Artificial Musical Intelligence and the Composer

I am cautious of but also excited for artificial musical intelligence, and not for reasons that might immediately spring to mind.

Is our increasing “belief in climate change” (a frequently championed phrase that makes little semantic sense) an opportunistic reaction to Christianity’s presently impoverished state? After all, climate change as a doctrine is predominantly Western, and the West’s most notorious cultural vandals, Foucault and Derrida, were no Christians. If modernism first challenged blind faith, postmodernism further persuaded those already infected by Nietzsche’s nihilism, and it was this predilection for, as Lyotard puts it, “incredulity towards metanarratives” that fuelled the atheistic passions of the late twentieth and early twenty-first centuries. These passions have not, in any rigid sense, endured.

Whilst there can be no doubt that Christopher Hitchens is still furiously pounding the pearly gates, lobbing fistfuls of quips at Saint Peter, there do seem to be fewer contemporary atheists of note; Richard Dawkins, if he was ever the movement’s nominal leader, now promotes himself as a “cultural Christian”. Rather, this faith of a kind, or the belief that there is no monotheistic god, has been superseded by a simpler, more ancient value proposition: that the soils of the earth are more meaningful than the possibilities of the sky. It is not Ouranos but Gaia who has awoken, leaving behind the coma dealt to her by the Industrial Revolution. And yet Andrew Forrest and Simon Holmes à Court are far more like James Watt and George Stephenson than they would make known. If Western capitalism did grow neglectful of the natural world, investors are now promised lucrative returns should they place their trust—their belief—in underdeveloped wind, solar and hydrogen technologies.

Like most things, it is not environmentalism that drives climate hysteria but pseudo-moral greed and, for some time, I have held the view that even the most underpaid of meteorologists should be required to declare any publicly traded securities they hold. And if such scrutiny is asked of them, it should definitely be asked of luminaries like Bill Weir, Cable News Network’s Chief Climate Correspondent; Justin Rowlatt, the British Broadcasting Corporation’s first-ever Climate Editor; Peter Prengaman, the Associated Press’s Global Climate and Environmental News Director; and even the Australian Broadcasting Corporation’s climate debonair, Michael Slezak. If these gentlemen are invested, their portfolios should be of serious interest. But I digress.

How ironic it is, then, that at the same time as this regressive yet commercial paganism, artificial intelligence (AI) threatens to, as metamodernists Timotheus Vermeulen and Robin van den Akker might be heard to say, catapult humanity towards a “futureless” future. Perhaps threatens is the wrong word. Whilst we must adopt an appropriately cautious stance towards AI, the field does imbue an innate excitement. What problems might this omniscient technology solve? And, by the same adrenaline-fuelled token, what problems might it create? In pursuit of answers to both questions, imagination collides with memory; I am transported back to 2008, to my local picture theatre, munching popcorn and taking in Andrew Stanton’s animated science-fiction film WALL-E. Incidentally, there is a great deal of digestible wisdom to be found in Pixar’s animated films of the early twenty-first century; one classic example is Brad Bird’s The Incredibles (2004), which, although libertarianism is not chief among my political philosophies, champions the arguments of Ayn Rand and includes one of cinema’s most prophetic lines: “When everyone’s super, no one will be.” (Not long after The Incredibles’ release, my primary school began awarding participation ribbons to all students who competed in its annual athletics carnival.) Four years later, WALL-E, infused with its Aristotelian flavours, offered another vital education to its viewers: that the purpose of technology must always be to aid human flourishing.

Stanton’s depiction of an unnatural humanity, crippled by lethargy and biological impairment, is a destination that AI cannot be allowed to journey us towards. Here, the irony of climate change only intensifies. Underdeveloped renewable technologies can do nothing to aid human flourishing. More, does man’s subconscious so fear the emergence of AI that he would manufacture a contemporary cult to worship Mother Earth? In any case, Western governments’ and corporations’ emphases on climate change provide for a poignant, if erroneous, counterpoint to our increasingly virtual lifestyles.

Within this emergent virtuality remain art and the creative industries, navigating the ever-stormy waters of our futureless future. Generative AI, specifically artificial musical intelligence (AMI), set sail many decades ago. Its voyage has led some to predict the eventual collapse of human music-making, or, at the very least, to adopt a pessimistic outlook. My position, as a composer of contemporary art music, is that I am cautious of but also excited for AMI, and not for reasons that might immediately spring to mind. Through considering existing compositional AMI systems—this article is chiefly confined to discussing compositional AMI systems, not those systems, though equally fascinating, associated with expressive performance or improvisation—I draw the hopeful conclusion that authentic human art music composition is, in fact, poised to flourish.

Contemporary art music composers have been using AMI systems for decades.

