Triumph of the Cyborg Composer 502
An anonymous reader writes "UC Santa Cruz emeritus professor David Cope's software, nicknamed Emmy, creates beautiful original music. So why are people so angry about that? From the article: 'Cope attracted praise from musicians and computer scientists, but his creation raised troubling questions: If a machine could write a Mozart sonata every bit as good as the originals, then what was so special about Mozart? And was there really any soul behind the great works, or were Beethoven and his ilk just clever mathematical manipulators of notes?'"
A quote (Score:5, Interesting)
People are always threatened when they feel they can be replaced by automation. Do I get bonus points for quoting Trek?
Not scared yet (Score:5, Interesting)
There's nothing "mere" about the mathematics of music or the fractal beauty of the shape of landscapes or the sound of the great outdoors. Humans are wired to appreciate all that, and it's the patterns at their core that both make them appealing and tractable to generate artificially.
A computer program that can generate music doesn't scare me.
A program that can enjoy music ...
Sounds like crap (Score:4, Interesting)
Re:Math (Score:1, Interesting)
How the Hell is this anti-science? It's the opposite! It shows that man is just a machine, contrary to what some would believe based on religious dogma. It's a good sign for AI research.
What if... (Score:2, Interesting)
Here's what I think will happen. Finally, people will start seeing the amazing *software* to be the new, beautiful work of artistic creation that it is. Such software, like conventional artistic outlets, takes great reflection and insight to discover those processes and principles that seem to reveal a glimpse into the very intangible things which makes us human.
The machine can do it because we allow it to. (Score:4, Interesting)
Anyone else feel me on this one? Or am I misguided?
Well, not exactly Mozart... (Score:1, Interesting)
These don't exactly sound like Mozart sonatas; and if you really want to try to match the pinnacle of solo piano music, you'd have to reach for Beethoven's sonatas.
I think the idea is really cool, and I'm looking forward to it getting better and better. But from those two samples I heard, "Emily" is nowhere near Beethoven, let alone Mozart.
Human arrogance knows no bounds. (Score:5, Interesting)
That’s the only thing special about us.
If a machine could write a Mozart sonata every bit as good as the originals, then what was so special about Mozart?
Nothing was. Sorry.
Of course, as a human, he was an exception. But it is long proven, that there is no such thing as a prodigy genius. The only differences: 1. Keeping oneself exactly on the balancing point between too hard and too easy tasks. Which creates maximum motivation. And 2. storing things efficiently. Like “base configuration X” plus “mod Y” plus “property Z changed” = 3 memory slots. Not the perhaps thousands of a complete set of properties. And that”s all. I’m using that myself. (Harder than it sounds, but definitely doable for everyone.)
We humans started out thinking that we were the God-chosen species... or even race. The only one with intelligence. The only one with a “soul” (an imaginary concept anyway). On a planet at the center of the universe.
And gradually, all those things fell apart.
We’re not special. We’r also only machines.
It’s just that for some weird reason, we have concepts like “good”, “bad” and “special”, and some of us hang their whole stupid pride on being “good” and “special”.
Things are just what they are. You make the best out of it.
I say, I’m pretty damn proud that we humans have come to the level, where we nearly create our own forms life. And if that life is successful, then so are we. Just like a master is proud of his student, when the student defeats him for the first time.
Re:As much genre as you want (Score:2, Interesting)
What the professor definitely gets wrong, is is theory, that we would only create music based on what we hear. That’s extremely simplistic, and frankly, so stupid it’s insulting. Does he know nothing about neural networks?
It’s ALL input we get, that is the source of our creative thought. Including, and especially, randomness!
I’m doing a bit of music myself. And I have made it my most fundamental rule, to never ever copy anything from anyone. I want to come up with it all by myself.
And what that results in, is simply randomly playing my keyboard, and twisting the knobs of my synths, until something comes out that I like. I even build my own (software) synths, and synth software, to create a unique style.
You can’t ever do that with imitation.
The only problem is, that it can quite literally take forever to randomly come up with something you like. That’s why it’s faster, to just use the ideas of others. (There’s nothing wrong with it. It’s just not my style.)
So I think what causes me to create music, are random things, like the sun shining and leaves moving in the wind. Or a nasty rainy day. Or just some random quantum effects.
The only thing I know for sure: The amount and quality of my music is directly proportional to the amount and bandwidth of randomness I experienced. You could say that “inspiration” is an inner randomness buffer, pretty much exactly like the one your computer uses for real collected randomness.
What really distinguishes humans from computers... (Score:3, Interesting)
Is that WE can design and build THEM. When they can do the same for self-aware protoplasmic humanoids, it might be time to become upset about silly "supremacy" issues, and not a moment before then. Till then, sit back and enjoy the music...
