i've been proven wrong:
http://www.guardian.co.uk/arts/features ... 51,00.html

this may have already been covered at some point. sorry for the redundancy if so.

Meh.

Who cares?

I have not seen this article before, but I have been theorizing that this is how they do things for some time...

Har-de-har campers! I linked to the article in another thread almost at the same time you posted this! We're either psychically bonded or we read the same paper. The latter's WAY too out there to even consider as an option.
This WILL be the funniest thing EVER said about the software:

Throw enough hits at a wall and some of them will stick

Great, now HAL can pick the next pop turd to float to the surface of the music industry toilet.
I would love to plug in some Wolf Eyes, Lightning Bolt, Sunn0))), etc. into this so called "Hit Song Science" Machine and see where they graph out at and what kind of "score" they garner.

There is actually something interesting about this.

It implies that the record companies consider that a record must be subjectively GOOD in order to sell units.

I would have thought that was CLEARLY not the case.

Do people buy Britney records because they are actually good? Is it not more about hype and marketing and manufactured POP-MYSTERY?

The creators of this software (and record companies) would have you believe that pop-buyers are fickle and know what they like.

Is that the case, or is it actually the case that they are told what they like?

If the software is based on previous successes (which were based on marketing and hype) then it's predictions are surely without value?

Strange days.

This was investigated to interesting effect almost a decade ago:
http://www.diacenter.org/km/musiccd.html

Champion Rabbit wrote:
If the software is based on previous successes (which were based on marketing and hype) then it's predictions are surely without value?

If you had enough computing power and were evil enough, it should be possible to use a massively parallel array to derive a non-averaged prototype 'hit' just by exposing the system to large numbers of training examples of previous hit and 'non hit' songs, normalised for context (hype, payola etc). I'm pretty sure that this is still too complex - even for the Pentagon - and they'll have used a series of algorithms. I'm also assuming that there is some control over genre / production style, rather than just whether it's a good song, in which case the outcome for us as listeners will be no different than some ponytailed coke fiend in LA making the decisions. Just what the world needs: An A&R robot.

YO31 wrote:

Champion Rabbit wrote:
If the software is based on previous successes (which were based on marketing and hype) then it's predictions are surely without value?

If you had enough computing power and were evil enough, it should be possible to use a massively parallel array to derive a non-averaged prototype 'hit' just by exposing the system to large numbers of training examples of previous hit and 'non hit' songs, normalised for context (hype, payola etc). I'm pretty sure that this is still too complex - even for the Pentagon - and they'll have used a series of algorithms. I'm also assuming that there is some control over genre / production style, rather than just whether it's a good song, in which case the outcome for us as listeners will be no different than some ponytailed coke fiend in LA making the decisions. Just what the world needs: An A&R robot.

You're probably right; I was getting over-excited!

Steve Reich concert tonight!!

Slightly off topic comment: I really cant stand Norah Jones.
Slightly on topic comment: Is anyone really suprised by this computing revelation? Werent we all waiting for the next level of music corruption, whether by computer or not.