4 ms·

I am not sure how to say this best, it probably sounds negative but I don’t mean it that way. It is really interesting because using these tools abstracts away

by j4yav 5y ago

I am not sure how to say this best, it probably sounds negative but I don’t mean it that way. It is really interesting because using these tools abstracts away the talent/skills part of creating art - you say your idea, and get the computer to keep showing you completed pieces until you find what you meant, or at least something you found interesting. It reminds me a bit of the role of the DJ to do the searching and curating, but not actually create the music. In fact you could imagine a musical version of this where the operator is describing feelings or styles and the AI is generating music to match where the operator wants to go.

suby 5y ago
You're right that it removes some (honestly perhaps most) of the craftsmanship that comes with creating art. I do think the resulting work is still interesting, valuable, and art from a practical point of view, though. There is perhaps a threshold where this is not true, perhaps related to how much curation/pruning/how indiscriminate the human is in determining the final outcome. That is to say, if there is some text to image generation program, and the human generates 1 image based on the first phrase they thought of, that image they generated is probably not going to be very interesting, at least based on the image generation stuff I've played around with so far. It often takes time and several attempts / variations on the phrasing to get something which is interesting and novel. People have been finding tricks like appending the words "Unreal Engine" onto the phrase in order to produce certain types of results. Some people actually go into the code to tweak things to get desired results. I do wonder how true this will be going forward. It feels like this media synthesis stuff is progressing at the speed of light, and perhaps it'll only take 1 iteration / attempt to craft the interesting and novel stuff that we're looking for. So at least at the moment it's maybe not so different from anything else. I think of it as exploring the possibilities in a given space. I work on video game development, and it's the same story there. If you're creating a game prototype, you might have an initial idea which sparks development, but once the initial seed is implemented the game often takes on and is guided by a life of it's own, shaped by the possibilities in that given space which you are narrowing down into a subset which seem to work and gel together. That's how I view it anyway.
- kingsuper20 5y ago
  >You're right that it removes some (honestly perhaps most) of the craftsmanship that comes with creating art. Perhaps art, going forward, is simply described as the process, not the result. I don't doubt that a computer could (or will) produce a perfectly convincing Charlie Parker solo or Richard M. Powers painting. There really was less there than met the eye (or ear) and most of human output consists of cliches in any case. My take is to become increasingly archaic and self-sufficient in interests. Greek red figure pottery looks like a good place to end up after the walls are filled with amateur art, but a kiln looks pricey.
  - germinalphrase 5y ago
    ‘Beautiful thing’ will indeed be massively available (as it is right now), but ‘beautiful thing touched by a human’ will always remain scarce. Only the latter will be truly considered an art object.
    - kingsuper20 5y ago
      In that case, the value of a thing is not it's physical reality but it's provenance. I guess it's no different than the high value given to a lot of 20th C. art, it's just another form of marketing. Maybe this bodes well for antique prices.
hutzlibu 5y ago
" It reminds me a bit of the role of the DJ to do the searching and curating, but not actually create the music" It very much depends on the type of DJ. But a good DJ performing live, is indeed generating music, as he feels the vibes of the crowd and mixes in the exact tracks fitting to that moment. Not predefined track after track, but they blend in and over on each other. If you do this with parts of a song from here and then there, with a small snippet from something else - it is indeed generating new music in my book. And I think alien dreams and ai generated art alike ... can move into that direction. Generating art, by knowing the various tools and combining them.
- j4yav 5y ago
  I am a DJ myself, but I would still give credit to the authors of the tracks rather than claiming it for myself (even when it’s a unique interesting combination or flow).
  - hutzlibu 5y ago
    Sure. You build up on existing art. Like is done here with the generated art, it probably does not make much sense for one person to claim he "created" and owns this image. Would be hard to judge, how much credit goes to the programmers, the basic AI researchers, all the people who created the input on the AI to learn ...
writeslowly 5y ago
To me it feels like AI-assisted divination. It's like staring into clouds of smoke or a crystal ball until you find something that looks significant, but using a computer algorithm instead. Divination has been around throughout human history, so it makes sense that it would emerge again through the use of machine learning algorithms.
- api 5y ago
  https://gizmodo.com/an-occult-history-of-the-television-set-1521354168 https://gizmodo.com/an-occult-history-of-the-television-set-... The spiritualist movement and similar late 19th / early 20th century occultism can easily be re-interpreted as oblique science fiction... or vice versa. The whole industrial revolution is at least culturally and historically inseparable from the occult revival in the sense that all that "natural philosophy" stuff was the primordial soup of ideas from which modern science emerged. Methodological scientific empiricism is the gold that was left in the crucible, and it yields magic that actually works. ... now back to writing obscure runes in an alien language to instruct the daemons in the ether ...
songeater 5y ago
>> In fact you could imagine a musical version of this where the operator is describing feelings or styles and the AI is generating music to match where the operator wants to go. OpenAI already did this with Jukebox! I'm just surprised more people have not used it as an "instrument" vs as a novelty. shameless plug: https://soundcloud.com/songshtr/sets/2106-ep https://soundcloud.com/songshtr/sets/2106-ep
- smthngwitty 5y ago
  One downside with Jukebox: it takes ~9 hours to generate ~90 seconds of audio (even on a NVIDIA V100 GPU) since it's an auto-regressive model making experimentation and 'co-creation' much harder
  - songeater 5y ago
    agreed. full waveform "music" (unlike say MIDI) is just many more variables than a "picture" or even "video". Also you have to "stitch together" a lot of these samples to get anything that resembles a "new" song. More akin to mining... but still its kind of crazy what can be done with it.
    - echelon 5y ago
      I'm not familiar with Jukebox, so maybe I'm wrong. It seems like you'd want a fast architecture to use a model to generate sheet music / MIDI, then layer another model on top to create instrumentals?
    - songeater 5y ago
      >> Why doesn't the architecture use a model to generate sheet music / MIDI, then layer another model on top to create instrumentals? there are various models that do this [0]... just doesn't have the same power to generate waveform music that jukebox does. Jukebox effectively is a 2 step model a) a 3 layer VQ-VAE compresses the music (so think of this as the MIDI equivalent) and b) a transfomer then learns/generates sequences. The compression is the expensive part of the model. [0] https://towardsdatascience.com/how-to-generate-music-using-a-lstm-neural-network-in-keras-68786834d4c5 https://towardsdatascience.com/how-to-generate-music-using-a...
omgwtfbbq 5y ago
I wonder if its not more similar to electronic music producers. They no longer have to learn how to play instruments and can focus on composing so maybe artists that learn these tools can compose art in a similar way.

