Prime Time Replay:
David Cope
on his music-composing computer
program EMI
MsgId: *brain_storm(1)
Date: Fri Sep 5 22:03:16 EDT 1997
From: Rob_Killheffer At: 205.198.117.128
Good evening and welcome to another edition of Brainstorms. I'm your host, Rob Killheffer, and joining me tonight is David Cope, a professor of music at the University of California at Santa Cruz. Professor Cope has created a remarkable bit of software called EMI (for Experiments in Musical Intelligence) which has cast an interesting new light on the creative process. Dave, are you here?
MsgId: *brain_storm(2)
Date: Fri Sep 5 22:03:49 EDT 1997
From: David_Cope At: 128.114.128.134
I am here. Thanks.
MsgId: *brain_storm(3)
Date: Fri Sep 5 22:05:37 EDT 1997
From: Rob_Killheffer At: 205.198.117.128
Welcome to Brainstorms, Dave. Perhaps to start you could tell us a little about what EMI does, and how it does it?
MsgId: *brain_storm(5)
Date: Fri Sep 5 22:07:48 EDT 1997
From: David_Cope At: 128.114.128.134
Sure. EMI composes music in the styles of the music it uses for databases. I started working on EMI in the early 1980s. It creates music by analyzing music the ddatabases very carefully and then composes using the rules it finds in that music.
MsgId: *brain_storm(6)
Date: Fri Sep 5 22:11:23 EDT 1997
From: Rob_Killheffer At: 205.198.117.128
From what I understand, it does an amazing job of composition -- producing works that could be taken for new pieces by Mozart or Chopin, which seems way beyond what other "creative" software has done. What's the key to EMI's prowess, do you think?
MsgId: *brain_storm(7)
Date: Fri Sep 5 22:14:50 EDT 1997
From: David_Cope At: 128.114.128.134
I think the main "key" would be that the program does not contain rules of composition. Rather, these rules are derived from the different databases themselves. Also, I think that by using databases of music composed by humans the program tends to emulate what we as humans like. In effect, I think the program inherits a great deal of its successes from the original music.
MsgId: *brain_storm(8)
Date: Fri Sep 5 22:16:53 EDT 1997
From: Rob_Killheffer At: 205.198.117.128
And that's an interesting point, because it answers some of the worries expressed by people like Douglas Hofstadter over what it might mean for a computer program to be able to generate high-quality musical compositions. EMI doesn't remove the importance of the composer's human experience etc., because it actually relies on that for its own education, right?
MsgId: *brain_storm(10)
Date: Fri Sep 5 22:20:06 EDT 1997
From: David_Cope At: 128.114.128.134
I think so. However, Doug does not agree. In his mind, I think, the process is so complicated and involved that the human inspiration does not survive. I, on the other hand, believe that at least a small part of it does.
MsgId: *brain_storm(12)
Date: Fri Sep 5 22:22:11 EDT 1997
From: Rob_Killheffer At: 205.198.117.128
It sounds to me like it would. I mean, since EMI's whole strategy is to analyze existing compositions, what it's really demonstrating is that at some level what makes the music "work," so to speak, is identifiable as a pattern -- but the origin of that pattern still lies with the original composer.The real test, I suppose, would be to have EMI try to compose something more original -- not in the style of another composer, but in its own (which could of course draw on all that it had seen in other compositions). Have you thought about trying that?
MsgId: *brain_storm(14)
Date: Fri Sep 5 22:26:15 EDT 1997
From: David_Cope At: 128.114.128.134
Doug would argue that humans (students in music theory primarily) have been trying to learn the rules of certain styles and then composing in those styles for many decades with far less success, on balance, than a lot of EMI output.Concerning original output: yes, EMI has composed 5000 new works in (arguably) its own style. These include 1500 symphonies, 1000 string quartets, 1500 piano sonatas, etc. I have not heard them all nor will I have time to do so in my lifetime. We will be performing and recording EMI's Symphony 1383 this coming spring, however.
MsgId: *brain_storm(15)
Date: Fri Sep 5 22:27:18 EDT 1997
From: Rob_Killheffer At: 205.198.117.128
On balance, how do you judge EMI original output? Is it good?
MsgId: *brain_storm(16)
Date: Fri Sep 5 22:29:27 EDT 1997
From: David_Cope At: 128.114.128.134
I am probably biased. However, I find a good bit of EMI output to be very musical and occasionally inspiring. I am not at all offput by the fact that a program created the music. However, many others are quite disturbed by that fact.
MsgId: *brain_storm(17)
Date: Fri Sep 5 22:32:41 EDT 1997
From: Rob_Killheffer At: 205.198.117.128
It seems to me that EMI would be comparable to, say, a master art forger -- the best ones, who not only can produce convincing copies but even "new" works in the style of other painters. And, as in the art world, there is some sense to the argument that perhaps it doesn't matter if a particular Rembrandt is "really" a Rembrandt, if it packs the same power. So with EMI's compositions.I was struck as I read about EMI by the similarities in strategy to IBM's Deep Blue chess playing computer. Did you follow the big match with Kasparov? And do you see any similarity between Deep Blue and EMI?
