Who is Work For?
This week, I saw that Palisade Research released an hour-long interview with a guy named Tao Lin, who, as far as I can tell, is not the poet, novelist and psychedelic drug enthusiast. This Tao Lin is a researcher at Anthropic, and is just the latest in a long line of AI researchers who have chosen to drone on in public about how AI will destroy the world and render most of us unemployable.
I've been annoyed for years by this kind of thing, but I finally put my finger on why: these viewpoints originate in a worldview which barely registers the existence of non-automatable work, and one that fundamentally has no respect for work that can't be automated, and I find that incredibly small.
My Priors, My Priors
Before I dive into that, I should put some priors on the table: I've worked in training, operations and learning/development for more than two decades, across a number of different industries. Some of these have been heavily tech-centric; I've worked on an API program for a major telecom provider, I've worked at one of the very big tech companies. Some have been decidedly low-tech; I've worked in manufacturing facilities, I've worked at industrial facilities. Some have been a mix; I've worked in call centers and for a financial provider. All of these workplaces used computers, but not all of them were computer-centric. They required human labor, they required heavy machinery, they required physical space and physical objects like desks and pens and gloves and trash cans, all of which are, themselves, made by human labor which is assisted and enabled by computers and machines.
Today, I work in a place where we have a display wall of items people have flushed down the toilet: blankets, bricks, toys and horrors the reader cannot comprehend — no amount of technology will prevent humans from doing what so many of us do, and we will always need humans to clean up after ourselves. Swinging a pickaxe at a fatberg is, for the foreseeable future, a strictly human endeavor.
I like learning about and talking to people about work. There's a good possibility that part of what you know about me started with WORK IS FOUR LETTERS. There are so many different kinds of work in the world; I could spend every day for the rest of my life talking to a different person about their work and I'd learn several new things every day, and were I to ever win the lottery, I might do exactly that with my time.
Also, I'm not particularly hostile to AI technology; I think it has upsides and downsides, just like any other technology. I'm probably less hostile to the companies producing this technology than the average person. But the researchers and the executives representing this work and these companies are missing some really fundamental things about the nature of work, and humans, and the humans who do work, and why they do it and why it's valuable.
The Hidden Life of Production
If you'll indulge me, dear reader, I'd like you to engage in an exercise from wherever you're reading this, whether it's at your desk, or on your couch, or on a bus or a train taking you to or from somewhere you want to go. Think about how each of the objects around you was made.
As I write this, I'm sitting at my rat's nest of a home workspace; above my desk are four framed, stylized maps of cities I've lived in. Each of these maps was drafted by a designer, who adapted them from older maps and city blueprints, and who then stylized them and created a layout and design to make them look pleasing in a frame on the wall. This work was likely done in a tool like Photoshop or Illustrator, but the designer did the work — these were simply the tools of the work.
Once the designer created these pieces, they then had to print them (or contract with someone to have them printed). The printer was designed by a human; prototypes were tested, adjustments were made, inspections conducted. Each machine passed through countless human hands in one way or another — thousands of people make a complex product like this. Humans use computers to manufacture the product. These are products made by humans and machines, for the benefit of humans. We use tools to make things for us.
You can fall down the rabbit hole thinking about this kind of thing — what about the ink cartridges? You could think about this all day, if you want. Sometimes, that's my job.
But back to the exercise: this describes four items directly in front of me, each of which has an impossibly complex chain of design, manufacturing, testing, shipping, implementation, etc., involving thousands or tens of thousands or hundreds of thousands of people — or more — before they arrived on my wall. Four simple items — printed maps in frames. I am sitting next to a wall of bookshelves with literally hundreds of books and hundreds of records exploding from the shelves. My desk is a disaster area of cables and cords and chargers and computers, each of which has its own production story. Everything around you is like this: look at your couch, or the emergency exit latch on a bus window, or the coffee cup on your desk. There's an enormous ecosystem of people and products and machines and things that all work together to bring you something as simple as the humble coffee cup you forgot to empty yesterday.
