Here's a 1,000-word summary of a 3,000-word excellent essay, so uh... let's just say I'm adding to noise rather than illumination today

Mr Albrecht, the "other" part of the duo that runs Economic Forces (I mostly read/quote from Josh Hendrickson, a money guy and pretty well-versed/respected Bitcoiner-ish), ran a great piece titled "You Are Not a Horse" yesterday.

Not at all unrelated to my newfound interest in Robert Gordon’s classic The Rise and Fall of American Growth (#1485162, #1476033)

The title is an excellent summary of the direction the article is taking: basically calling bullshit on the tractor-horse analogy, a historical development I wrote about for Human Progress two years ago. (Guess I promised to write this up for SN but never did! #886161, soz Siggy!)

And while it’s true that plenty of horses became unemployed through the mechanization of the American economy, 1870 to 1970, the process took lifetimes (peak horse is 1920-ish — and didn’t even fully remove horses from their working duties, but ultimately did turn them into high-value, high-maintenance leisure goods (luxuries!). Meaning: they don’t really have to earn their own way, their food and other needs subsidized by someone else’s value production.

“it’s certainly not a new fear, and not just an AI thing.”“it’s certainly not a new fear, and not just an AI thing.”

If AI really is a perfect substitute for human labor, any cost advantage drives to 100% AI. You don’t need an essay to prove it. But “AI will eventually be a perfect substitute” is doing all the work.
But that’s hiding a lot, lots of margins of adjustment and differences, heterogeneity that makes the world the world and not a simple model. How substitutable is AI right now? What would it take for that to rise high enough? What else has to hold?

We talk a lot about these things in econland (#1485186) these days because, I guess, it’s important… and because everyone else is!

Albrecht first walks through the extreme, edge-case scenario of human labor demand literally going to zero. That seems insane

Every dollar you spend lands somewhere. Some dollars land in activities with lots of human labor inside them: a restaurant, a therapist, a roofer. Some land in activities with almost none: a streaming subscription, an automated checkout, cloud storage. So when we are tracing out what happens when AI gets cheaper, it’s not just “Can AI do my job?” It is “When everyone saves money because AI did my job cheaper, what do they buy next?”

“The economy is not one production function. It is many activities. When AI makes some of them cheaper, people don’t just buy more of the same thing. They buy something else.”“The economy is not one production function. It is many activities. When AI makes some of them cheaper, people don’t just buy more of the same thing. They buy something else.”

Hashtag let’s do some microeconomics

When AI can do the things firms are actually buying, cheaper AI does two things at once. Firms substitute AI for workers, which reduces labor demand per unit of output. But cheaper AI also lowers output prices, output expands, and the expansion pulls labor demand back up. Whether labor demand rises or falls depends on which effect is larger. This is the Hicks-Marshall decomposition of derived demand into substitution and scale effects.

(I don’t know what “scale” is doing here… maybe some new/odd/fancy way of saying ‘income effect’…?)

The saved dollar doesn’t vanish when a task gets automated. It creates new tasks within the same job, such as more review, more client management, more judgment calls. Just as there’s not some fixed amount of demand so the scale effects matter, there is not some fixed job.

Surely, obvious, but also something I hadn’t really considered seriously before this latest AI-work craze (Our beloved Mr. Livingston over in Saylorboi-land drew my attention to this what-is-a-job phenomenon #1425743)

For some jobs, the extreme damage of an error can undermine any reduction of price in specific components (the “O-Ring logic”): “When failure on one component destroys the value of all others, you don’t care about the sticker price.” Meaning, for this portion of work no amount of AI improvement completely obsolete human oversight. Powerful to move us away from the edge, but less reassuring for the masses looking to earn a decent living.

So:

The question isn’t whether AI competes with some human goods. It’s whether any human-intensive island survives. Does anyone still spend money on something with a person inside it?

Substitution Ruling the Roost: Software as a Service ModelSubstitution Ruling the Roost: Software as a Service Model

Suppose substitution wins inside most jobs. The saved dollar escapes the workplace entirely. Where does it go?

Historically here, the software industry itself, a close echo of LLMs, has had the opposite tendency:

Heavy digital inputs didn’t drive out human labor. If anything, the industries that automated the most are the ones that spend the most on workers [...] The scale effect won within the sector most exposed to digital automation. The BLS could be completely off but the evidence so far points strongly toward the scale effect dominating in software-intense industries.

Also:

As manufacturing got cheaper, people didn’t just buy more stuff. They shifted spending toward healthcare, education, restaurants, personal services. That’s the saved dollar in action at a more not-quite macro but close level — the savings from cheaper goods flowed toward services.

things just shifted around, no biggie.

Stumbles on my major concern, then:

“migration alone doesn’t help workers unless the destination still has human labor inside it.”“migration alone doesn’t help workers unless the destination still has human labor inside it.”

Yeah, fair… maybe most of us just become decently remunerated servers (housekeeper, PT, chef, yoga instructor, cleaner, banker, personal care, therapist, entertainer etc etc) of, like, one insanely productive and high-value worker, I don’t know.
(yes, I’m playing with the word “worker” here, which I guess becomes the major point; take my meaning, not my lack of dictionary options). But also: infinity hits its attention limit… #1467035, #1449095: “Consuming human services has an opportunity cost: time spent at a live concert is time not spent on a superior AI experience.”

comparative advantage always pops up fighting against this. When automation makes some things cheap, the things that remain expensive tend to be the things that are hard to automate. And the things that are hard to automate are, almost by definition, the things where humans still have comparative advantage. The saved dollar drifts toward where humans are still worth paying. That’s not optimism. That’s what comparative advantage means.

The textile industry analogy, rather than the tractors, is a reassuring example:

cloth got so cheap that demand exploded, and total employment in textiles rose for decades. Same in early steel, early autos. Eventually demand saturated, prices stopped falling fast enough, and automation reduced employment in each sector. The question for AI isn’t “does automation destroy jobs?” It’s “which phase are we in, for which sectors?”

TLDR FINAL TAKEAWAY: Horses had nowhere to go. You do.TLDR FINAL TAKEAWAY: Horses had nowhere to go. You do.

The really optimistic AI story is, thus, this:

at every step, there’s a saved dollar looking for somewhere to land. And the question is always the same one. Where does it go next? For the horse outcome, you need that saved dollar to find nothing with a human attached to it.

A falling labor share is not falling labor demand. There is a range where labor’s share of income is declining but total labor demand is still rising, because the pie is growing faster than labor’s slice is shrinking.

Thanks for reading following along my reading notes to the very end!