Post seems very biased towards the now. Stable diffusion et al are very successful with a certain technique but it is foolish to think that is a method which will simply be improved indefinitely.
Generative "AI" will take many forms. Ultimately it will likely remove much of the "technique" element to creation, depriving artists and content owners of income and relevance.
Will this happen overnight? No. I suspect over the next, say, decade, AI will be a beneficial tool more than a threat.
At some point, I expect generative AI to become multi-sensory(sight, sound, touch). Such systems will work from physical models of subjects/environments to produce novel and accurate representations based on rich descriptions and deep contextual awareness of culture. These systems will not think in pixels but in objects and relationships which are then simulated, rendered and filtered to match the desires of the users.
I do applaud efforts of the writers and actors to protect themselves from competition but I believe it will ultimately be in vain. It will be interesting to watch the legal developments in this space. It may be necessary for future generative systems to provide an audit trail showing how they gained an understanding of the world to prove no unauthorized training was performed. This merely raises the bar slightly and does not prevent future generative systems from deriving important relationships via other means, such as 'clean room', high-level descriptions being given(perhaps by other automated processes).
For example, while it may be illegal to train an AI to reproduce Harrison Ford using his copyrighted works or even images captured in a public space, I can reduce Harrison Ford to a set of characteristics which can be passed to a generative system to produce something indistinguishable from the real Harrison Ford. If I am able to document this procedure I see few ways for the legal system to prevent it but then again I am no expert in this area.
For what it's worth, I'm not a fan of current "AI". I have found LLMs to be particularly unreliable and mostly useless. I also find most "AI" generated art to be either boring, inaccurate, or in some way not compelling. That said, I think the trend is becoming clearer.
Generative "AI" will take many forms. Ultimately it will likely remove much of the "technique" element to creation, depriving artists and content owners of income and relevance.
Will this happen overnight? No. I suspect over the next, say, decade, AI will be a beneficial tool more than a threat.
At some point, I expect generative AI to become multi-sensory(sight, sound, touch). Such systems will work from physical models of subjects/environments to produce novel and accurate representations based on rich descriptions and deep contextual awareness of culture. These systems will not think in pixels but in objects and relationships which are then simulated, rendered and filtered to match the desires of the users.
I do applaud efforts of the writers and actors to protect themselves from competition but I believe it will ultimately be in vain. It will be interesting to watch the legal developments in this space. It may be necessary for future generative systems to provide an audit trail showing how they gained an understanding of the world to prove no unauthorized training was performed. This merely raises the bar slightly and does not prevent future generative systems from deriving important relationships via other means, such as 'clean room', high-level descriptions being given(perhaps by other automated processes).
For example, while it may be illegal to train an AI to reproduce Harrison Ford using his copyrighted works or even images captured in a public space, I can reduce Harrison Ford to a set of characteristics which can be passed to a generative system to produce something indistinguishable from the real Harrison Ford. If I am able to document this procedure I see few ways for the legal system to prevent it but then again I am no expert in this area.
For what it's worth, I'm not a fan of current "AI". I have found LLMs to be particularly unreliable and mostly useless. I also find most "AI" generated art to be either boring, inaccurate, or in some way not compelling. That said, I think the trend is becoming clearer.