Experiments in Musical Intelligence

[Guest chax]

http://arstechnica.com/science/news/2009/09/virtual-composer-makes-beautiful-musicand-stirs-controversy.ars

 

http://arts.ucsc.edu/faculty/cope/mp3page.htm

 

When most of us think about a machine composing musical pieces, we think of primitive songs coming out of a HAL 9000 that could be suitable for a child's toy, but nothing that music lovers would actually enjoy. That's because most of us haven't heard of Emily Howell. No, that's not a person—it's the name of a computer program written by University of California-Santa Cruz professor David Cope that, after nearly three decades of work, is about to release, uh, "her" first CD through Centaur Records.

 

Cope is Dickerson Emeriti Professor at UC-SC—he attempted to retire years ago because he didn't want to go to meetings anymore—teaching graduate courses in music composition and computer-assisted composition. Cope is also an Honorary Professor of Computer Science at Xiamen University in China and is often ascribed as a computer scientist, though he insists that he is a music professor first, not a CS professor. However, given the work he has done on Emily Howell and "her" predecessor, EMI, it's clear that he has managed to mesh the two in ways that go far beyond a musical computer program.

How it all got started

 

You name the composer and EMI could analyze his works to spit out a new piece that sounded just like that composer had written it himself. Except he hadn't; a computer had.

 

Emily wasn't just born out of Cope's mind as a genius application that could compose pieces that compete with the likes of modern-day composers. Cope, a composer himself, said his work actually started around 1980 when someone commissioned him for an opera and he found himself facing mean case of composer's block. A friend who worked in artificial intelligence lightheartedly suggested he write a program to help him complete the opera. He had done some programming before—"I had to work on those large IBM mainframes with the paper cards that went in; there was no visual complement at all"—so he gave it a go.

 

"We assumed the program would create terrible music from the start. We assumed that, when I heard it, I would say 'I wouldn't do that, I would do this.' Sure enough, that's exactly what happened," Cope told Ars. The music was bad, and creating it was a pain; the team had to use numbers spit out by the computer and transcribe them onto a staff, then sing it or play it on a piano. So, the program got scrapped in favor of developing a new system that was "more my style." The problem was that this new program didn't quite understand Cope's style too well—"As time progressed and I got this idea of data-driven programming," Cope recalled. "Instead of me writing all the rules for the program, I would create a database of music and have the computer analyze that data to attempt to produce something like it."

Experiments in Musical Intelligence

 

Cope finally managed to complete his opera seven years late, but more importantly, he had developed a project that he called Experiments in Musical Intelligence, or EMI (pronounced "Emmy") for short. EMI was originally developed to analyze his own musical style by feeding it past compositions in hopes of finding patterns—little musical signatures that only he does—and replicating them in new ways. However, he found that the program wasn't just good for him—the analytical engine worked for nearly every composer in history. Bach, Beethoven, Mahler—you name the composer and EMI could analyze his works to spit out a new piece that sounded just like that composer had written it himself. Except he hadn't; a computer had.

 

Needless to say, once the rest of the world began to get wind of this, things got pretty interesting. On one side, academia embraced Cope's work. "Extremely positive feedback came from the science community," Cope said. "They really seemed to feel that my work has had great potential in many areas other than just music."

 

The music side was a different story. "Most musicians, academic or composers, have always held this idea that the creation of music is innately human, and somehow this computer program was a threat in some way to that unique human aspect of creation," Cope said. "I have always refuted that by saying a human built the machine, listens to the output, and chooses what's the best. What's less human about that than if I had taken years and years to just compose the whole thing myself?"

 

A lot, apparently. Because EMI is just a computer program, Cope needed musicians to play the pieces that EMI had created. A number of big-name classical performers expressed interest, but their agents wouldn't let them touch it with a ten-foot pole, citing industry controversy over Cope's work. "They thought it would blemish the name of the performer," Cope said. Apparently, the number of negative reactions "far outweighed" the number of positive ones.

 

(You can listen to a number of MP3s generated by EMI, but be forewarned: some are better than others, and they are all being performed by a machine, therefore making them sound much more mechanical and primitive than they would if being performed by a live person. Even an authentic Joplin piece would sound a little lifeless when being interpreted by a machine.)

The "birth" of Emily Howell

 

Cope isn't exactly the type of guy who lets detractors get to him. "I initially get five minutes of anger, and then I'm immediately back to work again saying 'I'll show those SOBs.'" But instead of improving upon EMI to create something that would please even the most vocal critics, Cope decided to scrap the entire project for something that he thinks is far more controversial.

 

Around the start of 2003, Cope decided to take his decades'-worth of musical databases to the virtual dump. He kept the fruits of EMI's labor, but not the data that generated those pieces—"I essentially stopped doing historical replications. That aspect of my work was finished," he said. From there, he created what he describes as the succession of EMI: Emily Howell.

 

Instead of feeding Emily a database of works that already exist, he gave her a collection of works that EMI had produced to get it going and, from there, she began working on her own musical style. Cope described Emily's style to be similar to modern composers, a "sort of an amalgam of all styles" and very contemporary. But what makes Emily interesting isn't just that; it's the fact that she can take audible feedback—musically or verbally—from an audience in order to modify her compositions. "The program produces something and I say yes or no, and it puts weights on various aspects in order to create that particular version," said 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."

