Here are some things I do not mean when I say that generative AI is probably here to stay:
- The companies currently boosting generative AI will succeed financially
- I am happy that generative AI is probably here to stay
- We should engage constructively with generative AI
- You should use generative AI
- I am using generative AI
I don't mean any of those things. What I mean when I say that generative AI is probably here to stay, is that I think generative AI is probably here to stay.
To some readers this statement likely seems trite. But I have plenty of friends and acquaintances who won't read it that way. These--all people I like and respect--dislike generative AI for understandable and quite possibly correct reasons.
I don't share the prevalent post-Romantic model of creativity, but I do share the dislike for generative AI. I fear, though, that on this topic many of my friends reason toward what we'd collectively like to happen, not toward what the available evidence suggests.
All of us tend to settle rather than to aspire in some creative fields.
For instance, I have no taste in music. Not bad taste, but no taste: I will listen to whatever is on. Dvořák, Anri, slowed-and-reverb Imagine Dragons, it is all the same to me. I dress badly, eat what is on my plate without comment, and have no more developed views on public sculpture than that I like vaguely realistic statues.
Everyone behaves like this in some creative fields. Everyone behaves like this, I think, in most creative fields. Those who make a witting effort to cultivate taste do so in only a few spheres. We all settle--or satisfice, if you want--so I see little point in getting toffee-nosed about each other's areas of ignorance.
Because we all settle for the good-enough, the good-enough matters more than the excellent. This effect intensifies in any commercial, mass-audience field. In such fields, we must assume an easy majority of the audience lack taste and have no eye for craft. I say this without contempt: again, in most spheres I sit firmly in the part of the audience that lacks taste and has no eye for craft.
This settling, or satisficing, gives us the tyranny of good-enough.
I don't know whether generative AI can, with enough iterative prompting, create excellent work. But excellent work doesn't matter. Good-enough work matters, and for better or worse generative AI makes work that's good enough for many.
When it comes to literary success, for instance, the question is not whether a LLM can write like George Eliot. Few people read George Eliot (more should!). The question is whether an LLM can write like the authors people do read: otherworld light novels (award winners among them), Mills & Boon, popular science and history with the 'buckle-up' tone of the airport bookstand. LLMs can write like those authors, whether or not they can match Eliot.
When it comes to visual art, generative AI doesn't have to compete with Velázquez; it has to compete with stock photographers and guys taking pin-up commissions--painting's session musicians.
And so on and so forth, field to field.
We have all stumbled here and there on people who resist generative AI production in their own field but lean on it elsewhere: human authors reaching for generated images, for example. Such people should think on the tyranny of good-enough: how sure do they feel that their field is special, insulated--not from generative AI, directly, but from our universal tastelessness?
Maybe a crash approaches for the big companies doing ballyhoo right now. A market correction of some kind at least seems likely to me, in my ignorance. Postulating many more paying users than Microsoft Office has, globally? Surely unrealistic. Some past technologies found mass adoption while sinking the companies that first pushed them, after all.
However, a crash won't stuff the genie back in the bottle. If it comes, it might slow, stall, or reverse the datacentre build-out. But workable generative AI models run happily on local, comparatively low-powered hardware.
We may or may not wind up in a world of coding agents profitably rented out by big AI specialists. I suspect we will remain in a world where the average person uses generative AI to chat to you on a dating app or write an application for a city small-grants scheme. Generative AI may never become economically dominant, but it will pervade and persist.
If you strongly feel that all this is an insult to life itself, I suggest making some plans to preserve life itself when the machines have won.