Course Navigation

 Course Home Page

Robot Lab Wiki

WeeklyReactions

Final Project

Tim Ambrogi

1) The Tournament 2) Computer-Assisted Composition (CC)

The Tournament

So I know you're all sore that I snuck my way into the finals by playing the losers bracket, and that there is an overall sense of gratification at my eventual downfall at the hands of Ananya's *Unholy Fast* Konane war machine. However, I think that it demonstrated clearly for me the inseverability of man and machine. While AI had alot to do with it, my human ability to win at a coin toss and to play the double-elimination system were more important overall than my static evaluation function. While some may question my findings, I found that with two identical AI, the player who goes first will win more often than the other. I don't know why this is, but that's just my scientifically untested observation.

By the way Ananya:

Your Konane-machine

Refuted my science

Now my static evaluator

Feels like some old appliance

But if we have a rematch

You better make way

With evolved evaluation

King Konane don't play

CC

I have been reading alot of posts on the Wiki about CC. By fluke it happens that my tentative thesis topic is CC. In response to a couple of Jackie's assertions:

The easiest types of music to have a computer create are a) Atonal music and b) Catchy music. If one had a solid, comprehensive education in tonal harmony and C++, it would take a matter of months to work out a piece of software that would craft catchy chord progressions. In fact, the catchiness of a piece is based quite alot on what *has already been done*, not the creative energy that has been poured into it.

Other pieces, that are not immediately accessible or enjoyable to the average listener, are usually so because there is material in the piece that has not been done very often and is fairly foreign to our ears. Thus, due to its unfamiliarity with the underlying patterns and systems, we find it less "catchy" than the latest nsync song. However, upon repeated listening, the brain decodes and begins to understand the more unusual patterns involved. At this point a unique gratification may be derived from a skillfully crafted piece of "creative" music.

I think the reason most people do not believe that music will ever be fully automated is because they do not understand the inner workings of music and music perception psychology. I do not profess to posess a mastery in either of these fields, but I have myself coded software that isolates relatively tonal subsets of music. Having gone through this research and experience, I will assert confidently that the role of "human touch" and "creativity" in the composition of music is incredibly overrated.

The computer's freedom from a human ear and a small brain will already allow it to do systematic composition more intricately than any human. But as far as crafting new musical ideas of its own, the computer has the unique ability to forget all the music it has ever heard, and to start composing with a clean slate. In eyes of many (notably Orson Scott Card in 'Unaccompanied Sonata') this is the key to truly creative composition.

Anyways, this is something I feel rather strongly about, but I am interested in hearing any other informed points of views that others might have. There are many in the field on both sides of this fence...

--Tim A.

P.S. In response to this comment:

"I do not feel threatened by AI in music because i know that human musicians aren't going ot stop composing just because a machine can do it too. I don't tink it will ever come to that."

If you a professional composer, trying to make a living, you will have a hard time beating the machine in the end. For one thing, jobs in the film and digital entertainment industries will be monopolized by machines who can compose an entirely different film score overnight. And as soon as one market dries up, there will be many more composers fighting for the same number of commissions. The only thing in the way of all this is mankind's acceptance of the inevitable dominance of computers and those can wield them effectively. P.P.S. I realize this post may have come off as sharp or attacking. Please note that this is not my intention.

Larry Bomback

Jackie Chew

Catherine Chiu

Jason Coleman

I thought I had a pretty clever evaluation function, but I got ripped apart at last weeks tournement. Sometimes, even, it seemed that my AI was making the absolutely worst move... which was probably the case sometimes. Why? My evaluation function was:

# of possible moves - 3 * # of opponents possible moves

This "weights" the opponents # of possible moves so that the evaluator thinks that a situation where the opponent and itself both have low #'s of possible moves is better than both having higher #'s of moves. This behavior kind of pushes whoever is losing into the grave. The problem is "What if the AI is losing?!?"

Well, I tried having an evaluation function that would tell if the AI was losing and in those cases weight the AI's # of moves (which would result in higher numbers of moves for everyone generally). But you can't have both of those evalutation functions floating around in the same tree because they conflict with one another... and you get situations where 3 moves for you and 4 moves for the opponent (9-4=5) is better than 3 moves for you and 1 move for your opponent (3 - 3=0)... Not a good evaluator.

So I went with just the aggressive (mine - 3 * yours) evaluator for the tournement. I got several comments about how well my AI played the middle game (probably because it seemed to be minimizing the numbers of moves for both players, so it made fairly "sound" move decisions). But by the end of the game it would fall apart.

Soon after I got home and got over my bitter sense of defeat, I realized a way to have both evaluations... Choose an evaluator at the beginning of the tree, and use that throughout the tree. So now you don't have conflicting evaluations in any tree, but you can decide by your "current board" what kind of strategy to pursue.

My AI now chooses either an aggresive (mine - 3 * yours) or defensive (3 * mine - yours) evaluation function before calling the Min_Max function. It plays a lot better than before. How much better? I don't know... maybe another tournement is in order.

Ian Harrison

Personally, I'm not particularly interested in whether or not a computer can replace Mozart. I'll be impressed when AI manages to replace Keith Moon. I think that when we talk about human creativity in music, a large part of this creativity is actually the emotional attachment of the composer to the piece. An artificial composer may be able to play the drums like Keith Moon, but would that same program have the inclination to knock the drums over at the end of the set? Much of the beauty we see in art is not in the finished product so much as it is in the process that led to the creation of the finished product. Our empathy with the human artist and our ability to speculate on an artists emotional state while they were creating a piece allow us to enjoy art. AI generated art of any kind, though it may be technically sound, currently lacks this empathy. Ray Kurzweil has written a program that writes poetry, and it seems like pretty good poetry, except that there is nothing behind it. With Sylvia Plath or Coleridge or any other great human poet, knowing the story of the person behind the poetry is necessary to appreciate the poetry, and allows us to appreciate it on a higher level. With a machine, the poetry is just a patterned arrangement of words. There is no human connection. A truly intelligent composer would have to not just create the music, but feel it.\ N'Sync is catchy? Um, yeah, right.

