[LAU] Google Magenta project's first composition

david gnome at hawaii.rr.com
Sun Jun 5 07:49:43 UTC 2016

On 06/04/2016 07:12 PM, Erik Steffl wrote:
> On 06/04/2016 12:21 PM, Ralf Mardorf wrote:
>> On Sat, 4 Jun 2016 05:09:02 -0400, tom haddington wrote:
>>> One might observe that the machine wrote bad music.  Well, humans are
>>> already doing that, too, so Magenta has gotten at least that far!  As
>>> with chess machines, it may be a matter of time.
>> The point isn't, if a machine is able to fake music, it doesn't
>> matter, if it's good or bad faked music. What the machine generates is
>> completely uninteresting to me, since a machine has got no
>> emotions I'm interested in. A machine has got no emotions at all, so
>> even if the machine would generate "good music", it would be faked "good
>> music", emotional fraud. Human impostors are able to e.g. fake love.
>> Victims often feel more loved by an impostor, than by somebody who
>> really loves them. Fraud could make us feel good, we anyway
>> dislike fraud. That just shows what kind of company Google is. A human
>> might be an untalented musician, but at least a human usually has got
>> real emotions. A machine that is able to fake "good music" has got
>> absolutely nothing to do with progress. It's a damage. Developing
>> something like this shows the unimaginativeness of the developers.
>> Nobody needs it, it's good for absolutely nothing and even not a
>> useful step to learn something for useful AI projects or something
>> like this.
>   great emotional impact on audience does not require great emotional
> investment of performer/author. Think of ocean or sunset or flowers - no
> emotions but they are beautiful/impactful. (you might now change
> argument to what's natural or not but that's a separate argument, only
> pointing out the emotions here)

Interestingly enough, color and music don't objectively exist. They are 
both processed perceptions. Color is heavily processed, even before it 
actually reaches the brain, by the optic nerve. Sound, on the other 
hand, goes from the nerve cells (connected to hair cells) directly into 
the brain, where parts of the brain analyze it into frequency components 
while other parts analyze it into temporal components while other parts 
analyze its timbre while other parts analyze the apparent spatial 
location of the source, etc. Even our old reptilian brain is involved, 
identifying emotional elements. (After all, one function of our 
marvelous audio processing system in the past was to separate out the 
single sound made by a tiger coming up behind you, and make you run like 
hell before it could pounce!) All of this is kept in memory and analyzed 
in real time by other parts of the brain as the brain builds an internal 
model of the music, constantly predicts what will come next, and gets a 
pleasure jolt every time its prediction proves correct. It also gets a 
pleasure jolt when its prediction is wrong but the brain successfully 
reworks its model of the music to achieve another correct prediction. 
(Damn, music processing is one of the most complex things brains do!)

All done so that we can take in a barrage of sound and separate it into 
the bass playing this, the keyboard playing that, the rhythm guitar 
strumming this way, the singer singing a line and the lead guitar 
playing a short riff at the end of each line - while being aware from 
how she sounds that the hot chick beside you at the concert is really 
getting turned on ...

David W. Jones
gnome at hawaii.rr.com
authenticity, honesty, community

More information about the Linux-audio-user mailing list