Comment by l33tman
3 years ago
You're wrong in your concept of how AI/ML works. Even trivial 1980's neural networks generalize, it's the whole point of AI/ML or you'd just have a lookup-table (or, as you put it, something that copies and pastes images together).
I've seen "infographics" spread by anti-AI people (or just attention-seekers) on Twitter that tries to "explain" that AI image generators blend together existing images, which is simply not true..
It is however the case that different AI models (and the brain) generalize a bit differently. That is probably the case between different humans too. Not the least with for example like you say those with photographic memory, autists etc.
What you call creativity in humans is just noise in combination with a boatload of exposure to multi-modal training data. Both aspects are already in the modern diffusion models. I would however ascribe a big edge in humans to what you normally call "the creative process" which can be much richer, like a process where you figure out what you lack to produce a work, go out and learn something new and specific, talk with your peers, listen to more noise.. stuff like that seems (currently) more difficult for AIs, though I guess plugins that do more iterative stuff like chatgpt's new plugins will appear in media generators as well eventually..
ML generalization and human abstraction are very different beasts.
For example, a human artist would have an understanding of how line weight factors into stylization and why it looks the way it does and be able to accurately apply these concepts to drawings of things they’ve never seen in that style (or even seen at all, if it’s of something imaginary).
The best an ML model can do is mimic examples of line art in the given style within its training data, the product of which will contain errors due to not understanding the underlying principles, especially if you ask it to draw something it hasn’t seen in the style you’re asking for. This is why generative AI needs such vast volumes of data to work well; it’s going to falter in cases not well covered by the data. It’s not learning concepts, only statistical probabilities.
I know what you're saying, and for sure existing models can be difficult to force into the really weird corners of the distributions (or go outside the distributions). The text interfaces are partially to blame for this though, you can take the images into Gimp and do some crude naive modifications and bring them back and the model will usually happily complete the "out-of-distribution" ideas. The Stable Diffusion toolboxes have evolved far away from the original simple text2image interfaces that midjourney and dalle use.
The models will generalize (because that's the most efficient way of storing concepts) and you can make an argument that that means they understand a concept. Claiming "it's not learning concepts, only statistical probabilities" trivialises what a modern neural network with billions of parameters and dozens of layers is capable of doing. If a model learns how to put a concept like line width 5, 10 and 15 pixels into a continuous internal latent property, you can probably go outside this at inference at least partially.
I would argue that improving this is at this point more about engineering and less about some underlying unreconcilable differences. At the very least we learn a lot about what exactly generalization and learning means.