by j4yav 5y ago

suby 5y ago

You're right that it removes some (honestly perhaps most) of the craftsmanship that comes with creating art. I do think the resulting work is still interesting, valuable, and art from a practical point of view, though. There is perhaps a threshold where this is not true, perhaps related to how much curation/pruning/how indiscriminate the human is in determining the final outcome. That is to say, if there is some text to image generation program, and the human generates 1 image based on the first phrase they thought of, that image they generated is probably not going to be very interesting, at least based on the image generation stuff I've played around with so far. It often takes time and several attempts / variations on the phrasing to get something which is interesting and novel. People have been finding tricks like appending the words "Unreal Engine" onto the phrase in order to produce certain types of results. Some people actually go into the code to tweak things to get desired results. I do wonder how true this will be going forward. It feels like this media synthesis stuff is progressing at the speed of light, and perhaps it'll only take 1 iteration / attempt to craft the interesting and novel stuff that we're looking for. So at least at the moment it's maybe not so different from anything else. I think of it as exploring the possibilities in a given space. I work on video game development, and it's the same story there. If you're creating a game prototype, you might have an initial idea which sparks development, but once the initial seed is implemented the game often takes on and is guided by a life of it's own, shaped by the possibilities in that given space which you are narrowing down into a subset which seem to work and gel together. That's how I view it anyway.

kingsuper20 5y ago
>You're right that it removes some (honestly perhaps most) of the craftsmanship that comes with creating art. Perhaps art, going forward, is simply described as the process, not the result. I don't doubt that a computer could (or will) produce a perfectly convincing Charlie Parker solo or Richard M. Powers painting. There really was less there than met the eye (or ear) and most of human output consists of cliches in any case. My take is to become increasingly archaic and self-sufficient in interests. Greek red figure pottery looks like a good place to end up after the walls are filled with amateur art, but a kiln looks pricey.
- germinalphrase 5y ago
  ‘Beautiful thing’ will indeed be massively available (as it is right now), but ‘beautiful thing touched by a human’ will always remain scarce. Only the latter will be truly considered an art object.
  - kingsuper20 5y ago
    In that case, the value of a thing is not it's physical reality but it's provenance. I guess it's no different than the high value given to a lot of 20th C. art, it's just another form of marketing. Maybe this bodes well for antique prices.

