JS Bach/AI Bach

How do you know that this essay was written by a human and not a computer? Well, I’d like to think that my writing style is un-robotic enough that you do know! But what about the autocorrect that my computer uses to help me with spelling? Maybe I’m even using Grammarly or some other program to assist my word choice and syntax. If I were would my computer be a co-author? 

In the art world we are very interested in the provenance of works. Scholars forensically examine handwriting, paper styles, watermarks in order to determine a piece’s proper attribution. And a cursory glance at musical economics will tell you that the attribution means much more than the work itself: a painting suddenly identified as a Leonardo jumps many millions in value, yet it remains the same painting. Blind studies of instruments have shown that the finest contemporary instruments are judged to be on par with Stradivarius and Guarneri violins, yet they are sold for a minuscule fraction of the cost. I remember how my heart sank when I learned that the “Haydn” D-Major Cello Concerto, which I loved passionately as a child (and still do), was likely composed by Anton Kraft and merely attributed to “Papa.”

Yet as much interest and intrigue have been invested in attributional research, there has rarely been any doubt that the provenance of artworks was human. Assuming that Hamlet was not, in fact, written by monkeys with typewriters, our museums and concert halls and theatre have been highly anthropocentric venues (excepting the brief fad of whale song and the third movement of Pines of Rome, with its nightingale recording). This may not be the case for much longer, and not because a new wave of octopus-literature is headed our way. Rather than animals, it is computers which are steadily eroding our human monopoly on intelligent creation. 

Someone could put together a lengthy documentary of confident-looking intellectuals chuckling and proclaiming: “Computers may be tremendous at long division, but they’ll never ever be able to do [x].” Those proclamations have tended not to age well. I spent a good portion of my teenage years obsessing over the ancient Asian board game Go, and the first line of my speech evangelizing for the game was always: “In chess they have created a computer that can defeat the best human, but Go simply doesn’t allow that. There are just too many possibilities.” I was able to enjoy that satisfaction for about ten years, until the program AlphaGo, developed by the company DeepMind, defeated a string of top professionals in 2016. Of course, there is plenty of footage of overly-optimistic (or pessimistic, depending on your viewpoint) AI researchers, as well, who since the 1950’s have predicted that human-like robots are just 20 years away. Those early prophecies, made in the initial rush of enthusiasm that accompanies frontier-opening discoveries, have proven naive: in particular, early AI researchers massively underestimated the difficulties of simulating basic human functions like vision and facial recognition. As cognitive psychologist Stephen Pinker has written, “The main lesson of thirty-five years of AI research is that the hard problems are easy and the easy problems are hard.” And yet unless you’ve been living underneath a rock, you’re probably aware that those intractable problems of vision and facial recognition have become, well, “tractable.” Fully-automated cars are driving our streets and Chinese security cameras are scanning citizens’ faces.

Music, and especially classical music, would seem an unlikely breeding ground for AI. Surely there is nothing more purely human, nothing more authentically expressive of our unique perch between animals and computers than music-making. A computer may be able to dominate in chess, or even Go, with its clear rules and objectives. But music requires the spark of individual expression, and that most human of qualities: creativity. With the glee of the zealous myth-buster, Prof. David Cope of Univ. of California Santa Cruz set out to prove those assumptions wrong by developing a composing computer. His program, Experiments in Musical Intelligence (EMI), first set out to master the Bachian Chorale, and after seven years of work was able to churn out 5,000 in a day. A performance of several of them in Santa Cruz was received very positively, that is until the identity of the composer was revealed—some in the audience yelled in anger. As EMI continued to develop it received a challenge from a skeptic: Univ. of Oregon Professor Steve Larson proposed a musical taste test, in which three pieces—by Bach, Larson, and EMI—would be performed anonymously, and the audience would vote on who composed what. The event certainly proved a point, though not the one Larson was hoping for—the EMI was judged to be by Bach, the Bach piece by Larson, and Larson’s piece by the computer.