 

(This is, of course, an extremely simplified explanation of how Emily Howell works. One of Cope's many books, Computer Models of Musical Creativity, goes into excruciating detail about how Emily operates if you're interested in learning more.)

 

It's for this reason that Cope believes Emily might—or at least should—be regarded as more controversial than EMI. "Most living composers feel exempt from the controversy with EMI because their music hasn't been affected by the fact that any music that sounds like Mozart is good or bad," Cope explained. "Now those same composers are looking at a competitor—a virtual composer competing in the same arena with 'her' own style and music that is really excellent. It seems to me that these composers should feel a little less smug and more defensive about their position."

 

Whether Emily Howell will generate the same level of attention as EMI remains to be seen, though. Because "she" basically represents a new composer, much of the world might actually brush it off as just another AI experiment whereas EMI was akin to bringing old composers back from the dead. Cope doesn't care though—it's all about furthering his research and learning how to challenge common beliefs in new ways. In fact, he's already working on his next project. "I'm currently writing a book that is attempting to claim that music is a science through mathematics. It involves game theory to create music, trying to prove that there are aspects of music aside from acoustics and sound itself—that composition is not just inspiration," he said. "To me they are as related as anyone can imagine; when I'm composing all day, I'm programming. When I'm programming, I'm composing."

 

 

I totally disagree with this guy thinking that he can actually create an original composition using artificial intelligence and computer programs. He's basically just telling the program to use 'samples' or influences from other composers and songs and then gets the computer to create variations of that based on other peoples work. I am absolutely sure that nothing will challenge or mimic or even come close to the realm human beings creating music. Other people might find this kind of thing a lot more interesting so I thought I'd share

[awepittance]

have you checked out any of his music? if you disagree without hearing it i think that's a little silly personally. (just noticed you posted an mp3 link so you probably heard it)

 

I found his computer generated Bartok concerto's to be ridiculously impressive, and his Chopin generated pieces amazing.

 

I am absolutely sure that nothing will challenge or mimic or even come close to the realm human beings creating music

 

you are very much living in the past my friend, i guarantee you will eat these words in your lifetime.

 

Copes' techniques and theory have had a huge influence on me, i sat in on his lectures many times at UCSC

 

edit:

I totally disagree with this guy thinking that he can actually create an original composition using artificial intelligence and computer programs. He's basically just telling the program to use 'samples' or influences from other composers and songs and then gets the computer to create variations of that based on other peoples work

 

isn't this basically what the human brain does though? We pick up influences around this and combine them and channel them musically. This could be compared to a computer 'sampling' a musical score.

[Braintree]

I think a computer is just as capable of writing good music as any person is. On that level, it's still a science. To be able to write a masterpiece to be appreciated for generations, I'm not so sure...

[jeremymacgregor87]

Generative Music?

 

Eno

[kaini]

well, it's very easy to feed a system a zillion midi transcriptions of jazz improvisation

and it's very possible to produce a system that analyses those midi files to spots common improvisational rules

and it's easy from there to produce a system that uses those rules to produce new compositions,

with parameters called horrible things like 'degree of improvisation'

 

...but the only things unable to spot the difference from a human being

 

 

interesting, a sort of musical turing test

 

 

 

...were other computer programs.

 

regarding eno, is he still using KOAN?

hint: no.

[awepittance]

i think it's hard to find people who are great computer programmers as well as good musicians. It has different less impressive results i think when a musician tries to use an algorithmic music generator made by someone else. I can only think of a few examples in my head of people who have done a really interesting fusion of the two: Cope, Autechre, Cylob and Vladislav Delay. The guy who did the 24 hours of Beethoven i would also put into this category but i forgot his name

[kaini]

every melody in http://www.project168.co.uk/music.php?p=1&e=jittery168&t=140&ap=yes is based on the first 16 bars of a gavotte by js bach

[chenGOD]

well i dunno much about much, but the chopin mazurka on that mp3 page is pretty fucking sweet.

[khov]

i think it's hard to find people who are great computer programmers as well as good musicians. It has different less impressive results i think when a musician tries to use an algorithmic music generator made by someone else. I can only think of a few examples in my head of people who have done a really interesting fusion of the two: Cope, Autechre, Cylob and Vladislav Delay. The guy who did the 24 hours of Beethoven i would also put into this category but i forgot his name

 

Leif Inge (24 hour beethoven symphony), it should be out on dvd soon. one of my favorite piece of music in recent history. not sure, if its proper music but it's fantastic.

[modey]

i think it's hard to find people who are great computer programmers as well as good musicians. It has different less impressive results i think when a musician tries to use an algorithmic music generator made by someone else. I can only think of a few examples in my head of people who have done a really interesting fusion of the two: Cope, Autechre, Cylob and Vladislav Delay. The guy who did the 24 hours of Beethoven i would also put into this category but i forgot his name

i wouldn't consider 24 hours of beethoven to be generative, that was just slowed down with an equivalent program to paulstretch, no?

