16.0 Music AI in context#
Before we dive into the technical ideas, it is worth stepping back to place music AI in context.
Music and technology co-evolve#
Music is an artform of change over time. So too does the technology that enables it.
When confronted with recent staggering developments in music AI, it is important to remember that music and technology have always co-evolved. As we discussed in Chapter 0, each new technological paradigm, from instruments to recording, electronics, and computing, has offered new creative possibilities to musicians and also challenged existing norms.
Many of these technologies were initially viewed as threatening. For example, even synthesizers were viewed with suspicion initially. The rock band Queen famously printed a disclaimer on several of its albums in the 1970s:
Fig. 131 The liner notes of Queen’s A Night at the Opera (1975), which insist that the album’s many layered sounds were made without synthesizers.#
Perhaps the most profound example is recorded music itself. In his 1906 essay “The Menace of Mechanical Music” [Sou06], the composer John Philip Sousa warned that recordings and other mechanical reproductions of music might make musicianship obsolete. Recording did dramatically change the economics of music, and it also offered tremendous new creative potential, from the studio as an instrument to sampling (Chapter 11).
AI may be the most powerful technology that humans have created, and its effects on music may accordingly be profound. Like recording technology before it, AI will almost certainly have a marked effect on the labor and economics of music. It remains to be seen how it will affect music creatively and culturally.
Music AI is multifaceted#
Another important consideration is that “music AI” means more than just generating music from text prompts.
Music is multifaceted, and encompasses far more than the recordings that we consume. It includes composing, performing, practicing, teaching, producing, and listening, among many other activities.
Similarly, “music AI” means more than generating music from short text descriptions. It encompasses a wide range of other emerging (and potentially transformative) capabilities, such as understanding the contents of music, integrating with the creative workflows of musicians, and interacting with musicians in real time. Many of these capabilities aim to assist the creative process of music, not just the consumption of music recordings.
Scope of this chapter#
This chapter is mostly a cursory overview. We will cover a few of the basic technical ideas of machine learning, and offer exposure to several research directions in music AI.
We will not go in depth into machine learning algorithms or modern AI systems. For a thorough treatment of modern machine learning, see Kevin Murphy’s Probabilistic Machine Learning [Mur22]. For a deeper dive into music information retrieval, see Meinard Müller’s Fundamentals of Music Processing [Mul21].
Unfortunately, there isn’t yet an authoritative text on modern music AI, primarily because it is still a rapidly developing field. Instead, the end of this chapter provides a list of research papers for those who wish to dive deeper.