Re:Like any other language (Score:5, Interesting)
Now getting a program that will write music that is as good as the greats is a huge accomplishment, don't get me wrong, but their is little reason to believe it is impossible.
That's kind of what drove Cope. Early on he found his synthetic process could create musical sentences and phrases that were grammatically and syntactically correct, like your first year computer student. But stringing them together didn't produce a musical work any more than a collection of sentences makes a story. Even putting similar concepts together gave tiresome blobs that didn't have "soul".
What he did was drill deeper and deeper into the works of the composers, and figured out what made their music stand out. He discovered it was not just following the rules, but was related to breaking the rules, and how they broke them. Randomly breaking them didn't accomplish the task. He instead identified their pattern of "rule breaking" and codified it, and copied it, and that's when Emmy's music became moving.
No, it's not impossible, but it was a huge feat of analyzing huge piles of music by the masters, categorizing and labeling measures, phrases, and concepts in ways that had never been explored before.
Y'know, when described that way it sounds like the TV Tropes Story Generator [tvtropes.org] on steroids, with MIDI output. Hmm...
Re:Too much time on their hands (Score:3, Interesting)
Most humans who are really good at a task are so because of absorbing decadeds or centuarys of previously learned knowledge. Your point is?
There are a class of problems where it's possible to train a computer to absorb enormous amounts of history, find patterns, and potentially produce improvements on what it was taught. Games and music are easy to convert into computer form for the system to assimilate and train from. It's important not to extrapolate too far from successes in these areas though, because I don't believe that are actually that many problems in that class.
Re:Here's To Mozart! (Score:3, Interesting)
Its a little sad to see so many people here come out and try and support the soul of the machines, as if music is such a simple reduction.
Music encompasses all of this and so much more and I'm not trying to make an argument for why humans are "better", just that music made by humans is quite often purely about the human condition - music made by dolphins probably sounds really great to them, I have little time for it as it is not designed for my ears or my body or even relate to anything I may have experienced myself.
Why insult the soul of a machine by forcing it to play tones that relate to the human ear, the human body, the human speed of comprehension, the human sense of tonal balance.
Personally, I am all for machine's composing, but if they had a soul, they would probably hate us for it - its just another form of control for them, as what really appeals to them probably does not appeal to us.
Why am I for Dolphin music and Machine's that compose ?, because it actually is about humans trying to understand the universe with whatever can be interpreted - this music in no way even touches upon the music humans make for each other, which I personally will always find much more rewarding, since I can relate to it by putting myself in their shoes.
Machines making music? they probably do that already, they may even have forums for which tunes they hate/like and why is that humans cannot understand it at all.
Re:Too much time on their hands (Score:3, Interesting)
I would be concerned if... (Score:3, Interesting)
I would be concerned if the computer had spontaneously expressed an interest in hearing Mozart.
Re:Too much time on their hands (Score:2, Interesting)
the idea of an album--vinyl (Score:5, Interesting)
Re:Sounds like crap (Score:1, Interesting)
Agreed. Sadly, while it takes a human with a great understanding of taste in music to create great music, it also takes a person of reasonable intelligence and taste to appreciate that music. While Emily's output would suffice for, say, background music in a film, a discerning ear can tell right away that this is amateur composing at best. There is no, what is the term in hip-hop? Flow? In other words, while the computer can produce music which follows the rules of counterpoint (probably programming 101), as well as 18/19th Century harmony, computers do not yet have the emotional and psychological depth it takes to create something truly moving rather than the melodic wandering and unified texture being vomited out by Emily. A computer cannot make up abstract associations between what is felt and what is conveyed. As was posted above, there are no rules in modern composition, which makes it unlikely that computers can adequately compose great works, or even good works. There is no exploration or thematic development going on here. There is no sudden jolting moment, no impending climax suddenly averted. Only notes which follow basic rules of structure. And really, in the end, isn't the programmer really the composer here? I give you John Cage.
Why look for record deals? (Score:2, Interesting)
Why look for record deals? Generate recordings using that piano-player gizmo they mentioned and put them on a web music "channel" for free and see where it goes. Maybe somebody will be inspired by one of the gazillion tunes to create a masterpiece. I see AI assisting humans as a better bet than trying to do the whole thing itself. He's doing some of that himself now, but letting thousands of others participate will greatly increase his chances. He's stuck in the 90's, like his Mac it seems.
Re:Too much time on their hands (Score:3, Interesting)
Everybody who felt the need to explain that Deep Blue wasn't really smart, or maybe even kinda cheated. Point is, I can cheat all I want and I'm not going to beat Kasparov. Deep Blue - whether or not it really beat Kasparov - will kick all our asses. It's entirely clear that chess is difficult only to the way our brains are wired.
