My test of DALL-E produced crap.
The image above was the best result of 16 attempts, each with 4 variations. This one was in the very first set of 4. The prompt for this one was “space station with stationary long axis, docking ports at both ends. Large circular habitation wheel in middle, flanked on each side by a smaller wheel”. I simplified and reworded the prompt a few times, and each iteration was worse. Most mounted the wheels transversely, or at the ends of up to 3 arms.
It would be interesting to trace back the origins of the source materials. I wonder how many human artists had their creations stolen to produce this. All the passes that DALL-E produced had clear common themes and techniques, ones that Chesley Bonestell would have recognised. There is no creativity here, only theft.
But that’s now. What happens from here onward is unknowable. But we can be sure of a few things. First, the visual quality will continue to improve. Second, the responsiveness to the prompts will continue to improve, particularly as the Large Language Models such as ChatGPT become more integrated with engines such as DALL-E. Third, … actually I can’t think of a third positive thing.
I am of the opinion that “machine intelligence” will never come about as a result of current approaches. Humans have achieved intelligence through millions of years’ worth of natural selection. Some of us even claim to be self-conscious. I will grant that a very advanced “AI” might pass as intelligent, but will never be self-conscious. But how is one to know? I think I am self-conscious, and I’m pretty sure some of my family and friends are, but I can never be sure because I can’t experience the world from their perspective. (Ironically, my Spotify stream is currently playing “Daisy” by Karine Polwart, in which a critical line is “some don’t think at all”. The Universe’s serendipity at work?)
If a machine ever becomes both intelligent and conscious, I suspect it will be as a result of a full-brain upload from a human. And that is so far beyond any current technological capabilities as to be ludicrous to expect within many generations of development. That approach would require quantum-level scanning and faithful reproduction of every synapse in our brain, all at the same time. And we’re not even sure if our “consciousness” is an exclusively physical phenomenon.
The other possible route I’ll grant at a low level of likelihood is for researchers to recapitulate the process we went through - millions of generations of increasingly complex AIs, each living in a simulated complex world and learning from previous generations. To a very small extent this is already happening, but still under human control. One the evolution process is solely controlled by the machines themselves, this may change the trajectory. If it does, we’ll only know of their success when they take control of our world. Let’s hope they learn compassion in the process.
Without both intelligence and self-consciousness, DALL-E and its’ ilk will continue to regurgitate materials stolen from human artists. I’ll have to commission my space station design from a human.