4 ms·

> I see AI not as a replacement, but as a tool – one that opens doors for artists to express themselves in new ways. New ways? Aren't current AI extremely con

by josefx 5y ago

> I see AI not as a replacement, but as a tool – one that opens doors for artists to express themselves in new ways. New ways? Aren't current AI extremely conservative? For example github copilot which simply copy pastes decades old code encoded in its model to complete a text.

tgv 5y ago
Precisely. I'm not interested how to make some theme sound like his previous works, but how he would have created something new. Of course, he might have failed there.
- selfhoster11 5y ago
  I don't think that's fundamentally possible. We just don't have enough information about his creation process to know that.
  - mdp2021 5y ago
    The idea would be to backward engineer such "creation process" to the extent given by tools and abilities. Unfortunately, the enemy to be avoided, "pale imitation", lurks. It depends on the level one manages to reach: in a context of speech, for example (attempting to give the idea), copying the wisdom > copying the semantics > copying the syntax > copying the words. There are foundations grounding what they support, so just identifying patterns (e.g. of words) without creating their "understanding" misses the core. Generating the latter (words, syntax) without having reached the former (wisdom, semantics) will mean obtaining "rambling" which may suggest the style of some author - it would be creating an idiot impersonating some real figure, an "impersonating idiot" like any fool attempting to disguise as Churchill (or whoever). In theory, elements of the «creation process» could be entailed. In practice, that must be a goal, and it presumably requires a much higher understanding of those methods to be feasible.
    - tgv 5y ago
      It totally would, and it would also require an insight into how (in this case) Beethoven found new ways of expression, even though he already could draw on so many. In other words, it would require strong AI, and then particular knowledge about Beethoven's thought processes, in detail.
      - mdp2021 5y ago
        Yes, but even sticking with a selected amount of material - not in the direction of "simulating the full Beethoven person", but in that of "obtaining knowledge [as if tentative scientific laws - informational compression from observation to underlying structures which can be exported and productively reapplied] to be squeezed out of some material", there is a dramatic difference between identifying "apparent superficial patterns" and identifying "the foundations of such patterns". Those "laws" are meant to be "the structures of reality", not the "structures of phenomena": similarly for understanding, which is not "how this behaves" but (like in science) "why this behaves", using "how" (the observation) to deduce "why". Any rambling may superficially make sense - and to get that superficial patterns are enough -, while underlying understanding (as a broad and powerful category) is required to actually build sense. To continue the example, one could impersonate Churchill by appropriating the terms and mannerism, or by deducing his ideas - "think like Churchill" is not "sound like Churchill". The idea of "intelligence" is that "intus legere", which explicitly excludes reading the surface where there is no "into". Otherwise, identifying superficial patterns is "Artificial Skimming", AS. It is far from impossible to deduce deep thought structures from limited material, but it must be the explicit goal.

by josefx 5y ago

tgv 5y ago

Precisely. I'm not interested how to make some theme sound like his previous works, but how he would have created something new. Of course, he might have failed there.

selfhoster11 5y ago
I don't think that's fundamentally possible. We just don't have enough information about his creation process to know that.
- mdp2021 5y ago
  The idea would be to backward engineer such "creation process" to the extent given by tools and abilities. Unfortunately, the enemy to be avoided, "pale imitation", lurks. It depends on the level one manages to reach: in a context of speech, for example (attempting to give the idea), copying the wisdom > copying the semantics > copying the syntax > copying the words. There are foundations grounding what they support, so just identifying patterns (e.g. of words) without creating their "understanding" misses the core. Generating the latter (words, syntax) without having reached the former (wisdom, semantics) will mean obtaining "rambling" which may suggest the style of some author - it would be creating an idiot impersonating some real figure, an "impersonating idiot" like any fool attempting to disguise as Churchill (or whoever). In theory, elements of the «creation process» could be entailed. In practice, that must be a goal, and it presumably requires a much higher understanding of those methods to be feasible.
  - tgv 5y ago
    It totally would, and it would also require an insight into how (in this case) Beethoven found new ways of expression, even though he already could draw on so many. In other words, it would require strong AI, and then particular knowledge about Beethoven's thought processes, in detail.
    - mdp2021 5y ago
      Yes, but even sticking with a selected amount of material - not in the direction of "simulating the full Beethoven person", but in that of "obtaining knowledge [as if tentative scientific laws - informational compression from observation to underlying structures which can be exported and productively reapplied] to be squeezed out of some material", there is a dramatic difference between identifying "apparent superficial patterns" and identifying "the foundations of such patterns". Those "laws" are meant to be "the structures of reality", not the "structures of phenomena": similarly for understanding, which is not "how this behaves" but (like in science) "why this behaves", using "how" (the observation) to deduce "why". Any rambling may superficially make sense - and to get that superficial patterns are enough -, while underlying understanding (as a broad and powerful category) is required to actually build sense. To continue the example, one could impersonate Churchill by appropriating the terms and mannerism, or by deducing his ideas - "think like Churchill" is not "sound like Churchill". The idea of "intelligence" is that "intus legere", which explicitly excludes reading the surface where there is no "into". Otherwise, identifying superficial patterns is "Artificial Skimming", AS. It is far from impossible to deduce deep thought structures from limited material, but it must be the explicit goal.