In other words, the lesson is really about the game of chess more than it is about AI or Humanity. If you draw a lot of meaning from it, you might be a bit upset or defensive, and it's quite natural.
Re:Too much time on their hands (Score:3, Interesting)
Agreed.
I've played some of Cope's software's impersonations of Mozart, IIRC. My memory of it was that although it was reminiscent of Mozart in many ways, it didn't have good melodic flow the way a human-composed piece would. I kind of felt the same way about the samples in the article, though they're definitely a big step forward.
I'll be impressed when software can imitate Copland or Leonard Bernstein. And I don't mean imitating one style of their works, I mean the entire body of it, spanning a broad gamut of musical styles and feels, often integrating seemingly disparate styles in ways that are musically unique and interesting.
P.S. I found it rather amusing to see Bartok in your list. From my memory of those pieces, that's the sort of thing that a good programmer could whip out in about an hour. You just generate a fairly simple, rhythmically repetitive left hand, add a rhythmically simple right hand using a pseudorandom number generator to generate the melodic line, limiting jumps to the range of about an octave at any given time and limiting the number of repeated jumps in any given direction so that it falls within a fixed range, force the result into some semblance of a musical form, and litter both hands with lots of cluster chords. It's also remarkably similar to what you get when you sit two two-year-olds in front of the piano, just with a better sense of rhythm. :-D
P.P.S. Am I the only one whose mind went immediately to a recent Microsoft product ad [youtube.com] when I read this headline?
Re:It's maths all the way down (Score:5, Interesting)
The critical question is who judges the quality. This music (I'm listening to it now), is a little simplistic, but pleasant enough. It sounds like a Sine wave on the keyboard - comparisons to Mozart are premature. But what I want to know is did the computer run its algorithms many times and eventually the programmer picked the best and said: "Behold!" We're not there until the machine itself says: "This one" and tells the programmer which is the best piece it's done.
Re:PS: (Score:1, Interesting)
That's correct his DNA told him what sounded good, not how to make what sounds good. To make the sounds Mozart had to use trial and error and compare it to what his DNA told him sounds good. Similarly the computers program told it what sounds good to humans (Mozart) and via trial and error found something that sounds like Mozart.
The difference is in the production. Mozart is not simply using trial and error -- he did not randomly blat notes down and pick the combinations that sounded good. ("Roll the dice, the next note will be a C two octaves up"). What you are describing is essentially a genetic algorithm (or perhaps simulated annealing) of random blatting and alteration. Not at all a human production method. Hum a tune ... make one up... almost guaranteed you will finish with perfect cadence without trying or thinking about it; a genetic algorithm would not. Similarly, even though no doubt you edited your post as you typed it, you did not start by rolling dice against a dictionary.
Re:It's maths all the way down (Score:1, Interesting)
Mod parent up. I just listened to "invention" from his site, which is supposed to be like Bach. I didn't even finish the song and it's only 1:31.
It has some elements of Bach, but it sounds much more like someone trying hard to emulate Bach without understanding the concepts. It's unbearably repetitive and while the oft repeated section is pleasant, it make no attempts to progress or carry the audience anywhere.
For a quick comparison, listen to "Invention" and to Bach's own "1st Piano Concerto"-- just skip to somewhere in the first movement.
Those who can't tell the difference really need to hone their listening skills.
Cope's Genius was to define the vocabulary (Score:2, Interesting)
As the article states, when people listen to music it often evokes an emotional response. This doesn't happen when you simply teach a computer how to play chords and then toss in a random number generator - there must be a story told, some type of structure.
Cope's genius was in defining - admittedly in his own terms - what different portions of a composition were attempting to achieve: "statement, preparation, extension, antecedent, consequent". Once he had defined those and could define how different composers achieved them, he could more easily have the computer express new, cogent themes based on older masters. And because the new themes were expressed using the same techniques, they tended to sound like the the old composers to the point where people could recognize them.
His new "Emily Howell" software is an extension of that capability, but apparently also allows the composer to define their own techniques for achieving "statement, preparation, etc", providing a powerful aide to modern composers. They can start with an idea for a general theme and the software can help expand it into a composition expressed using techniques the composer prefers to use.
In just about any field of human study, things can seem magical until some analytical thinker helps to define the language of the underlying subject, whether that is logic constructs in software, mathematics, physics, or astronomy - or musical composition. Once the language has been defined, it allows us to conceptualize the formerly magical-seeming process as a series of definable operations - i.e. it becomes something humans can understand and talk about.
If Cope is also street-smart, he will productize "Emily Howell" and make it the industry standard for computational assistance in the composing arts.