Archives - 1990

Archives Main | 1990

 

SOURCE BOOK

Spring 1990
ROBO BEETHOVEN

In the spring of 1957 a couple of Bell Telephone Laboratories scientists, Max V. Mathews and John R. Pierce, went to a piano recital at a nearby New Jersey college. The music-lovers were unimpressed by parts of the performance that night, and as they left, one said to the other, half-jokingly, "The computer can do better than this."
Back in the labs Mathews decided to test that casual remark. He wrote a computer program to translate simple instructions into a sequence of computer-generated binary numbers that represented the sound waves of musical tones. Then Mathews enlisted a colleague, Newman Guttman, to compose a tune to be played with the tones the computer produced. In May of that year the first-ever piece of music performed by a computer, "In the Silver Scale," was played at Bell Labs.
Mathews had turned a computer into a musical instrument, but that wasn't the only way of making music with computers. The year before at the University of Illinois, Lejaren Hiller Jr. and Leonard Isaacson wrote a program to compose sequences of musical notes, which they then combined into a string quartet score called "Illiac Suite." Both of those early experiments were pretty rough, but they represent important steps in the evolution of electronic music.
A Russian named Leon Theremin made some of the earliest such music in 1919 (see the sidebar on page 55). Music from machines became more sophisticated when acoustic research made advances in analyzing the wave forms of every instrument, and electronics research made it possible to reproduce them. In 1963 Robert Moog teamed up with the experimental composer Herbert Deutsch to investigate voltage-controlled electronic-music modules. The resulting Moog synthesizers, first sold in 1965, were widely used by rock bands, and even classical composers took note of the new sounds made possible.
As the early examples show, a human being can certainly put his fingers on a computer keyboard and make recognizable music come out. But can a computer ever compose music? The question has two possible answers: "yes" and "maybe" -the "maybe" depending on what you are pleased to define as music.
For instance, the school of aleatory music has a history that includes practitioners as far back as Wolfgang Amadeus Mozart and as modern as John Cage. Aleatory, or found, music involves random composition, and computers are great at random operations. But Mozart didn't need a computer when he invented a musical dice game called Musikalische Wurfelspiel for the amusement of those who wished to compose easy pieces. A roll of the dice provided the random results.
Many possible algorithms can be used for aleatoric composition by a computer. But to make the music "creative," or at least unpredictable, you need to use what mathematicians call the Monte Carlo method, which generates random numbers. When John Cage composed his "Imaginary Landscapes No. 5," he took sounds from 42 phonograph records and combined them in an order directed by a chance operation.

 

 

In the 1960s scientists and composers at Bell Labs followed the early inspiration of Pierce and Mathews and produced a record, 'Music From Mathematics'. Their premise, according to the album's liner notes, was that "any sound can be described mathematically by a sequence of numbers," and the selections demonstrated various ways of doing so. On one track, a computer-produced voice sings "A Bicycle Built for Two" (a performance many of us heard reproduced in the film 2001: A Space Odyssey, when the ditty is sung by the "dying" computer, HAL 9000). Three of the album's composers let the computer itself generate the music. "The notes and rests, of various lengths, were selected by chance," says Pierce of his "Five Against Seven-Random Canon." After the first measure, he continues, "the subsequent measures are, by chance, either new music or repetitions of earlier bars."
Found numbers have been used frequently in compositions. As recently as August 1989 Michael Lee Thomas composed such a piece when he took the telemetry signals of dust particles striking the spacecraft Voyager II during its flyby of the planet Neptune and enhanced the sounds by computer to produce his "Voyager-Grand Tour Suite."
Those examples support the "yes" answer to the original question. Computers can generate a kind of stochastic or aleatory music, though usually they need a lot of help from their human operators. But can computers compose in the sense that human beings do, by following rules of counterpoint and harmony? Maybe.