 

and regarding autechre, i thought they only used generative stuff very minimally, and controlled the parameters to such extent that it basically wasn't very generative after all? correct me if i'm wrong!

[khov]

no leif inge work isnt generative, its an extension, based on something created years ago & regenerated by computer skills (slow down without changing the pitch, its called granulation i think). well, with Ae, from the patch found via GBIII & Eno i think the "generative" means that the software auto-adjust some parameeters (beats/ambient lines) given by the artist. i've played with graphics software generative machines & its more like a (fed with bits of random/rules)sequencer than a wizard software, creating symphony, just like that, out of nothing.

[mcbpete]

i wouldn't consider 24 hours of beethoven to be generative, that was just slowed down with an equivalent program to paulstretch, no?

Yep, whatever they used on 24 hours of beethoven was basically like a precursor to the methods used in Paulstretch - some of the transitions of the granules in the aforementioned beethoven doo dah are pretty jarring, it's a shame the window size wasn't larger.

[chim]

computers are useless, they can only give answers

[Dementia]

I'm not yet sure why, but I dislike generative art very much.

[kaini]

well, it depends how 'generative' it is.

 

eno's recent methods at installations and stuff is to make [lots of short] or [a few long] remixes of existing tracks, and some other textures and things, and then compile these into cds, but leave some tracks on the cd as just silence. then he puts cd players looping these cds in various spots in the installation.

 

this is generative, but it's also an interesting exercise in making a big-ass stereo pad, you could stick some percussion in there, field recordings... i think it's a good strategy - you can be very intentional about the content of the music - the generative element is mostly in the timing

 

http://en.wikipedia.org/wiki/Extracts_from_Music_for_White_Cube

http://en.wikipedia.org/wiki/Kite_Stories

http://en.wikipedia.org/wiki/Music_for_Civic_Recovery_Centre

http://en.wikipedia.org/wiki/Compact_Forest_Proposal

 

etc

[Guest ezkerraldean]

pisses me off when people dismiss things like this out-of-hand because it's "electonic" and "soul-less" and "cold" and all that wank. some people say that about electronic music in general.

 

i want to kill them

[mcbpete]

i want to kill them

That's a bit harsh....

[Guest hahathhat]

"It was always this kind of metaphysical, almost magical thing, about literally thinking things to the point where they would happen," says Herb Deutsch, a Hofstra University music professor who worked with Moog to develop the first Moog synthesizer in 1964. Deutsch, who also worked for Scott, remembered one of his colleague's visionary objectives. "He wanted to take the work out of being a musician," Deutsch says. "That used to really get me upset. He said, `Look, I just want to sit here, and I'd like to turn this machine on, and whenever it does something good, I just want to record it at that point.' It was not that he was a lazy guy - far from it. He worked incredibly hard to take the work out of being a composer."

 

http://www.raymondscott.com/em.html

[awepittance]

i wouldn't consider 24 hours of beethoven to be generative, that was just slowed down with an equivalent program to paulstretch, no?

Yep, whatever they used on 24 hours of beethoven was basically like a precursor to the methods used in Paulstretch - some of the transitions of the granules in the aforementioned beethoven doo dah are pretty jarring, it's a shame the window size wasn't larger.

 

true, i didnt mean to imply it was generative, just a very good fusion of musical art and computer programming . things like it had exsisted before like the Kyma spectral time stretch but it's my understanding the guy made his own algorithm in something to make the Beethoven piece

[awepittance]

pisses me off when people dismiss things like this out-of-hand because it's "electonic" and "soul-less" and "cold" and all that wank. some people say that about electronic music in general.

 

i want to kill them

 

well that's kind of my thinking except i don't want to actually kill them i just think it shows a form of human exceptionalism, meaning that its the belief that things human's create cannot autonomously on their own create things more diverse/complex/or as interesting as an actual human. I think in my lifetime that theory will mostly be proven wrong although there will still always be doubters. I don't think that the 'singularity' will happen necessarily but things very much capable of fooling humans into thinking they are 'real' or not completely made from a computer is already happening. I think Cope has actually proven this hypothesis already by tricking classical music enthusiasts who are well studied on for example every piece Mozart has ever written into thinking a piece he generated on the computer was genuine.

[awepittance]

 

and regarding autechre, i thought they only used generative stuff very minimally, and controlled the parameters to such extent that it basically wasn't very generative after all? correct me if i'm wrong!

 

again this is not what i meant (looking back on ehwat i said it was kind of confusing). I just meant that the way autechre uses things like max/msp so heavily to generate things like the trademark AE farting bassdrum sound is a GREAT example of really talented musicians who also have a knack for programming musical toys or instruments.

 

and i was comparing it to say Brian Eno using Koan, which he did not program himself. i think i can have some rather beautiful results when the artist is really adept at both and uses them together to make music. Tim Exile is another good example, but it's more about his live shows not the recorded music.

[Guest ezkerraldean]

lol i don't really want to kill them. i'd just invite them out with a bunch of friends, and buy everyone except them an ice cream.