Nick Kerr

tim = sillymouse

he thinks computer composers are awesome. i doubt that's even possible. whatever.

Ananya Misra

I don't know much about music composition, so I thought of that question in terms of writing, instead. I took a fiction writing class last year, and it was serious work. We had to revise, and revise (and the piece was still never flawless). We had to make sure the characters were well-defined and had internally consistent actions, the plot arose from the characters, the stories were meaningful, the language was appropriate, there were no cliches...and important things were shown, not told. I don't see how a computer could follow all these instructions and produce an interesting story unless the humans behind it did a huge amount of work, both creative and tedious. Does anyone else?

This reminds me of the "play" in the video we saw at the beginning of the semester. But most of the creative work there was done by humans, and it would probably not win a literary award, right?

In any case, I searched for computer-generated writing and found mostly humorous links. Here are two. The second one is funnier, I think. www.evolutionzone.com/kulturezone/c-g.writing/ familygames.com/features/humor.html

Jamie Racanelli

The only music related to computers that will ever be truly awesome is music that is created by humans through the use of computers. People are making some amazing stuff that is entirely digital, involves no humans playing instruments directly, but it is people who are making it. I think that computers can't compete with humans because not all composition is formulaic. There can be a great deal of methodology and science to writing songs, but there isn't always. I have been writing songs for about 8 years now, and I don't know anything about any sort of music theory or what not; I have no formal training. And I know and have played with many other musicians who similarly just write songs by feeling them out. One might argue that there is a method to our madness, we just don't know what it is. But even if someone could look at this unformal music and find out some sort of formula behind it that they could feed into a computer, humans would still be one step ahead because they could just look at the formula and change it up here and there. I guess computers could do that to, but people would do it first. So if computers do become great composers, then they will never drive people out of music. They may just push people towards really spastic, dischordant stuff, like wicked-ninja-acid-free jazz or something. Humans will always somehow be slightly better than computers is my point, because we'll do it first, like Elvis, and we'll never run out of new things to do. And I'll be damned if any computer music master is gonna keep me down. Fight the tyranny I say, Fight it.

Juan Ramos

Tim, I like your argument about how computers might evolve to the point where they may replace human composers in films. Though I think this is true, I believe that this would be more of an option for films that do not wholly depend on music to really create an atmosphere and engross the viewer into the film. In my experience, I haven't heard a computer synthesizing a whole orchestral suite, full of violins, pianos, drums, etc, plus a choir. I know I'd pick the 'human' version of the Imperial March from Star Wars over the MIDI of it I just think that as it stands now, computers lack the emphasis and the 'umph' to back up a film that uses music to its advantage.

The same applies with music in general. Ian points out that emotion is a big part of music, and I couldn't agree more. The Friday before break, in Haverford's La Fiesta, I could care less if my feet hurt from so much dancing. Why? Because the music was just begging to be danced and followed. Although a computer might be able to compose a Merengue or a Salsa, I'm skeptical in believing that such a song could echo the cultural roots and the emotional attachments of a piece of music composed by, say, Celia Cruz or the late Tito Puente.

I will admit, though, that this topic seems to hearken closely with the question of whether computers are intelligent by human standards. Mabye we just haven't gotten to the level of complexity necessary to make computers replace human composers. It could be that a few years from now a computer would be able to compose the score of the next Lord of the Rings and fool the audience into thinking it was done by Howard Shore's orchestra. Or maybe in a few years the La Fiesta's will have a computer onstage instead of a band (though I HIGHLY doubt that'll happen!)

Matt Rushton

First the konane tournament was fun, though I wish I would have done a bit better. Even after playing though I'm still not sure what the best static evaluator would have been. Ananya's seemed solid and amazingly fast, and I do like Jason's idea of an offensive and defensive player, with the computer being able to switch between the two during the run of play. That seems like it could be a pretty good strategy. It will be interesting to see how much better the AI gets when people implement a learning algorithm along with their static evaluators. We'll see. My final project is coming along, I've been working on it a bunch the last couple days. I'm feeling pretty confident that it will be pretty accurate, but I'm still crossing my fingers. So far the biggest problem has been all the tedious typing, of stats and their manipulations. As far as the whole music issue goes that everyone is talking about I don't really know enough about music to have too much of an opinion, but it does seem like computers can create some decent sounding pieces. And for people like me, we probabaly wouldn't even be able to tell the difference musically. However, I feel like people would never listen to it, knowing it was artificial. Of course big movie studios could save some money, but would it be worth it. In the future they will probably not need actors and actresses anymore either and just use computer generated ones who look beautiful, and never mess up their lines. Yet, I feel as though this will never happen, for all the obvious reasons people have been discussing.

Tina Tan

I think with computer generated music, not only will people always continue to compose it, but people will also continue to support these composers. The technology that allows computers to write westerns like in the video has been around for years, and it hasn’t come to the point yet where all or even most of our plays now being written by computers. People will keep creating art forever, especially those artists whose first concern was the sake of expression and not profit. Maybe there will be a point where using computer generated scores in movies and recordings will be cheaper but for those true artists that won’t stop them. Like Matt, I probably won’t be able to tell the difference between music written by a human or a computer but I know I would appreciate the music much more if I knew that it was written by a person. I think of music being created now, and people have always given much more respect to artists that compose their own music and write their own lyrics than those that have been produced by a studio team. I don’t think appreciation for expression will ever die out.