Larson’s challenge was a variation on the Turing Test, named after its creator Alan Turing of “Imitation Game” fame. Turing predicted, way back in 1950, that machines would indeed develop the ability to think, and he created a method for testing it: the tester would send written questions to a machine and a human, respectively, and based on the responses would guess which is which. This test makes a presupposition about thinking: that its output is the sign of its existence. That is to say, if a machine yields the same output as a human mind, there is no reason not to call it intelligent.

This presupposition was challenged by philosopher John Searle in his famous riposte to Turing’s concept of intelligence: a thought experiment known as the Chinese Room. A man who knows no Chinese sits in a room and receives instructions on bits of paper slipped under the door, which tell him: “whenever you see [character] write down [character].” The instructions turn out to be an AI program for answering questions about stories in Chinese. The man’s output, which he then slips back under the door, seems to indicate that he knows Chinese. But he clearly doesn’t.

Philosophers could spend years spinning yarns of this ilk, and they have: “But what if the man is sitting in a vestibule and learning Quechua?” We don’t need to go down that rabbit hole, but the opposition of Turing and Searle raises a question with great relevance for music: do origin and intention matter? or just the output? When I think about the “Haydn” D-Major Cello Concerto, I chide myself for being disappointed that Anton Kraft is the possible composer: what should it matter who wrote it? The music remains just as beautiful! But when I think about the Bachian Chorales composed by David Cope’s algorithm, my gut tells me that their provenance does matter and should matter profoundly. They are like the uncomprehending scribbles of the man in the Chinese Room, assembled without intention, meaning, or spirit.

Cope might object, however, saying: “Wait a minute! Are you so sure that your vaunted human composition is so different? Sure, it feels when we compose as though we are expressing something original from within our authentic selves. But in fact we are merely assembling flotsam and jetsam from our memories into seemingly new arrangements, a sort of poetics of plagiarism.” There may be some truth to that. But surely composition flows from a deep-seated desire to represent and express, a desire that drove all artists from the cave-painters of Lascaux to today, and which is inextricable with consciousness, not merely intelligence. Following that logic, perhaps computers could indeed ingest terabytes upon terabytes of already-composed music and learn not only to imitate the styles of great composers brilliantly but also to synthesize elements of their styles in new and interesting ways. One could even go so far as to call this genuine creativity. But even with a nearly unlimited means of expression, will computers ever have something to say? 

It’s certainly not inconceivable that they will, even in the not-so-distant future. It’s very possible that computer-generated compositions will be indistinguishable from those of humans in their “sense of utterance.” Yet even if that is the case, I would argue that it shouldn’t matter. It shouldn’t matter, because output is not all that counts—provenance and intention matter, too. Sure, the work of Cope and his cohort has something important, perhaps even something profound to say about human creativity: theirs is a Copernican revolution of sorts, shaking our sense of our own centrality and specialness. And yet, people’s anger at discovering that the Bachian chorales they so enjoyed were written by an algorithm is not just some sort of logical fallacy, a bubble to be popped by rational myth-busters. The humanness of art isn’t just an arbitrary feature—it’s the essence of the thing itself. It’s humans figuring out what it means to be human. That isn’t to say that octopus art, or chimpanzee art, or AlphaGo art wouldn’t be of interest, but it would belong more in a cabinet of curiosities than in a museum. In chess, in presidential elections, in online commerce, the ends are largely unambiguous and independent of the means: checkmate, electoral victory, hits and sales. They may contain immense complexity, but these are all, at root, massive logic problems, problems which are amenable to being solved by algorithms: hence, Deep Blue, the data-driven micro-targeting of Cambridge Analytica, and the diabolical success of Amazon. The inherent difference of art may be overblown by some, but I would say that in art, the means are more inherently bound with the ends: when we see the paintings of Jackson Pollock, we feel his vigorous pouring and dripping and spattering onto the canvas, and that is part of the work’s meaning and impact. In encountering artworks, we see/hear/sense other humans coming to terms with their humanness and their being in the world, and that resonates with our own lived experience, whether consonantly or dissonantly. 