In 1957, composers Lejaren Hiller and Leonard Isaacson, after two years of experimentation with the University of Illinois’s ILLIAC I supercomputer, published Illiac Suite for string quartet. The work, regarded as the first serious advance in all-computer composition, was created through ILLIAC I’s random generation of musical parameters—pitches, rhythms, expressive techniques, and so on—and its retrospective organisation of those parameters according to rules-based problem-solving. “Our primary aim,” noted Hiller and Isaacson in Illiac Suite’s foreword, “is not [emphasis in original] the presentation of an aesthetic unity—a work of art. This music is meant to be a research record—a laboratory notebook.” Thus, even then the composers struck upon a foundational, quasi-paradoxical principle that has underpinned AMI to date: it is an artistic endeavour achieved only through scientific procedure, and a scientific procedure that strives towards artistic endeavour. This principle is examined throughout.

Since Hiller and Isaacson, a plethora of varying historical compositional AMI systems have followed. Systems of interest include Kemal Ebcioglu’s CHORAL (1993), which uses over 270 rules and heuristics to harmonise four-part chorales in the style of Bach; and SICIB (2001), designed by Roberto Morales-Manzanares et al, which composes music based upon choreographic gestures relayed as data through sensors attached to dancers. Another notable contribution is David Cope’s Experiments in Musical Intelligence (EMI) project, which Cope commenced in 1983 simply to, in his words, “have help finishing a commission”. EMI analyses discovered patterns, or “signatures”, across a composer’s catalogue to generate new music that emulates that composer’s unique style. Interestingly, Australasian Performing Rights Association and Australasian Mechanical Copyright Owners Society (APRA AMCOS) have voiced contemporary criticism of this survey-like methodology, concerned that generative AI systems have displayed a “complete lack of transparency” in acknowledging the existing, human-created work upon which they are modelled. Though this criticism seems real enough, how one goes about addressing it in our present age, in which digital accountability continues to degenerate, is not at all clear.

In a way, contemporary art music composers have been using, within composition software, expressive performance AMI systems for decades. Music notation programs like Avid’s Sibelius and MakeMusic’s Finale, as well as Steinberg’s competitive Dorico and the free, open-source MuseScore, all support playback functions, allowing typed scores to be synthetically realised through sampled sounds. For composers of seasoned aural skill, these realisations have proven quaint companions, occasionally yielding the odd insight. They can also provide colleagues and artistic planners with a tangible (if inaccurate) proof of concept. But playback functions have also provided a crutch for those of lesser aural skill, and, far from ennobling their musical technique, students of composition are deceived into the dangerous assumption that the sounds a computer can spit out are always achievable by a real musician. The truth is that Samuel Adler’s The Study of Orchestration (1982) remains far more instructive than a music notation program ever could be, just as a crisp, blank page offers limitless possibilities to the creative mind. The pen is mightier, not just than the sword but also the mouse.

To date, the vast majority of computational creative processes, such as compositional AMI systems, function according to predefined sets of rules, themselves often heuristic. The aforementioned principle, that AMI is at once artistic endeavour and scientific procedure, can now be more closely examined. The weight that should be ascribed to these, respectively, artistic and scientific characteristics is contested. John Searle’s Chinese-room argument—that while computers “manipulate formal symbols”, the meaning of those symbols “is totally beyond the scope of the computer”—suggests the scientific eclipses the artistic. As Margaret Boden writes in Artificial Intelligence and Natural Man (1987), extending one’s creative capacity depends upon the sustained accumulation of knowledge, or intelligence. Logically, such is equally a prerequisite for the machine. However, Geraint Wiggins asserts that Searle’s Chinese room has since been clarified as relating to machine consciousness, not machine intelligence, and that distinction between these two concepts is seldom accurately drawn. “The machine is musically intelligent,” asserts Wiggins of AMI, “but not musically conscious.” Conversely, although Ramón López de Mántaras concedes that machines “possibly will never have conscious thinking”, he argues that such a concession does not “deny the potential” for machine creativity and intelligence.

Accepting that the composer is more dependent on consciousness than intelligence in creating quality art, we may resolve that consciousness, not intelligence, is the greater contributor towards the aesthetic sensibility that underpins art music.

It seems to me that the question then becomes whether quality art, insofar as I have previously defined the term in these pages, can exist of pure intelligence without consciousness; moreover, as quality art is itself inanimate, this question must be withdrawn of the artwork and asked of the artist. I contend that, through its creator, quality art is, contextually, both logical and empirical, with the capacity to be both at once. For AI to achieve empiricism in its art, as human artists can, it must not only simulate human thinking and reasoning but develop an irrational, spontaneous approach to the human senses. Indeed, without the latter, the former cannot be transcended. Thus, to compose quality music, AMI requires consciousness more than it does intelligence. The human composer is bound to the same truth. Machine consciousness, such as imagined in Spielberg’s AI: Artificial Intelligence (2001), remains a fantastical proposition—though, as López de Mántaras implies, only for the present.

