habits and judgment criteria. The pervasive influence of AI-generated music on popular taste creates a feedback loop, where public preference is shaped by what the algorithms produce, rather than genuine artistic innovation.
I do not, however, claim that humans are unique in their creative abilities; I acknowledge that other systems might achieve similar results through entirely different means. Nevertheless, I am convinced that the scientific, logical, and mathematical approach increasingly promoted by contemporary computational developments is misguiding and more damaging than beneficial. The focus on efficiency and quantifiable results often leads to the neglect, or the suppression, of the more subtle, qualitative aspects of creativity that are essential to art. As is unfortunately too often the case in Machine Learning research, results that appear spectacular can blind both researchers and the public, who may overlook the importance of critically evaluating the value and quality of the content produced. The allure of technological advancement can obscure the need for a deeper, more nuanced understanding of what makes art truly valuable to humanity.
The outcome, in my opinion, is a significant regression, which disregards decades of artistic, philosophical, epistemological, and cultural inquiry by prioritizing models and hypotheses over observation and critical thought. This regression risks reducing art to mere data processing, where the richness of human experience and the diversity of artistic expression are sacrificed for the sake of computational convenience. The challenge lies in reclaiming the space for critical thought and creative freedom in an era increasingly dominated by algorithms and mathematical models.