In “Why Do Americans Work So Much?“, Rebecca Rosen poses some answers to the question in the title, most notably, “American inequality means that the gains of increasing productivity are not widely shared. In other words, most Americans are too poor to work less.” I’m not convinced this is true; one problem we have involves the difficulty or illegality of building and selling relatively inexpensive housing in high-demand areas (see here and here for two discussions, and please don’t leave a comment unless you’ve read both links thoroughly). Some of what looks like financial “inequality” is actually people paying a shit ton of money for housing in New York, Seattle, L.A., and similar places, rather than living in cheaper places like Houston or Phoenix. Homeowners who vote in those areas vote to keep housing prices high by strangling supply.

Plus, I’d add that, per “The inequality that matters II: Why does dating in Seattle get left out?“, financial inequality isn’t the only kind, though for some reason it’s gotten an overwhelming amount of play in the press over the last ten years. I’ve seen people speculate that financial inequality is fun to attack because money can easily be taken from someone at the point of a gun and given to someone else, while other forms of inequality like beauty or a playful disposition can’t be taken so easily.

Still, there’s one other important factor that may be unexplored: Demanding and remunerative cognitive jobs may not be easy to partition. That is, one person doing a cognitively demanding job 40 hours per week is way more efficient than two people doing the same job for 20 hours a week. And that same person may be even more efficient working 50 or 60 hours a week.

Let me explain. With some classic manufacturing tasks—let’s imagine a very simple one, like turning an hex key—you can do x turns per hour times y hours. With many high-value jobs, and even ambiguously defined median-value jobs, that isn’t true. In my not-tremendous-but-not-zero experience in coding, having one person stuff as much of the code base—that is, the problem space—into their head as possible makes the work better. The person learns a lot about edge cases and keeps larger parts of the codebase in their mind. The cost of attempting to explain the code base to another person is much higher than keeping it all in one’s head.

Among professors, the ones who’ve read the most and written the most usually exponentially better than those who have read 75% and written 75% as much. They’re 5x as valuable, not 33% more valuable.

One sees similar patterns recur across cognitively demanding fields. Once a person has put in the 10,000 hours necessary to master that field, each additional hour is highly valuable, and, even better, the problem domain is better understood. That’s part of the reason law firms charge so much for top lawyers. Those top lawyers have skills that can only be developed through extensive, extreme practice.

I see this effect in grant writing: we don’t split proposal tasks because doing so vastly increases the communication overhead. I’m much more efficient in writing an entire proposal than two or three people could be each writing parts. We’ve rescued numerous doomed proposals from organizations that attempted this approach and failed.

Many of you have probably heard about unfinished and perhaps unfinishable projects (often initiated by government). Here’s a list of famous failed software projects. Some of those projects simply become so massive that latency and bandwidth between the workers in the project overwhelm the doing of actual work. The project becomes all management and no substance. As I mentioned in the previous paragraph, we’ve seen many grant proposals fail because of too many writers and no real captain. At least with proposals, the final work product is sufficiently simple that a single person can write an entire narrative. In software, thousands of people or more may contribute to a project (depending on where you draw the line, hundreds of thousands may contribute: does anyone who has worked on the compiler or version control system or integrated development environment (IDE) count?).

Put these trends together and you get people working more because the costs of splitting up tasks are so much higher. If you put five junior lawyers on a project, they may come up with a worse answer or set of answers than a single senior lawyer who has the problem space in his head. The same thing could conceivably be true in software as well. The costs of interconnection are real. This will increase inequality because top people are so valuable while simultaneously meaning that a person can’t earn x% of the income through x% of the work. A person must do 100% or not compete at all.

This is also consistent with changes in financial remuneration, which the original author considers. It’s also consistent with Paul Graham’s observations in “The Refragmentation.”

Finally, there may also be signaling issues. Here is one Robin Hanson post on related concerns. At some point, Hanson described working for Lockheed before he did his Ph.D., and if I recall correctly he tried to work fewer hours for commensurately lower pay, and that did not go over well. Maybe Lockheed was cognizant of the task-splitting costs I note above, or maybe they were more concerned with what Hanson was communicating about his devotion to the job, or what example he’d set to the others.

So earning may not be scalable. It may be binary. We may not be “working” less because we’re poor. We may be working less because the nature of many tasks and occupations are binary: You win big by working big hours or don’t work much at all.

EDIT: See also “You Don’t Need More Free Time,” which argues that we may not need more free time, but rather the right free time—when our friends are free. I also wonder if too much “free” time is also enervating in its own way.