The late Richard Gill was often known to refute the notion that music is a language; I once heard him use this pithy phrase: “How do you say: ‘I want a cup of tea, please’ in music?” Wiggins agrees, describing music as, among its other qualities, both “anepistemic” and “autoanaphoric”; that is, almost always, music is “without denotational meaning” and “refers to itself”. What does, for instance, a bassoon denote when it sustains the pitch C4 for three seconds at a forte volume? The answer is that it denotes exactly that utterance, nothing more, nothing less. More, this utterance does not explicitly relate to any other utterance the bassoon has produced, whether similar, identical or starkly different. With no denotational meaning, it exists—and, thus, means—entirely in the moment. Therefore, as Wiggins argues, “aesthetic sensibility is almost everything” in anepistemic and autoanaphoric music “because … there is almost nothing else”.

In other terms, or in terms that Quadrant Music readers are more familiar with, a piece of music’s craft and style—its aesthetic sensibility—are its musical substance. Its musical substance is not to be found, for example, in its extra-musical themes. This is even true of music which overtly and successfully employs such themes. Ultimately, the tale told in Till Eulenspiegel is not realised by the work’s narrative plot but, rather, its musical parameters, like the staccato quavers and augmented seconds of its famous horn solo. And those staccato quavers, ultimately, denote no more than, and certainly do not refer to, the solo staccato quavers found in, for instance, any Mozart horn concerto, or even the staccato quavers assigned to the horn by Strauss in his other works.

Herein lies the rub. Accepting that the composer is more dependent on consciousness than intelligence in creating quality art, we may resolve that consciousness, not intelligence, is the greater contributor towards the aesthetic sensibility that underpins art music, which is largely anepistemic and autoanaphoric. Further, I contend that the more denotational and referential a piece of music, the more it may be explained and replicated not by consciousness but intelligence. If this is true, contemporary art music composers should pause and reflect. A composer whose music is of quality and depth, demanding consciousness and not just intelligence, is far more likely to successfully compete with intelligent yet unconscious computational creativity. Indeed, the AMI tsunami, if it is coming, will strike first at the musical foreshore. Popular music inclusive of lyrics is, through its employ of language, the more denotational and referential creation and, thus, a creation more easily replicated through intelligence alone. Fundamentally, the singer-songwriter must answer the same call for artistic authenticity as the symphonist—though with greater haste. Contemporary composers, of any musical genre, must insist their work becomes more crafted and more stylistic, and they must strive towards these ambitions with not just expanding intelligence but intensifying consciousness.

Even more philosophically, as a brief interlude, let us accept the validity of Leonard Bernstein and Stefan Bauer-Mengelberg’s “infinite variety of music” formula. (Simply put, “How long is a piece of string?” And so it is with a piece of music.) This being the case, there exists for humanity to explore a limitless universe of musical creativity. AMI can accelerate our exploration of this universe, certainly. But, if it is truly infinite, which Bernstein and Bauer-Mengelberg, using mathematics, resolve that it is, then AMI can never completely explore it. That is not to admit it cannot and would not try, one galaxy of possibility at a time. Rather, humanity must decide whether the journey of exploration is one worth accelerating through. The quandary is artistic, not scientific; whereas there exists an imperative in medicine to progress as hastily as possible towards cures for rare forms of cancer, does the same imperative exist in music to, for example, discover new moods? Perhaps, subconsciously and because of consciousness, we have already acted upon our answer to this quandary. The innovation of compositional AMI systems, specifically those concerned with generating art music, does seem to have stalled in recent years; the most significant programs, like EMI and CHORAL, find their geneses in the late twentieth century. All in all, in the words of Bax: “These are deep matters.”

Wiggins also draws attention to the present inaccessibility of compositional AMI systems, writing that “those wishing to practice AMI must become at least knowledgeable about, if not expert in, music perception and cognition”. To use now-familiar terms, the contemporary art music composer is ever-striving to develop musical intelligence and human consciousness. These pursuits are herculean in themselves. Almost certainly, a correlation exists, and will continue to exist, between the complexity of compositional AMI systems and their popularity among human practitioners.

It is a moral question, not a scientific one, as to whether human listeners should assign value, of any degree, to the art produced by intelligent, unconscious machines. Of course we could, but why would we, especially as such value has been hitherto reserved, with significant success, for human art?

One of the great joys of having the good fortune to serve as this magazine’s Music Editor is that I am ever refining my ideas; this column, through its regularity, offers a kind of demonstrable progress of my many expeditions. For instance, I wrote an article on the relationship between art and AI for the Epoch Times in May last year. In it, I presented arguments that were too heavy-handed. The article’s final sentences read:

It is my belief that AI constitutes a legitimate threat to authentic artists, as well as societies that rightly prize art as vital—if not now, then certainly in the not-so-distant future. The solution seems to me to be a kind of global stance: artists everywhere must first identify AI-generated art and, second, reject it and its beneficiaries.