Automation is not cost-free. Jobs get eliminated: typing pools dry up and disappear, switchboards stop lighting up, call centers go dark as AI agents take over. All of these are real problems, and they shouldn't be minimized. But jobs ending and work ending are two very different things, and they're too often conflated. Better mousetraps haven't ended the vocation of the exterminator. The work continues.
Willful Blindness
It's perhaps unfair to pick on Lin here; after all, other researchers — including other Anthropic researchers — have been saying different variations of the same thing for years. But I'm going to pick on him anyway. Given how wealthy the average Anthropic employee became this week, he can dry his tears with $100 bills. In the interview, Lin says (lightly edited):
"I think there's a less than 50% chance of AI taking over. I don't know. But I think the most predictable effect is everyone losing their current jobs, everyone needing to move to a different occupation briefly, and then sometime after that being completely unemployable, needing to rely on the state to provide for them financially, and having no hope of ever finding employment again. [...] I think it's very likely that pretty much all humans on Earth will become permanently unemployable and will need to strictly rely on government welfare to sustain themselves, and human employment will go away."
We have the same conflation of jobs and work here. This is a preposterously myopic thing to say, and it's a deeply offensive thing to say, but, mostly, man, it's a pretty stupid thing to say.
But it does get at the root of my long-winded complaint here, which is that researchers like Lin simply do not value, or even imagine, a world with work that an LLM cannot do. The cleaning crew who take out the trash and vacuum the floors at Anthropic's San Francisco office don't seem to exist to him. The technicians who maintain the elevators? Completely invisible. The person from whom he orders coffee in the morning, or lunch in the afternoon, or dinner in the evening? Irrelevant and meaningless, as are all of the people involved in growing the coffee or producing the food or making the plates and the cutlery used to consume it. Later in the interview, he throws out the movie WALL-E as one possible example of what the future could be like, but WALL-E is an allegory for children.
All of these people seem to have some half-baked thought about how AI-powered humanoid robotics are just over the horizon, but that train's been late for five decades and isn't getting here any sooner. The world of robotics is an exciting one, but we're a very long way off from Rosey the Robot cleaning our homes and cooking our dinners and minding our children. In all likelihood, none of us will live to see that. Your children might not live to see it, either. Try something simple like writing cursive with your non-dominant hand and you'll start to appreciate one small, tiny fraction of the complexity of the endeavor. These are hard problems into which billions (or trillions) of dollars and countless human hours of labor have gone, and while we've made great strides, we still have a very, very long way to go, and humans will be doing that work, using machines as tools, just as we always have. To be fair to Lin, he recognizes that we're nowhere near this reality yet, but he treats it like an afterthought, rather than, you know, the whole ballgame.
Lin's by no means alone in this view; it seems pretty endemic among everyone working in this field, and it's wildly disconnected from reality. Systems and things and physical objects and entire fields of work get completely papered over and disappeared in this worldview; anything that can't be automated by the LLM simply isn't worth consideration. From this perspective, most or all humans will have skills that no longer matter in a world where AI is the thing doing the work, and humans are the tools of that work. People like Lin are paid extraordinary sums of money to think about the world in which their products are used and how they are used, and seem to have neither awareness nor even interest in how that world actually works or the people working within it.
A popular criticism of AI researchers and executives is that they're all galaxy-brained: big, huge ideas that are too big for reality. But I think it's the opposite: these are very small thinkers, with limited curiosity, and even more limited interest in anything that isn't on their screen. The industry could use much bigger thinkers, who can imagine a world beyond that of the next model, or the next earnings report. It needs many more people who can picture a world they haven't already seen on a television screen. It needs people who like, and who understand, and who, most importantly, value people and what they do.
There's a lot of discussion about "human in the loop" in regard to technical questions about AI safety and governance, and the technical discussion is beyond my expertise. But, from a practical standpoint, the human is always in the loop, because if the work isn't, in the end, for the human, who would it be for?