But is it Mozart?

A computer in California has just written Mozart's 42nd Symphony, a Chopin mazurka and over 5,000 works of its own.
Click to follow
The Independent Culture
The image on the cover of the New Scientist for 9 August was lurid: a head with a great cello-shaped bulge on one side of its brain and bars of (upside down) music protruding from the other. The caption read: "Watch out. This machine could steal your soul." Admittedly, these offered a somewhat sensationalised gloss on the actual response of the eminent cognitive scientist Douglas Hofstadter (author of the once-cult brain-teaser Godel, Escher, Bach) to the news that a computer has recently generated what its programmer is pleased to call Mozart's 42nd Symphony. Concerned that a machine which "has no model whatsoever of life experiences, has no sense of itself, has never heard a note of music" could imitate the creative process of a great composer to a degree that allegedly convinced at least some of the experts some of the time, Hofstadter wondered whether this meant that the composer's soul is irrelevant to the music, because "if that's the case - and I'm not saying it is - then I've been fooled by music all my life. I've been sucked in by a vast illusion. And that would be for me an absolute tragedy..."

The source of this perturbation turns out to be the 56-year-old American composer and academic David Cope, whose student orchestra in the University of California at Santa Cruz premiered the "new" Mozart symphony last April. In fact, Cope began to develop his Experiments in Musical Intelligence programme - or (rather confusingly for British readers)EMI for short) - some 15 years ago, not to fabricate Mozart but to help himself over a spot of composer's block. By analytically breaking down and programming the gamut of procedures in his previous works, Cope hoped he could generate a choice of logical continuations at any point where he got stuck. In a way, Cope's initial motivation hints at what may be the fundamental limitation of artificial intelligence as applied to musical composition to date: its circularity. For other composers, by contrast, might well consider that such blocks constitute an invitation, even a sanction, to follow through with something new, unexpected, or positively illogical. As Stravinsky, who seems to have foreseen the whole issue three decades ago, once remarked: "I expect in my own case that when the computer has quantified my musical characteristics, I shall try to do something different."

Actually, EMI represents only the latest stage in what, by now, is quite an extended history of efforts to simulate music by machines, ranging from such current programmes as Brian Eno's Koan for generating ambient music, back via the methods of mathematical patterning promoted between the wars by Joseph Schillinger (to the profit of sundry movie composers and even George Gershwin) all the way, as Cope acknowledges, to the musical dice game attributed to Mozart for composing endless waltzes "without the least knowledge of music". Nor should these recurrent attempts to construct such labour-saving mechanisms surprise any save the most incorrigibly romantic. For Western composition has traditionally rested upon teachable - and hence potentially programmable - skills, supplemented, or supplanted in our own time by such constructivist systems as serialism or the musical exploration of probability and chance. One might indeed hazard the assertion that up to 90 per cent of the musical fabric of some of the greatest masterpieces could have been generated by some putative programme drawn from the composer's previous works and the prevailing compositional practices of his time. The real challenge to the advance of artificial intelligence in music presumably lies in those 10 per cent of idiosyncratic strokes and procedures, complexities of affect, and so on that distinguish a Mozart from a Salieri, and which duly evoke from performers, and for listeners, the soul that Professor Hofstadter so hankers for.

The matter is rather complicated by a phenomenon long familiar to music schools and university faculties: the periodic appearance of students with seemingly faultless talents for pastiche in the style of Bach, Mozart, or whoever. To suggest that such assimilative abilities often correlate rather precisely with the absence of strong individual creativity might seem tendentious; but it is difficult to think of a composer of any standing who, in attempting to write in the style of another, has been able to resist doing something of his or her own to it. It might have been easier to determine whether Cope's programme transcends such powers of pastiche, penetrating that last 10 per cent, were his Mozart 42nd out on disc: attempts earlier this week by a succession of friendly computer experts to extract the bits that are apparently lodged on the Internet proved fruitless. What leads one to suspect that he has not got that far is his frank admission in the New Scientist that he cheats. For while his programme is evidently capable of throwing up any number of generic classical symphonies and then Mozartifying them through an input of sampled personal characteristics, Cope still relies upon his own composerly intuition to spot the likeliest version amid the incessant turn-over of hit-or-miss permutations, and to give it the finishing tweak. As he confesses: "It would horrify me personally to walk into a hall and hear someone else's EMI Mozart symphony and to hear it was as good as the one I produced."

Maybe that is why, for all EMI's simulations of other composers - including the production of a Chopin mazurka which has almost convinced even Hofstadter - and Cope's declared ambition to try, no less, for a Mahler's 11th, he has so far avoided what one might have thought would offer the most stringent, quantifiable type of test of his programme: the completion on computer of the sketches for Mahler's 10th to set beside the human completion prepared with such musicianly care by Deryck Cooke, with the assistance of the late Berthold Goldschmidt and the Matthews brothers, David and Colin, or an EMI Elgar's Third to compare with the forthcoming completion by Anthony Payne; or a Mozart Requiem to contrast with the many attempted fillings-out since Sussmayr - or, beyond that, any of the dozens of other works for which Mozart left only promising beginnings.

In fact, as quoted in a lucid discussion of EMI by Steven Poole in last Friday's Guardian, Cope seems to accept that his programme may ultimately benefit the present more than the past: "Since I've finished EMI, I have composed more works than I ever thought possible. Having an EMI around will, I hope, inspire more human composition, not less." Yet sociologically, this would surely compound the problem already raised by the computer- cloning of past masters. For if the manufacture of "new" works by Mozart seems questionable when there is already enough of the real thing to keep most performers and listeners happy for life, the multiplying of today's already multifarious compositional activity in the face of a music industry and a public that seem less and less inclined to absorb the bulk of it will likely lead to some sort of crisis - at least for composers.

Which is not to say that if one were attempting to compose, say, a complex canon, involving many redraftings, much pulling and pushing of material to find the best fit, one might not welcome a programme that instantly presented all the possible solutions. Except that this could well raise the question of the extent to which the ultimate character of a piece of music is dependent upon the very gropings and doubts, the time-consuming process of trial and error that went into its making - a character that might well be lost were the salient choices too quickly presented, too facilely made. But whatever the future role of artificial intelligence in the art of musical composition, its present developments at least raise the most searching questions about the nature of creativity, the traditional notions of personal style and the unified work, about function, reception and not least such dread philosophical conundrums as intentionality, which are quite enough for mere composers to be getting on withn

David Cope's Web page is: http://arts.ucsc.edu/faculty/cope/home