March 21, 2020
Home  »  Magazine  »  Arts & Entertainment  » Essays  » AI In Music »  Sounds Like Bach

Sounds Like Bach

But the day when music is finally and irrevocably reduced to syntactic pattern and pattern alone will be, to my old-fashioned way of looking at things, a very dark day indeed.

Sounds Like Bach
illustration by Jayachandran
Sounds Like Bach
Back when I was young -- when I wrote "Gödel, Escher, Bach" -- I asked myself the question "Will a computer program ever write beautiful music?", and then proceeded to speculate as follows: "There will be no new kinds of beauty turned up for a long time by computer music-composing programs... To think -- and I have heard this suggested -- that we might soon be able to command a preprogrammed mass-produced mail-order twenty-dollar desk-model 'music box' to bring forth from its sterile circuitry pieces which Chopin or Bach might have written had they lived longer is a grotesque and shameful misestimation of the depth of the human spirit." I went on and on in this vein.[1]

What do I make of such speculations now, a quarter-century later? I am not quite sure. I have been grappling for several years now with these issues, and still there is no clear resolution.

In the spring of 1995, I came across the book "Computers and Musical Style" by David Cope, a professor of music at the University of California at Santa Cruz, and in its pages I noticed a mazurka supposedly in the style of Chopin, written by Cope's computer program EMI (short for "Experiments in Musical Intelligence"). This intrigued me because, having revered Chopin my entire life, I felt certain that no one could pull the wool over my eyes in this department. So I went straight to my piano and sight-read through the EMI mazurka several times, with mounting confusion and surprise.

Though I felt there were a few little glitches here and there, I was impressed, for the piece seemed to "express" something. Had I been told it had been written by a human, I would have had no doubts about its expressiveness. It sounded slightly nostalgic, had a bit of Polish feeling in it, and it did not seem in any way plagiarized. It was new, it was unmistakably "Chopin-like" in spirit, and it did not feel emotionally empty. I was truly shaken. How could emotional music be coming out of a program that had never heard a note, never lived a moment of life, never had any emotions whatsoever?

The more I grappled with this, the more disturbed I became -- but also fascinated. There was a counterintuitive paradox here, something that obviously had caught me enormously off guard, and it was not my style to merely deny it and denounce EMI as "trivial" or "nonmusical". To do so would have been cowardly and dishonest. I was going to face this paradox straight on; I was going to grapple with this strange program that was threatening to upset the apple cart that held many of my oldest and most deeply cherished beliefs about the sacredness of music, about music being the ultimate inner sanctum of the human spirit, the last thing that would tumble in AI's headlong rush towards thought, insight, and creativity.

Had I only read about EMI's architecture and not heard any of its output, I would have paid little or no attention to it. Although Cope has put in far more work on EMI than most AI researchers ever do on any one project, its basic principles simply did not sound radically new to me, or even all that promising. What made all the difference in the world for me was carefully listening to EMI's compositions.

Over the next few months, I lectured about EMI in many places around the United States and Canada, and what I found truly surprising was that hardly anyone in my audiences seemed upset at Cope's coup in the modeling of artistic creativity; hardly anyone seemed threatened or worried at all. I, on the other hand, felt that something of the profundity of the human mind's sublimity was being taken away. It seemed somehow humiliating, even nightmarish, to me.

The deepest underlying principle behind EMI is what Cope terms "recombinant music" -- the identification of recurrent structures of various sorts in a composer's output, and the reusing of those structures in new arrangements, so as to construct a new piece "in the same style". One can thus imagine feeding in Beethoven's nine symphonies, and EMI coming out with Beethoven's Tenth.

EMI's central modus operandi, given a set of input pieces, is:

(1) chop up; (2) reassemble.

There are, of course, significant principles constraining what can be tacked onto what, and these principles are formulated so as to guarantee coherence. I summarize these two principles as follows:

(1) Make the local flow-pattern of each voice similar to that in source pieces;
(2) Make the global positioning of fragments similar to that in source pieces.

These could be likened to two types of constraints that a jigsaw-puzzle solver naturally exploits when putting together a jigsaw puzzle:

(1) The shape of each piece meshes tightly with those of neighboring pieces;
(2) The stuff shown on each piece makes sense in the context of the picture.

The former of these constraints might be characterized as "syntactic meshing", or meshing based solely on "form", while the latter could be characterized as "semantic meshing", or meshing based solely on "content". In isolation, perhaps neither of them would be too impressive, but when used together, they form a powerful pair of constraints.

Lack of space unfortunately prevents me from describing here the many types of intricate mechanisms by which EMI picks up stylistic characteristics and carries out the many-leveled "recombination" that Cope has programmed.

In my lectures on EMI, I nearly always let my audience hear a handful of small two-voice pieces for the audience. The listeners are forewarned that there is at least one piece by Bach in the group, and at least one by EMI in the style of Bach, and they should try to figure out which ones are by whom (or by what). After the pieces have been performed, I ask the audience to vote. Usually, most of the audience picks the genuine Bach as genuine, but usually it is only about a 2/3 majority, with roughly 1/3 getting it wrong. And it is not by any means always the less sophisticated audience members who make the wrong classification.

EMI is evolving -- it is a moving target. Cope began work on his program in 1981, and in all these years he has not let up on it. EMI's early pieces are, like any fledgling composer's, pretty amateurish affairs, but its later output sounds increasingly impressive, and Cope has grown more and more ambitious over time. Whereas initially he was proud of EMI's production of short two-part inventions and short mazurkas, he now has EMI producing entire sonatas, concertos, and symphonies. There is even a "Mahler opera" under way now -- something that would certainly be a challenge for any human composer to carry off.