COMPUTERS CAN UNDOUBTEDLY learn such rules and follow them. For instance, one program written in LISP (short for list processing) constructs a counterpoint to a computer-generated cantus firmus (or melody) according to the classic rules of contrapuntal composition. Each rule is a function of the program which rejects any note that violates a rule.
But computers are capable of even more. They can also be programmed to compose in a particular style. At Stockholm's Royal Institute of Technology in 1976 Bjorn Llndblom and Johan Sundberg analyzed the structure of Swedish folk tunes, and by programming rules into a computer, they produced new but authentic-sounding Swedish folk music. Similar rules can be constructed for almost any kind of music and its composer.
But some experts in the field-such as Marvin Minsky, a skilled pianist and the presiding genius over the Artificial Intelligence Laboratory at the Massachusetts Institute of Technology-think that analyzing the brilliance behind the music has serious limitations. "Music theory has gotten stuck by trying too long to find universals," says Minsky. "We would like to study Mozart's music the way scientists analyze the spectrum of a distant star. But we must view these findings with suspicion, for they might show no more than what composers then felt should be universal. If so, the search for truth in art becomes a travesty in which each era's practice only parodies its predecessor's practice."
Nevertheless, composers' styles can be analyzed and reproduced. Would musicologists listening to these synthetic compositions recognize them as at least in the style of, for instance, Beethoven? Probably so, and they would go on to call them bad Beethoven - pieces attempted on some of his worst clays and then abandoned.
But that isn't really the point. To be sure, objectively measuring the "best" music, or even the passably good, is not easy. Some of the greatest standards of the contemporary symphonic orchestra were detested at first hearing. The premiere of even so incontestable a classic as Tchaikovsky's Violin concerto caused the leading music critic of the time to speak of "music that stinks in the ear." Art of any kind is a subjective phenomenon.

 

Most people think of composition-or any form of creation-as an act of the will. Whatever that trait, computers do not possess it now. Hardly anyone in the field will speculate that one day a computer may "wake up" and "decide" to compose a 10th Beethoven symphony.
Yet surely one could program a computer to generate musical themes at random and provide all the counterpoint, harmony, and orchestration. Just as human composers have been doing for centuries, the computer could submit its finished works to human critics to see which are worth publishing and performing.
With a little more progress in computer capabilities, we can imagine a computer critic- a program with so much information about human musical tastes that it can listen to a thousand of the most promising themes to find one recording that might go platinum.

PERHAPS THE BEST ANSWER TO THE question of whether computers can creatively compose music is that it's just the wrong question to ask. Music's worth depends on how the mind values the sound vibrations that strike the ear. It doesn't particularly matter how those pulses in the air are generated. Maybe the entire process of composition should be rethought.
All over the world musicians and scientists are exploring our symbiotic relationship with the computer to find new kinds of music. At MIT's Experimental Media Facility, David Rosenthal programs computers to "listen" and "react" to music as we do so they can "learn" to harmonize with humans. Not limited by physical restrictions, the computer can play superhumanly fast. Tod Machover, a cellist, a noted composer, and the director of MIT's media lab, anticipates a time when a single performer, computer aided, can play an entire synthesized orchestra of hyperinstruments. If so, the performer can also be the composer (with the help of the computer), so that the human being who creates the idea of the music can be the same one who delivers it to the audience.
Many theoreticians of the computer's long-range future don't imagine a time when independent artificial intelligences will put human beings out of work; they suppose an ultimate equal partnership between man and machine. Even so, some of the best of them foresee what Hans Moravec, the author of Mind Children (Harvard University Press, 1988) and the director of Carnegie Mellon's Mobile Robot Laboratory, calls the postbiological phase of human history. Moravec believes that within 40 years, computers will become fully intelligent machines. They will combine increased calculating ability with the emergence of humanlike reasoning powers, learned through interaction with people and progressive robotic efforts to master perception and motion. Once the machines attain this level of competence, they will improve themselves, accelerating the evolution of this learned intelligence far beyond the slow pace of our biological selves.
"Someday we will bequeath to our machines the mechanisms (hundreds of millions of years in the making) that allow us to appreciate, and thus create, music," says Moravec. "Shortly thereafter the machines will inherit the sound-interpretation sense of mockingbirds and humpback whales. Music will never be the same."