Just as there's an aesthetic response to specific mathematics and specific code — not just to mathematics or code as such — there's an aesthetic response to specific AI designs. I don't know of any terminology for it, but
- There's a form of brutalism in the scaling up of data sets and parameter counts, despite the elegance of improvements in training methods...
- ...that's very different from the sheer abstract elegance of something like AlphaZero, the data irrelevant as the network unfolds itself from the minimalistic seed of the rules of the game...
- ... which is also very different from the organic feeling of using reinforcement learning to train an agent against an external environment or simulator.
There are overlaps, of course, and sub-styles, facets of skill (are you stacking layers just because you can, or is there some deep domain intuition guiding your architecture?) and even issues of originality - the aesthetics of the nth application of the same architecture to a very similar data set is very different from a bold new approach. The new optimizer feels different than the new way to reparametrize data (at least to me, but they probably feel different to you too, although neither might feel to same to us).
And there are even deeper and wider divides: how to account for the difference between a sparse and small Bayesian network and a dense and large deep learning model attempting to model the same system? One would be tempted to call the former a limit case of poetry and the latter the most prolix essay, but how come this poetry is then more understandable than the essay?
Speaking of essays, this isn't one; it's at beat a ramble. The differences I mention are technical, mathematical, and pragmatic, but what I'm pointing towards is that there are also differences in how I feel learning about, building, or using these different sorts of models: learning about AlphaZero (and its older, less successful predecessors) I felt a happy thrill similar to an illuminating mathematical proof, while GPT-3 feels more like the Ulysses, and not just because of the obvious jokes one could make.
Is any of this relevant to the industry? Not to any business application or technical decision, no, but the edge where technology ends and the possibility of application begins is the lone person at the intersection of the idea and the problem, and if or as long as this is an activity that requires some degree of creativity, the aesthetics of it may not be irrelevant to our practice.
(There's also a different, sadder question: we speak less now about the aesthetics of code. We write more code in more languages, but it's more transient, more collaborative, in languages and libraries we no longer have many years to know as intimately as we could. All of that reflects changes in technology and economics; they are probably marks of success, and I wouldn't want to be another Ruskin decrying the loss of crafts when industry improved standards of living so much for so many. But even a loss that's worth the trade remains a loss.)