The One Thing The Everything Machine Doesn't Know.
Cold comfort is still comfort.
I shouldn’t have been surprised that Robin Sloan wrote the most provocative piece I have yet read about the philosophical question of large language models (hereafter referred to as “the model”). His title question of “Is It Okay?” is approachably elegant, and his answer (maybenotreallypossibly) seems grossly right to me. Or if not right, at least all that can reasonably be concluded at the moment.
It is his distillation of what we are talking about that stuck: the model has everything in it.
“…language models are not merely trained on human writing. They are the writing: all those reasons, granted the ability to speak for themselves. […]
To make this work — you already know this, but I want to underscore it — only a truly rich trove of writing suffices. Train a language model on all of Shakespeare’s works and you won’t get anything useful, just a brittle Shakespeare imitator.
In fact, the only trove known to produce noteworthy capabilities is: the entire internet, or close enough. The whole extant commons of human writing. From here on out, for brevity, we’ll call it Everything.
This is what makes these language models new: there has never, in human history, been a way to operationalize Everything. There’s never been anything close.
And therein lies the whole freakin’ ballgame. You know why your head hurts and you feel like you’ve eaten a bag of Skittles on an empty stomach when you think about AI? This is why. The operationalization of Everything is a rollercoaster on a Ferris Wheel on the Nurburgring of existential vertigo. Even that business suit verb is an MC Escher drawing in sheep’s clothing, because what it means is to be able to turn Everything into Anything.
The math of creativity as we have sort of known it goes something like this. [Individual experience] x [individual personality] = [stuff only that individual can make]. Put extremely crudely, you could not infant swap me with F Scott Fitzgerald and still get so much stuff about gin.
Only the key of me can unlock whatever is behind door number me. Same for you. Imitation, then, is suicide, as Emerson wrote, because it chucks in the rubbish that singular junk drawer of preferences and potentialities that is you.
The Everything Machine is different. It starts with more stuff than you can ever have taken in and can twiddle with it faster and with a wider range of techniques than you can ever hope to master. The sum total of all you have watched or read or heard is but the first word on the first page of the first book in the library of libraries that is in the Everything. And it’s ready to party—all the time, forever.
For someone like me, and for a newsletter like this, the all-singing, all-dancing, Everything Machine’s potential to do humanity’s best act at the speed of electrons is, it must be said, a 10.0 on the Richter Scale of bummers.
Perhaps counter-intuitively though, I don’t fear an Everything Machine cultural supremacy because I’ve already seen what it looks like. Extant math-derived cultures (aka the algorithms that play act as “social media”) are so encouragingly terrible. Entertaining for sure. Finely honed attention cigarettes? To our very peril! But masterpieces of art and learning to topple singular artistic vision? You all know the answer, say it with me now: “hell no.” Gimmicks, memes, clips, “trending sounds” abound, but I’m going to go out on a Banyan tree-sized limb here and predict none of it is going in anybody’s eulogy for Grandma. The algorithms aren’t trying to make anything great: they are trying to serve up just enough to keep you on the phones.
There is one last bulwark against the Everything Machine supplanting human creativity, though what a bitter one it is. Implicit in the “if the Everything Machine can suck up everything we have ever made, then it can reverse engineer what makes us tick” is that “everything ever made” is approximately congruent with what it’s like to be every kind of human, everywhere, forever. But as anyone who has tried to write, paint, build, say or otherwise communicate something, that just isn’t the case:
“Every moment happens twice: inside and outside, and they are two different histories.”
― Zadie Smith, White Teeth
The yawning abyss between what we think and feel and what we then manage to squeeze out for someone to take a swing at understanding might not be the happiest obstacle to being existentially annexed by math, of course, but it a sturdy one. The Everything Machine might soon know what is already knowable. But it can only sift through what we have made so far. It doesn’t know that everything at its disposable is but the leavings, fertile, various, and remarkable though they be, of what it means just to walk around on a regular Tuesday as a person. So it cannot make what it is we really want to make—some arrow that will fly across the great whatever from me to you, and stick there, if only for a moment.