AI-generated Art, part 3:

Beyond Art—Will AI Become Conscious or Forever Remain Inscrutable?

I read widely about the topics that interest me—Nautilus, Quanta Magazine and New Scientist are good sources—and then try incorporate what I’ve learned into my fiction (novels and short stories).

Also, I’m an engineer by both training and nature; that is, if something isn’t working I try to find out why and then fix it (if possible, of course). And as a science fiction writer I try to see what might happen in the future, often leading me to extrapolate how it could be done right (not that necessarily happens, but as an upbeat SF writer you have to dare to be wrong).

That’s why, for this particular topic of AI—both as art generator and in general—I not only try to follow (as much as is doable) when it becomes trending, but also try to dig deeper. How do DALL-E2, Stable Diffusion and ChatGPT work? And particularly in ChatGPT’s case, how much are they corrected by their human programmers? And why do they have certain shortcomings, and is there a way to fix those shortcomings?

As such, I see the following major shortcomings of current AI:

• Lack of agency

• Lack of creativity and/or lateral thinking;

• Lack of an ingrained sense of narrative (story);

• Lack of consciousness and its predecessors sentience and self-awareness;

  • (not necessarily feelings, as feeling are evolutionary shortcuts);

There is one thing art-generating AIs have in common, namely that they are all run like neural networks. Not just art-generating AIs such as the above-mentioned DALL-E2, Stable Diffusion, ChatGPT and their ilk, but anywhere from speech recognition, language translations, chess- and Go-playing robots, face recognition, medical diagnosis, and self-driving cars¹. Such neural networks have advantages—the more specific, the more tailor-made, the better they work²—and disadvantages—outside the specific environment, their solutions are brittle, and they’e inscrutable. More on that under the “Creativity” sub-chapter.

---