I spent half a year doing (very) part time gig work “training” AI, and while I’ve written about it in previous posts and remarked vaguely on what I was doing for that work, I haven’t focused in on the grit of how much water that word carries around such work: training.

I make the rounds of the various commentary that shows up as grassroots reporting on places like Youtube, so recently watching some commentary from “Eli the Computer Guy”^[1] sparked some connections between other dots in my head on this topic.

Strictly speaking I was doing this thing called “data annotation” in the aforementioned role, and if you had images of me-as-instructor, tutoring an AI in the lessons of human literature and history I would suggest that this is (a) likely intentional by (or at least a happy by-product of the marking effort of) the part of large AI companies and (b) so far from the truth as to be laughable. 

In the video, Eli was talking about how we label “data conversion” as training, and remarks (to paraphrase) that it paves the way for all manner of sins and ethical or legal violations. He says something like, if one was to create a new kind of file type, like a better version of PDF, and then scraped the entire contents of human writing into this format just because it’s a new technology, you would almost instantly be hit with a vast array of copyright lawsuits.

But AI is doing basically this same thing. From a technical perspective, an AI model is (over-simplified, sure, but) just a big, complex database. It has scraped the entire contents of human writing into a unique format just because it’s a new technology, and simultaneously we’ve anthropomorphized this action with verbs like “training” and “learning”—as much marketing gibberish as technical terms, and you know this because the companies like the one I worked for don’t actually call it that, they call it “annotation”—and a lot of people collectively shrug at the ethical and legal steamrolling.

But AI isn’t a people—it’s a tool and a technology made by people, and those people have broken our laws and told us their systems are just going to school so that we don’t begrudge content theft disguised as education.