Style, of course, is a multi-layered phenomenon. There are shallow and deep aspects of style. It is quite possible that someone could be capable of capturing many of the shallower trademarks of a composer and yet miss the bull's-eye as far as essence is concerned.And so, how much are we being fooled when, on hearing a piece of music, we respond to some gestures that in the past we have come to associate with composer X, and then exclaim to ourselves, "This piece sounds like X"? Can we even distinguish clearly between responses at a shallow level and a deep level? Indeed, what is the difference, in music, between "shallow" levels and "deep" levels of style, between "syntax" and "semantics", between "musical form" and "musical content"? Is there really any such difference at all?

In my lectures, I usually have a second musical interlude, this time involving mazurkas -- one by Chopin and one by EMI. One time, when I gave this lecture at the world-famous Eastman School of Music in Rochester, New York, nearly all the composition and music-theory faculty was fooled by the EMI mazurka, taking it for genuine Chopin (and the genuine Chopin piece, by contrast, for a computer-manufactured ditty). An Eastman music student, Kala Pierson, wrote me an email about this event in which she said, "I voted real-Chopin for the second piece, as did most of my friends. When you announced that the first was Chopin and the second was EMI, there was a collective gasp and an aftermath of what I can only describe as delighted horror. I've never seen so many theorists and composers shocked out of their smug complacency in one fell swoop [myself included]! It was truly a thing of beauty."

One stunning lesson from my Rochester lecture (and indeed, from all of the times I've lectured on EMI) is that people with deep musical gifts and decades of training can, on occasion, mistake an EMI product for "the genuine article". And remember -- we are just embarking, we humans, on the pathway toward the realization of the dream of "preprogrammed mass-produced mail-order twenty-dollar desk-model music boxes" -- those boxes on whose "sterile circuitry" I heaped so much scorn, back when I wrote GEB.

Where will we have gotten in twenty more years of hard work? In fifty? What will be the state of the art in 2084? Who, if anyone, will still be able to tell "the right stuff" from rubbish? Who will know, who will care, who will loudly protest that the last (though tiniest) circle at the center of the style-target has still not been reached (and may never be reached)? What will such nitpicky details matter, when new Bach and Chopin masterpieces applauded by all come gushing out of silicon circuitry at a rate faster than H2O pours over the edge of Niagara? Will that wondrous new golden age of music not be "truly a thing of beauty"?

Consider "Prokofiev's tenth sonata", as Cope calls it. In the liner notes to EMI's first compact disk, called "Bach by Design", Cope wrote, "This computer-composed Prokofiev sonata was completed in 1989. Its composition was inspired by Prokofiev's own attempt to compose his tenth piano sonata, an attempt thwarted by his death. As such it represents another of the many potential uses of programs such as EMI (i.e., the completion of unfinished works)." To me, this remark comes close to blasphemy.

What worries me about computer simulations is not the idea that we ourselves might be machines; I have long been convinced of the truth of that. What troubles me is the notion that things that touch me at my deepest core -- pieces of music most of all, which I have always taken as direct soul-to-soul messages -- might be effectively produced by mechanisms thousands if not millions of times simpler than the intricate biological machinery that gives rise to a human soul. This prospect, rendered most vivid and perhaps even near-seeming by the development of EMI, worries me enormously, and in my more gloomy moods, I have articulated three causes for pessimism:

(1) Chopin (for example) is a lot shallower than I had ever thought.
(2) Music is a lot shallower than I had ever thought.
(3) The human soul/mind is a lot shallower than I had ever thought.

Let me briefly comment on these. Pertaining to (1), since I have been moved to the core for my entire life by pieces by Chopin, if it turns out that EMI can churn out piece after piece that "speaks like Chopin" to me, then I would be thereby forced to retrospectively reassess all the meaning that I have been convinced of having detected in Chopin's music, because I could no longer have faith that it could only have come from a deep human source. I would have to accept the fact that Frédéric Chopin might have been merely a tremendously fluent artisan rather than the deeply feeling artist whose heart and soul I'd been sure I knew ever since I was a child.

That loss would be an inconceivable source of grief to me. In a sense, the loss just described would not be worse than the loss
incurred by (2), since Chopin has always symbolized the power of music as a whole, to me. Nonetheless, I suppose that having to toss all great composers out the window is somehow a bit more troubling than having to toss just one of them out.
The loss described in (3), of course, would be the ultimate affront to human dignity. It would be the realization that all of the "computing power" that resides in a human brain's 100 billion neurons and its roughly ten quadrillion synaptic connections can be bypassed with a handful of state-of-the-art chips, and that all that is needed to produce the most powerful artistic outbursts of all time (and many more of equal power, if not greater) is a nanoscopic fraction thereof -- and that it can all be accomplished, thank you very much, by an entity that knows nothing of knowing, seeing, hearing, tasting, living, dying, struggling, suffering, aging, yearning, singing, dancing, fighting, kissing, hoping, fearing, winning, losing, crying, laughing, loving, longing, or caring.

Although Kala Pierson and many others may hail its coming as "truly a thing of beauty", the day when music is finally and irrevocably reduced to syntactic pattern and pattern alone will be, to my old-fashioned way of looking at things, a very dark day indeed.

Douglas Hofstadter is, inter alia, director of the Center for Research on Concepts and Cognition, Indiana University, Bloomington. His books include the Pulitzer winning Goedel, Escher, Bach, Metamagical Themas, Le Ton beau de Marot, translation of Pushkin's novel in verse Eugene Onegin and The Mind's I (with Daniel C.Dennett).

1. Editor's Note: Please see pp 676-677 of Godel, Escher, Bach

Next Story >>
Google + Linkedin Whatsapp

The Latest Issue

Outlook Videos