hutzlibu 5y ago

" It reminds me a bit of the role of the DJ to do the searching and curating, but not actually create the music" It very much depends on the type of DJ. But a good DJ performing live, is indeed generating music, as he feels the vibes of the crowd and mixes in the exact tracks fitting to that moment. Not predefined track after track, but they blend in and over on each other. If you do this with parts of a song from here and then there, with a small snippet from something else - it is indeed generating new music in my book. And I think alien dreams and ai generated art alike ... can move into that direction. Generating art, by knowing the various tools and combining them.

j4yav 5y ago
I am a DJ myself, but I would still give credit to the authors of the tracks rather than claiming it for myself (even when it’s a unique interesting combination or flow).
- hutzlibu 5y ago
  Sure. You build up on existing art. Like is done here with the generated art, it probably does not make much sense for one person to claim he "created" and owns this image. Would be hard to judge, how much credit goes to the programmers, the basic AI researchers, all the people who created the input on the AI to learn ...

writeslowly 5y ago

To me it feels like AI-assisted divination. It's like staring into clouds of smoke or a crystal ball until you find something that looks significant, but using a computer algorithm instead. Divination has been around throughout human history, so it makes sense that it would emerge again through the use of machine learning algorithms.

api 5y ago
https://gizmodo.com/an-occult-history-of-the-television-set-1521354168 https://gizmodo.com/an-occult-history-of-the-television-set-... The spiritualist movement and similar late 19th / early 20th century occultism can easily be re-interpreted as oblique science fiction... or vice versa. The whole industrial revolution is at least culturally and historically inseparable from the occult revival in the sense that all that "natural philosophy" stuff was the primordial soup of ideas from which modern science emerged. Methodological scientific empiricism is the gold that was left in the crucible, and it yields magic that actually works. ... now back to writing obscure runes in an alien language to instruct the daemons in the ether ...

songeater 5y ago

>> In fact you could imagine a musical version of this where the operator is describing feelings or styles and the AI is generating music to match where the operator wants to go. OpenAI already did this with Jukebox! I'm just surprised more people have not used it as an "instrument" vs as a novelty. shameless plug: https://soundcloud.com/songshtr/sets/2106-ep https://soundcloud.com/songshtr/sets/2106-ep

smthngwitty 5y ago
One downside with Jukebox: it takes ~9 hours to generate ~90 seconds of audio (even on a NVIDIA V100 GPU) since it's an auto-regressive model making experimentation and 'co-creation' much harder
- songeater 5y ago
  agreed. full waveform "music" (unlike say MIDI) is just many more variables than a "picture" or even "video". Also you have to "stitch together" a lot of these samples to get anything that resembles a "new" song. More akin to mining... but still its kind of crazy what can be done with it.
  - echelon 5y ago
    I'm not familiar with Jukebox, so maybe I'm wrong. It seems like you'd want a fast architecture to use a model to generate sheet music / MIDI, then layer another model on top to create instrumentals?
  - songeater 5y ago
    >> Why doesn't the architecture use a model to generate sheet music / MIDI, then layer another model on top to create instrumentals? there are various models that do this [0]... just doesn't have the same power to generate waveform music that jukebox does. Jukebox effectively is a 2 step model a) a 3 layer VQ-VAE compresses the music (so think of this as the MIDI equivalent) and b) a transfomer then learns/generates sequences. The compression is the expensive part of the model. [0] https://towardsdatascience.com/how-to-generate-music-using-a-lstm-neural-network-in-keras-68786834d4c5 https://towardsdatascience.com/how-to-generate-music-using-a...

omgwtfbbq 5y ago

I wonder if its not more similar to electronic music producers. They no longer have to learn how to play instruments and can focus on composing so maybe artists that learn these tools can compose art in a similar way.