While others may wring their hands about the coming Jobs Apocalypse, we arty types have always felt secure . No bot could do what we do, could it? Hahahahahahahahaha!
The Art of Algorithms
I’m listening to music composed by Emily Howell as I write—from her album From Darkness, Light www.amazon.com/Emily-Howell-CHABRIER-CHOPIN-DEBUSSY/dp/B003JTUE1Y. I’m afraid that her headlong Prelude gallops a bit too fast for my taste but her Fugue II www.youtube.com/watch?v=wd__bsIDih0 is pleasantly complex, reminiscent of Bach. Emily was just a teen when this album was released and I think she shows promise. I’m tempted to send her a fan note but she wouldn’t be able to read it. You see, Emily is the brainchild of David Cope http://artsites.ucsc.edu/faculty/cope, a composer and professor of music at UC Santa Cruz. She’s a computer program.
David Cope is something of a digital provocateur. Back in 1980 he was a rising star in the music world, his work earning praise as it was performed around the world. Then he got stuck on a commission for an opera. To ease his composer’s block, he started playing around with programs to analyze the masterworks of classical music. The eventual outcome of his work was EMI http://artsites.ucsc.edu/faculty/cope/experiments.htm, short for Experiments in Musical Intelligence. Turn EMI (or Emmy, as some call her) loose on a body of work, say that of Johann Sebastian Bach www.jsbach.org, and it . . . um . . . she will deconstruct it into parts, then look for common signatures in those parts from which a unique style emerges and then recombine those signatures into new work.
This is easy enough for me to type, but it took years for Cope to make EMI work. Key to the process is what he calls recombinancy. In 2003 the composer Patrício da Silva www.patriciodasilva.com interviewed Cope www.thesoundstew.com/2010/07/interview-with-composer-david-cope-part.html and asked:
“At the core of EMI is the idea of recombinancy. Does a composer compose or recompose?”
Cope’s reply: “Music composition consists of a combination of what we hear and what formalisms we bring to bear. If I compose a work freely (i.e., without a prescription for voice-leading, allowable verticalities, etc.) then I will most likely integrate various ideas that I’ve previously heard. If I compose a piece strictly using a mathematical formula, then I won’t be re-composing music that I’ve heard but following strict rules. Most music consists of a combination of these two factors. The notion that humans have some kind of mystical connection with their soul or God, and so on, allowing them to produce wholly original ideas (not the result of recombination or formalisms) seems ridiculous to me.”
Soulless music! OMG! In another interview, this time with the Guardian www.theguardian.com/technology/2010/jul/11/david-cope-computer-composer, Cope said, “The question isn’t if computers possess a soul, but if we possess one.” Carried to its logical conclusion, Cope seems to be arguing that all music is essentially inspired plagiarism. If that seems too extreme, it’s worth interrogating our understanding of plagiarism. Is it basically an on/off function? Or is it a continuum, progressing from outright theft to unconscious borrowing to inspiration, allusion, homage, and satire? Did George Harrison plagiarize the Chiffons’ “He’s So Fine” when he wrote “My Sweet Lord” www.ultimateclassicrock.com/george-harrison-my-sweet-lord-plagiarism? A judge said he did. And what to make of postmodernism’s doctrine of intertextuality http://en.wikipedia.org/wiki/Intertextuality that underlies so much of our sampling culture?
Emmy and Emily
Once Cope perfected Emmy’s routines, he used her to create new work in the style of Bach, Chopin, Beethoven, and more. Bach was a particular favorite, so Cope turned Emmy on to the master’s chorales www.wikipedia.org/wiki/Chorale. One day he powered her up and went out for a bite to eat. When he came back that afternoon she had created 5000 new Bach chorales http://artsites.ucsc.edu/faculty/cope/5000.html. Now you may be wondering how good such instant music could possibly be. Judge for yourself here www.youtube.com/watch?v=PczDLl92vlc. Or how about a new Bach fugue (no. 44) www.youtube.com/watch?v=6BDEHMBsKfw, from Emmy’s Well Programmed Clavier? In 1993 Cope released an EMI album, Bach by Design, followed by Virtual Mozart (check out the Allegro www.youtube.com/watch?v=Kc9Iw5vnCwI and Virtual Rachmaninoff. While I am by no means an expert when it comes to classical music, I am an avid listener. I am quite certain that without prompting I couldn’t tell the difference between Emmy’s virtual Mozart and sonatas by the man himself.
But lest you attribute this to my tin ear, consider the sort of musical Turing Test http://en.wikipedia.org/wiki/Turing_test that the cognitive scientist Douglas Hofstadter www.theatlantic.com/magazine/archive/2013/11/the-man-who-would-teach-machines-to-think/309529 arranged in 1997.
(The Turing Test, for those not yet up to speed on Artificial Intelligence, was first proposed by the great Alan Turing www.turing.org.uk in 1950 and posits an answer to the thorny question “Can a machine think?” To overly simplify Turing’s idea: if a human in a protracted conversation with a machine programmed to mimic human responses can’t tell that she is talking to a machine, the machine passes Turing’s test.)
