Article on the BBC website today:

http://www.bbc.co.uk/news/science-environment-18449939

"Professor Armand Leroi from Imperial College London explains why he thinks a Darwinian computer program that can evolve music from noise could kill off the composer"


I'm wondering what music he's come up with.....

There's a science fiction novel (Look To Windward) by Iain M Banks in which artificial intelligences have evolved to the point of being sentient. In it, a famous composer asks one: Could you compose something as well as I could? The AI says that yes, it could, but listeners would know that it had been written by an AI, and that essentially they'd see it as less valid because of that.

I'll try and find the quote when I'm at home.

There was a bit on Toady this morning and it was a bit disappointing. Essentially they're starting with random noise and getting people to select bits which they like and then merging those bits together with some random element, then over many thousands of iterations they've produced something that sounds like a mobile phone ringing.

The algorithmic approach with harmonic rulesets that attempts to produce chorales and fugues seems to be far more successful.

The more I think about this, the more I think that human composers aren't in any danger of going anywhere.

The point of human achievement is more than just to be able to perform tasks. It's to push the boundaries of what humans can do. If you think about it from the point of view of just getting somewhere fast enough on either foot or in water, or throwing things, then the entire Olympics is pointless (some of us may think it's pointless anyway but that's not the argument here ). Any automobile can drive 100m faster than anyone could run it, and machines can lift any number of things heavier than a human ever will.

Besides which, I work in computers - for years people have been saying, computers can write novels, understand language, etc etc etc - well they can simulate it but they certainly aren't that close to being able to do true creative stuff. I have absolutely no doubt that a computer program can put things together to make a tune that sounds pleasant - it might even do better than some of the generic pop songs these days - but can it do things like use the particular technique of hand-stopping glissando at the end of the Harbison horn trio that gives it such a plaintive sound?

Maybe one day, but I don't think that the BBC article shows that that day is coming anytime soon. #

I did find the relevant section of Look To Windward, but it's too long to post here.

I certainly don't think composers are at too much risk. "Genetic algorithms", which I think is what this is, have been quite trendy for about 2 decades now. Basically it's playing Battleships: you start with a few stabs at random, then you pick the ones that seem most successful, and put more guesses in their vicinity. There are two clever steps: how you define "in the vicinity" (i.e. how do you select new points close to the ones that look good), and how you define "good". In this instance they seem to have used humans to assess "goodness" of their music, which means the product is still, actually, the product of a human, not the product of a computer.

I'd be very impressed if they'd come up with a quantitative, objective measure of "goodness" of music; many have tried, and I'm not aware of any succeeding, but maybe someone here knows a success story?

It's always hard to assess whether a report on something is actually slightly disappointing science, or fairly unimpressive reporting of science. One marker is the way an article ends. This one ends with the classic statments of a scientist: (1) give us more money and a bigger computer and then it will really work; (2) but we're not promising anything.

TV and newspapers have this strange habit of taking one person's opinion, overstating the case, and making controversial national news about it. What you have to remember is that their most important objective is to deliver eyes to advertisers. It is certainly not to apply intelligence, logic, and experience to their choice of news and its interpretation.

Generally, once you learn to stop taking the "news" too seriously, or accepting it uncritically, then it is all very amusing. But it gets dangerous when they give credence to cranks and charlatans or start to spread baseless health scare stories.

Barry-clari ATCL, from Clari Towers, South East London explains that a Darwinian computer program will produce music with no soul, and therefore it cannot in any way replace human composers.

It'll be alright for organists though, seeing as their instruments have no soul...
*runs*

I guess I'll just have to wait until I become a superior being like VH2, and I can find the news merely "amusing" from my ivory tower...

Computers, evolution and music - My three favourite topics

I hope this experiment helps us understand the origins and evolution of music and language. After all, music itself is an evolutionary phenotype of humans. Ultimately, this is about understanding ourselves.

I think it's a very neat idea to use natural selection to create music. This is one of the few cases where a machine could pass "Turing Test". The other example that come to my mind is the the "painting computer" I saw on BBC the other day. I have to admit that it WAS artistic, much more than many human artists.



If I have a say, I would have the program modified to give it "ears". It should listen to currently "trending" (annoyingly common word these days) music and incorporate elements of it. Then, the program will, hopefully, be able to learn various techniques such as glissando.



I don't think you can call the output as a human product. It's not produced by a single person but chosen by a sizeable population over many generations. The human volunteers are mere consumers and not producers. Theoretically, this experiment could have used whales or dolphins (only if we could teach them to "rate" the music). The result would be very different.

Ah, but the best organists DO give their instruments a soul...

I don't think I've read anything more to the point than that for a long time!! Well said VH2