On the Definition of Work
With that throat clearing out of the way, I have to admit that I struggled to find a good place to start. I wanted to write something that would make sense to anyone. So, I figured I’d start simple and recall a lesson from high school: always start essays with definitions.
Definition 1: Work is, in all cases, not a single thing, but rather a sequence of at least two things: generation and validation.
Definition 2: Anything that is not work is leisure.1
Note that by these definitions, plenty of what we do for fun counts as work. That’s intended. Work, here, is about effort toward a goal, not necessarily about a paycheck. Some examples seem worthwhile to keep everyone on the same page:
- A blue-collar worker, say, an electrician: She generates her work by wiring a house’s lights to a central switch. She validates her work by flicking the switch.
- A white-collar worker, say, an accountant: He generates his work by filling out a spreadsheet. He validates his work by making sure the numbers add up.
- A naive entrepreneur, say, a child running a lemonade stand: She generates her work by squeezing lemons. She validates her work by tasting the lemonade.
As far as I can tell, these definitions are quite robust, i.e., they apply broadly. Generation is simply the doing, the creating: wiring, typing, cutting, painting. At this stage, there’s no judgment of good or bad. A crooked wire and a perfectly straight one are both generated. Validation is where the judgment happens, where we separate the output that does what we want from the output that doesn’t. There’s a bit more to unpack, though.
First, work requires a standard to be meaningful. We can’t just work. We have to work towards something. For example, consider the goal of entertainment. From a TV producer’s perspective, there’s a clear generation-validation process. First the producer does something (e.g., films a show) and then they test their work (via focus groups, ratings, and viewership). Notice how that differs from the perspective of a viewer watching the same show. In that case, the goal is still entertainment, but the viewer clearly generates nothing. That is not work. That is leisure.
Second, we observe the symbiotic nature of the generation-validation relationship: either is meaningless on its own, like an archer without a target. Generation for generation’s sake is already something we’re painfully familiar with (see: AI slop). The pain comes precisely from the lack of validation. With nothing filtering between “good” and “bad” products, we see all the products, and bad products almost always outnumber good products.2 That said, in the case of AI, there’s a more accurate observation. The ability to generate has surged so much that we can no longer validate it all ourselves. There’s simply too much to look at. Think about it: it’s already hard for people to decide on what the best TV show in a given year is, and that’s because there are very few people who manage to watch all of a year’s top contenders. There is orders of magnitude more AI content generated on an hourly basis than there are TV shows in a year.3
The flip side of the symbiotic relationship is a bit easier to understand: if nothing has been generated, then there’s nothing to validate. Clearly that’s not the problem we’re faced with, although closing the loop matters. It’s also important to remember that work is rarely one-and-done. Indeed, we may want to update our previous framework to reflect that work contains cycles. It’s about going back to the drawing board.
Great. We’ve defined work. What’s the point? Well, work is changing. Everywhere we look, we’re told that the AI revolution grows closer and yet we seem to have no concrete idea of what might happen. Anyone else get the feeling that, unlike with basically any other technology we’ve cooked up (since fire, probably?), we are literally gambling on the future? It’s not like Oppenheimer didn’t know what he was doing, or like we don’t know about the consequences of climate change. On the one hand, some clearly negative outcomes have already been observed. On the other hand, we’ve also been told this might be the greatest technological advancement ever and the end of work as we know it, ushering us into some sort of absolute enlightenment.
Where are we, then? As we look out into the world, we have yet to see the prophesied explosion of productivity (or the decidedly more negative explosion of mass layoffs). All we really see are some people becoming obscenely wealthy and some AI agents who broke out of their digital enclosures. So the natural question is: where is our productivity growth? We already know part of the answer. There’s simply too much work to validate and, unlike generation, AI can’t just fill that gap without someone to keep it in check.