As proffered earlier, threat, it really does seem, is not the right word. Without consciousness, the compositional AMI system cannot, in my view, truly rival the creative potential of the human art music composer. Wiggins extends this view, arguing that “it simply does not matter whether a piece of music was written by a human or a computer”. In his estimation, “what matters is whether the music succeeds as a piece of music, in its own right”. This, he concludes, “does necessarily depend on human listeners”. Although he offers the point critically, López de Mántaras states humanity’s “social rejection” of computational creativity is based not upon science but morality. All the arguments produced in this paper would seem to point towards this being the case. It is a moral question, not a scientific one, as to whether human listeners should assign value, of any degree, to the art produced by intelligent, unconscious machines. Of course we could, but why would we, especially as such value has been hitherto reserved, with significant success, for human art? More, it is a moral question as to whether human listeners should accelerate their exploration of the infinite musical universe when there, seemingly, exists no imperative to do so. And, if, as Wiggins argues, AMI systems have, since their inception, failed “to consider the function of listening”, it is a moral question as to whether the gift of listening, itself requiring both intelligence and consciousness, should be bestowed upon and demanded of machines. Furthermore, even if humanity resolves to act affirmatively upon all these matters—to value the artificial, to accelerate the exploration, and to bestow and demand listening—I contend that such resolutions would occur not distinct from but in relation to previous human knowledge, attitudes and experiences. It may be that, one day, a machine generates—or, more accurately, regenerates—a cycle as colossal as Der Ring, or a symphony with a climax as powerful as the Resurrection. But if the purpose of art is to express the human condition, in all its complexity and uncertainty, then do these works derive an additional, innate strength simply because, respectively, Wagner and Mahler, humans, created them? I think they do.

With some added clarity, then, a final comment can be made concerning the principle that AMI is simultaneously an artistic endeavour and a scientific procedure. Whether artificial or organic, the artistic endeavour cannot be reduced to mere scientific procedure. Rather, in order to succeed as human composers have, compositional AMI systems must develop not only consciousness but humanity. Without humanity, AI is not an honest representation of the human experience. And that AI should be required to develop humanity would not only seem to be oxymoronic but, in any human sense, counterproductive.

Based upon these realisations, autonomous computational creativity, such as compositional AMI systems that achieve total independence, would not seem to threaten the craft of human artists to any existential extent. Humanity likely will not “reject” AI-generated art but, rather, contextualise it as a subsidiary of human creativity. And my view is that organic authenticity will be the more prized artform. It is the organic that has capacity for both intelligence and consciousness, and it is through these symbiotic qualities that there exists the potential for authentic expression, which relies upon the creator’s knowledge (that is, intelligence and logic) and honesty (consciousness and morality).

Whilst the emergence of AI challenges authentic artists to become even more so, it, conversely, creates yet another advantage for inauthentic artists. Semi-autonomous compositional AMI systems are susceptible to exploitation by human composers wishing to deceive others of their natural musical abilities. Such deceptive practices already occur. Composers of lesser aural skill are reliant upon the playback functions of music notation programs. Additionally, as Quadrant Music has previously addressed, some concert hall art music composers, unbeknownst to listeners, employ orchestrators to complete their scores. It is not impossible to imagine that this charlatanism could become amplified through computational creativity. By the same token, those profiting from the development and sale of AMI systems, just like those profiting from the development and sale of renewable technologies, must be closely scrutinised. Private experimentation with innovative software is one thing; using technology to achieve widescale commercial dominance is another. But all in all, let these concerns not sustain too much worry. The trajectory of Western art music—and, indeed, the West itself—has always been that of improvement, optimism and hope. Just as any latest technological feat follows in the wake of some previous advancement, human flourishing, including artistic and moral flourishing, is built upon the successes and struggles of the past. Generative AI, I really believe, cannot dissuade human creativity. It can only compel us towards new and greater heights, towards a futureless, human future.

In this issue of Quadrant Music, Dr Bradley Voltz shines a light on contemporary efforts to revive the noble yet ignored stradella piano accordion. They say that a gentleman is one who can play the accordion but chooses not to. My own experiences in writing and arranging for the instrument, however, have proved profoundly rewarding. R.J. Stove also returns to these pages, this time to explore, upon the fiftieth anniversary of his death, the life and work of French composer and Les Six member Darius Milhaud. Contributions to Quadrant Music, as ever, are encouraged.

Alexander Voltz is a composer and the founding editor of Quadrant Music.

Contribute to Quadrant Music: [email protected].