As reported in the New York Times http://www.nytimes.com/1997/11/11/science/undiscovered-bach-no-a-computer-wrote-it.html, Hofstadter arranged for a recital at the University of Oregon at which three pieces were performed. One was by Bach, one by Emmy composing as Bach, and one by Dr. Steve Larson www.gazettetimes.com/news/local/obituaries/steve-larson/article_705f1118-bccc-11e0-9a96-001cc4c03286.html in the style of Bach. Dr. Larson was a professor of music theory as well as an accomplished composer and a devotee of Bach. “Bach is absolutely one of my favorite composers,’’ he said at the time. ‘‘My admiration for his music is deep and cosmic. That people could be duped by a computer program was very disconcerting.’’ Larson’s wife performed all three works and then the audience of music lovers was asked to decide who (or what) had written each. Their opinion? Emmy got the most support for being the real Bach. Bach came in second and Dr. Larson was third.
In Gödel, Escher, Bach http://wikipedia.org/wiki/Gödel,_Escher,_Bach, his classic inquiry into the nature of consciousness and intelligence, Hofstadter speculated whether music might someday be written by computers, but he believed that it would take more than mere algorithms to create anything worth listening to. He found Emmy’s achievement disturbing. ‘‘The only comfort I could take at this point comes from realizing that EMI doesn’t generate style on its own. It depends on mimicking prior composers. But that is still not all that much comfort.”
Emmy’s facility in mimicking the greats and her prodigious output disturbed Cope as well. He told the Guardian, “Because my program was continuing to pump out music like a spigot, it became a problem of: ‘Why play this sonata and not that one?’” Eventually, after eleven thousand chosen pieces, he decided to shut Emmy down and move on to developing her “daughter” Emily Howell. Cope writes on his website, “Since the early days of Experiments in Musical Intelligence, many audiences have heard its output in the styles of classical composers. The works have delighted, angered, provoked, and terrified those who have heard them. I do not believe that the composers and audiences of the future will have the same reactions. Ultimately, the computer is just a tool with which we extend our minds. The music our algorithms compose is just as much ours as the music created by the greatest of our personal human inspirations.”
The difference between Emmy and Emily is that Emily doesn’t imitate. Rather she is developing a style of her own based on feedback. Cope has given her a huge musical memory that consists of a collection of Emmy’s compositions. This means that she is intimately familiar with all the composers her mother analyzed and recombined over her career. Working from this reprocessed knowledge of music history, “The program produces something and I say yes or no, and it puts weights on various aspects in order to create that particular version,” says Cope. “I’ve taught the program what my musical tastes are, but it’s not music in the style of any of the styles—it’s Emily’s own style.”
Ask most composers or visual artists or writers—indeed, anyone in the arts—whether a computer could do their job and you’ll get flat out no way, perhaps with a snicker of scorn. Sure, truck drivers and cabbies should be worried, and maybe it’s time for cashiers and accountants to ponder second careers. Do you recall my previous column “Welcome Our Robot Overlords” www.jimkelly.net/blog/2016/11/29/m7w8c1ph5o4xcpzgtypq1ywoo23f6w, in which I pointed to several studies that warned of huge job losses due to automation and expert systems? But do poets need to worry about computers?
Consider this article: Artificially intelligent painters invent new styles of art www.newscientist.com/article/2139184-artificially-intelligent-painters-invent-new-styles-of-art, which reports that researchers from Rutgers University have created a generative adversarial network (GAN) to mimic the various artistic styles like Baroque, Pointillism, and Abstract Expressionism. In a GAN, two neural networks interact: one creates a solution and the other judges it, continually testing the product against a knowledge base. In this case, the product is art; one neural net generates images and the other consults a database of paintings in the chosen style and plays art critic. But the researchers also programmed a creative adversarial network (CAN) designed to make art that does not follow any one style but instead is “maximizing deviation from established styles and minimizing deviation from art distribution.” Their discriminator network had a database of some eighty thousand paintings from a thousand different artists drawn from WikiArt www.wikiart.org with which to steer the creator network.
Doesn’t this sound like Cope’s Emily Howell project? But wait, there’s more! The researchers then asked test subjects to look at four different groups of images and guess which were created by algorithms and which by humans. Two groups were by neural networks: a collection of GAN (mimicking) art, and a collection of CAN (original) art and two were made by humans, masterworks of Abstract Expressionism www.theartstory.org/movement-abstract-expressionism.htm and an assortment of non-figurative art on display at the Art Basel https://artbasel.com international art fair.
Yes, you’ve heard this story before. While 85 percent of respondents correctly guessed that humans had painted the Abstract Expressionist paintings, 53 percent thought the CAN images and 35 percent thought the GAN images were made by real people. Just 41 percent thought the art hanging in the Art Basel show had come from human hands.
Okay, maybe composers and visual artists might be facing some algorithmic competition someday. Maybe. But certainly not writers!
Meet Alena, short for Artificial Life Emerging from Natural Affinities. She’s another brainchild of David Cope; she’s software specializing in haiku. And she’s been published. Comes the Fiery Night www.amazon.com/Comes-Fiery-Night-D-Cope/dp/1466219157/ref=sr_1_1?ie=UTF8&qid=1320082013&sr=8-1 is an experimental volume—or is it a stunt? You see, it contains two thousand haiku by humans and a machine, only there are no author attributions. So which were written by Alena and which by flesh and blood poets?
You’re human. You figure it out.