We've Been Using AI for Twenty Years

Every line of work has a task people would pay never to do again. For visual effects artists it’s roto, tracing the edge of a moving actor frame by frame so they can be cut out of a shot. For accountants it was adding up columns by hand. For film editors it was cutting and taping celluloid with a razor blade. Every time a tool came along that took some of that work away, the people doing it lined up for it. Nobody picketed the first spreadsheet. Nobody mourned the razor blade.

Lately I’ve been poking at the newest roto and keying tools, and they’re a big jump. They also come with a word on the box the old ones never had: AI. That one word turns a welcome upgrade into a threat. Same work, same relief, and now the people it’s helping are being told it’s coming for them.

Which is strange, because we’ve been living with AI for about twenty years, and we loved it. Think about everyone who drives for a living: the courier, the cabbie, the long-haul trucker, the guy bringing your pad thai. Google Maps is machine learning in their pocket, predicting traffic and rerouting them around the wreck on the 101 before they hit it. It made every one of those jobs easier, and I have never heard anyone say it stole their work. The opposite happened. A phone that can route a stranger to any address on earth is the reason rideshare and food delivery exist at all, and millions of people now drive for a living because of it. Everybody in Los Angeles used to keep a Thomas Guide in the glovebox, a fat spiral-bound atlas of every street in the county, and nobody held a funeral for that either. Spam filters, autocorrect, fraud alerts, face unlock, the “because you watched” row on Netflix. All of it is AI. We didn’t blink, because back then we called it features.

So what changed? A feature is something you use. Somewhere along the way AI stopped being a feature and became a character, something with intentions that’s coming for you. Four decades of Terminator movies didn’t help. Nobody ever made a movie where the spam filter becomes self-aware. Tools don’t come for anyone. Characters do. And once a story has a villain, every lost job looks like the villain’s handiwork.

Strip out the word AI and the story is one of the oldest we tell: an outside force is coming to take our jobs. People have told it about machines, about trade, and loudest of all about immigrants. So here’s a mirror worth holding up. If you roll your eyes at “they’re taking our jobs” when it’s said about immigrants, check whether you’ve started saying it about AI. It’s the same story with a new villain.

The comparison only goes so far, and I want to be careful with it. An immigrant is a person, with a family and a whole life, and AI is software. I’m not weighing them on the same scale. What they share is the story we tell about them, and in both cases the story gets the economics backwards. When the National Academies reviewed the evidence on immigration, they found it grows the economy over the long run and has a very small effect on native-born workers’ wages overall, with the most pressure falling on earlier immigrants and workers without a high school diploma. More people doing more things makes more work.

That idea is the one the AI panic keeps missing, and once you look for it, it shows up everywhere.

Start with the picture the fear runs on: a fixed amount of work in the world, the machine takes a slice, and people are left with less. That picture has been wrong every time we’ve tested it. When ATMs arrived, everyone assumed bank tellers were finished. Instead, as the economist James Bessen documented, teller jobs held steady and even grew for decades, because ATMs made branches cheaper to run, banks opened more of them, and tellers shifted from counting cash to dealing with customers. Spreadsheets did the same thing to finance. They wiped out a mountain of hand calculation, and the number of people doing financial analysis grew, because once modeling was cheap, everybody wanted more models. When a tool makes work cheaper, we end up doing a lot more of it.

Then there’s the bar, which never stays put. When desktop publishing let anyone make a flyer, the world spent a few years drowning in Comic Sans, and then a plain flyer stopped being acceptable. When Squarespace turned a website into a weekend project, every business needed one, and the standard for a decent one went up every year after. Film and visual effects ran the same arc for decades: every tool that made a shot cheaper got spent on more shots and better ones, and audiences came to expect it. Productivity mostly gets absorbed into higher expectations, and somebody has to meet them. AI will do this too. Whatever it makes easy this year becomes the floor next year.

And making AI work is work. The dream of typing one prompt and watching your whole job get done is the Staples Easy Button, and it’s a fantasy, mostly sold to people who don’t understand the tech. A demo is the first one percent. The other ninety-nine is wiring it into how a company runs, feeding it clean data, checking what comes out, and keeping it running when something upstream changes. People do all of that. You can see it in the job market. The forward deployed engineer, whose whole job is to sit inside a company and make the AI fit, went from an obscure title to one of the hottest roles in tech, with demand up roughly tenfold in a year and a half. Around that role is a growing crowd of people evaluating what models produce, cleaning the data they run on, writing the rules for how they get used, and teaching whole teams how to use them at all. Every company that tries to put AI to work finds out it needs more people who understand it, and there aren’t nearly enough of them.

So why do the headlines say otherwise? For several months this spring, AI was the most-cited reason for US job cuts. That sounds damning until you notice that “we’re becoming more efficient with AI” plays far better with investors than “we over-hired” or “demand is soft.” Those counts measure the press release.

None of this means AI stands still. It’ll keep getting better, probably fast. The scenario where it shrinks the job market is the one where AI keeps growing and the world doesn’t grow with it: no new demand, no rising bar, no new work built on top. Theoretically that could happen. It hasn’t happened with any tool before, and there’s no sign of it now.

And none of this means the pain is imaginary. People are hurting, with immigration and with AI. The diagnosis is just wrong. When the economy grows and your slice doesn’t, the newest arrival is the easiest thing to blame. What people are feeling is inequality: the gains are real, and they’re landing unevenly. The sharpest version right now is at the start of careers, where Stanford’s researchers find young workers in AI-exposed fields being hired less than their peers, though they note interest rates are tangled up in it too. That’s worth naming plainly. It’s also a problem about who captures the gains, and that problem is a lot older than AI.

Here’s my favorite thing about Google Maps: nobody thinks of it as AI anymore. It’s just how you get somewhere. That’s where every tool that lasts ends up, absorbed so completely that we forget it was ever new, let alone scary. In the meantime, forget whether AI takes the jobs. Ask who gets the time it saves. If those hours go to better work and a fair share of what that work earns, this is Google Maps all over again, and in ten years nobody will remember being afraid. If they go somewhere else, the fix lives in how we split the gains, and it always did.