MsgId: *brain_storm(19)
Date: Fri Sep 5 22:38:15 EDT 1997
From: David_Cope At: 128.114.128.134
I agree with your analogy to art "forgery." Yes, I followed the Deep Blue match with Kasparov with great interest and zeal. When Deep Blue finally won the match this year I shouted with joy. A number of individuals have asked me why I think that was such a wonderful occasion. My reply has been that Deep Blue's victory as a victory for humankind. After all, humans created the computer, the program, and the hundreds of thousands of games on which Deep Blue based its judgments. It's one thing to produce a living chess master, quite another to create a program which can defeat that master.
MsgId: *brain_storm(20)
Date: Fri Sep 5 22:41:02 EDT 1997
From: Rob_Killheffer At: 205.198.117.128
Deep Blue also had the uncanny ability to surprise the experts -- Kasparov, and even its own programmers -- which it sounds like EMI does, as well. That seems to be a landmark in the AI field -- even though some of the pie-in-the-sky goals haven't nearly been reached, the fact that we have now produced machines that can be partners in such rarefied activities as chess and musical composition certainly indicates a big advance.You started working on EMI mainly as a tool to help you in your own work, not as an experiment in AI, but clearly you've become as much an AI researcher now as a music professor. EMI is an amazing accomplishment, especially because, as I understand it, it runs on a fairly normal, everyday computer (not the huge monsters used for Deep Blue). Did you draw on current AI work when you were developing EMI, or did you arrive at your strategies independently?
MsgId: *brain_storm(22)
Date: Fri Sep 5 22:47:26 EDT 1997
From: David_Cope At: 128.114.128.134
Yes, surprise is one of EMI's most endearing qualities. I learn a great deal from such surprises. When I began EMI I used both my own and traditional AI ideas. For example, at EMI's heart is an association network (something I develop independently). At the same time, EMI employs standard AI pattern matching techniques for finding musical signatures (for example) and I use a number of ATN (augmented transition network) techniques borrowed from natural language processing.
MsgId: *brain_storm(23)
Date: Fri Sep 5 22:49:52 EDT 1997
From: Rob_Killheffer At: 205.198.117.128
From your earlier comments it seems you'd view advances in AI as basically benign, even beneficial -- that, as with EMI, AIs would be more like partners, not usurpers. Do you think there "is" anything important to worry about in programs like Deep Blue and EMI, any threat to humanity, as some people fear?
MsgId: *brain_storm(24)
Date: Fri Sep 5 22:53:57 EDT 1997
From: David_Cope At: 128.114.128.134
No, I don't believe such programs are a threat. I believe they are a challenge - and for that I am grateful. There are, of course, viruses and other programs which are not so benign, at least to other computer programs. For the most part, however, the good work that's being done today in AI is a step forward for humanity.
MsgId: *brain_storm(25)
Date: Fri Sep 5 22:55:37 EDT 1997
From: Rob_Killheffer At: 205.198.117.128
We're just about out of time, but I wanted to cover one point before we're through: as in the case of Deep Blue, EMI doesn't do what it does all by itself, even once all its data have been fed in. You still take a very crucial role in tweaking the program along the way, and of course you're the ultimate arbiter of what compositions are "working" and which are headed for the scrap heap. Do you think a program can ever be made that has that discriminating capability, that can not only compose but judge its compositions?
MsgId: *brain_storm(26)
Date: Fri Sep 5 22:59:48 EDT 1997
From: David_Cope At: 128.114.128.134
The aesthetics of such decisions are hard to define. There is a version of EMI which performs a kind of statistical analysis of the database music prior to composition and then, after composition, compares new works with that analysis. If they fail, the program tweaks itself and tries again. Ultimately it creates new works that satisfies its input analysis and I get to hear them. I am not sure that this passes for independent decision, however.
MsgId: *brain_storm(27)
Date: Fri Sep 5 23:01:51 EDT 1997
From: Rob_Killheffer At: 205.198.117.128
Possibly not. Perhaps that's how EMI and Deep Blue have truly advanced AI research, by narrowing down our sense of what the crucial issues are -- like that ineffable aesthetic sense, that kind of decision that we do all the time. That's the next hurdle. At any rate, our time is up. Thanks for appearing here on Brainstorms, Dave. I think it was a very interesting discussion. Viewers, if you're interested in learning more about EMI, you could look up David Cope's books, "COMPUTERS AND MUSICAL STYLE" and "EXPERIMENTS IN MUSICAL INTELLIGENCE," both published by A-R Editions (and probably available through Amazon.com), and you can check out some of EMI's compositions on the web at http://arts.ucsc.edu/faculty/cope/home. Thanks for coming, and see you next week!
MsgId: *brain_storm(28)
Date: Fri Sep 5 23:02:47 EDT 1997
From: David_Cope At: 128.114.128.134
Thanks for having me!
MsgId: *brain_storm(29)
Date: Fri Sep 5 23:04:30 EDT 1997
From: Rob_Killheffer At: 205.198.117.128
Thanks again for coming, Dave. It was fun! I'll be looking forward to hearing more about EMI's accomplishments in the future. Good night!
Home || Prime Time || Live Science || Machine Dreams || Project Open Book || SF-Fantasy-Horror
Continuum || Antimatter || Mind-Brain Lab || Interactive IQ || Gallery || OMNI ToonsQuestions, comments and suggestions can be mailed to the webmaster.
Copyright © 1998 by Omni Publications International, Ltd. All Rights Reserved.