Some may argue that tools have always displaced creations further and further from the direct touch of the creator—an architect these days barely lays a hand on the buildings that rise according to her designs—and that this is simply the next stage in that process. The algorithms doing the composing/painting/rhyming were originally developed by human ingenuity, and so their fruits are therefore ours also. Yet surely there is somewhere a line between subservience and creative autonomy, even if it’s shrouded and difficult for us to locate: I doubt that many Brits in the late 18th century were taking credit for the Declaration of Independence. Tizian didn’t claim ownership over the works of Tintoretto, his rival, even though he had provided him training as a young pupil.

In any case, what relevance does all of this hemming and hawing about AI hold for symphony orchestras? Well, at a very basic level, any arts institution has a responsibility to express, critique, and change the world in which it lives, and so a development which has the potential to change the essential fabric of human life and even human nature should surely hold relevance for art-makers of all kinds. But more particularly—and this is the point I really wish to home in on—the arts may well begin serving as a last refuge of sorts, as human superiority over algorithms continues to dwindle. We remain confident in our status as masters, with algorithms serving as highly-capable servants. But it doesn’t require sci-fi visions of a malicious, conscious AI run amok to see our agency already being eroded. Our social interactions are highly structured by Facebook’s algorithms, our opinions and political preferences are powerfully influenced by algorithms feeding us micro-targeted content over social media, our purchasing decisions are heavily influenced by algorithm-generated recommendations, and even our biological organism is beginning to be regulated by the biometrics of FitBits and their like. Yuval Noah Harari has been perhaps the most prominent voice in recent years warning us of this algorithmic sapping of our agency: he often stresses that a world in which humans can be “hacked,” in which algorithms understand us better than we understand ourselves and can implant opinions and desires in us, will require an entirely new paradigm. The steady advance of algorithms will be difficult to reverse, in part because there are compelling arguments in favor of it: particularly in the realm of health, algorithms are already outstripping human doctors in diagnosis and prescription. There are certainly discussions to be had about the balance of benefits and dangers entailed by algorithms’ rise: some gush about the technological possibilities before us (some in Silicon Valley have begun openly talking about immortality as a legitimate goal), while others like me react with more Luddite instincts. But as I have argued above, music should be a bastion for humans. And it should be a venue for our agency and free will, our self-expression and authenticity, as fraught and questionable as those concepts may be. 

In that spirit, and zooming in on the particular case of symphony orchestras, I would suggest that there should be a concerted effort to nourish those human qualities of agency, creativity, and self-expression within our orchestras. Flesh and blood as the hand of the conductor may be, there is an undeniably algorithmic quality to the paradigm of “one guy stands on a box and tells others what to do.” The human beings sitting below with instruments are coded into a logic from above. I certainly don’t wish to oversimplify: I have known both the joys of belonging to the orchestra organism and the agency and creativity that orchestra musicians do, in fact, wield within certain boundaries. But few would claim that traditional orchestral work, in its current iteration, is fertile soil for the full blossoming of the creative individual. I have heard it referred to as the “golden handcuffs.” Certain renegade ensembles—Orpheus Chamber Orchestra and A Far Cry among them—have gone to the other extreme, discarding those handcuffs and abolishing the conductor. But surely there is enormous territory between anarchy and traditional orchestral totalitarianism that remains to be explored, territory that could perhaps yield a healthy balance between efficiency and broader individual flourishing. 

I would love to see orchestras take up the explorer’s astrolabe (not the AI of Google Maps) and begin to venture into that unknown territory. There are new ways of working, new structures of leadership and decision-making yet to be experimented with. And hopefully some of them will better equip orchestras to be a fortress of human agency and natural intelligence, as AI continues its “advance.”