The Coming Robot Reckoning

by Michael Dorf

I've been thinking and writing a lot about robots lately. Prompted by a Washington Post reporter's questions about artificial intelligence, last December I wrote a column comparing rights claims for robots with rights claims for non-human animals. The use of a "bomb robot" by the Dallas police in July prompted this Verdict column. Last week, a government report prompted me to contemplate the best legal approach to self-driving cars. Most recently, the discussion of job losses during the presidential campaign has led me to try to take the long view about the future path of the economy. Here I want to explore the likely impact of continued automation on American politics.

First, the facts. Donald Trump is wrong that U.S. manufacturing jobs have fled to Mexico in large numbers because of NAFTA, but he is not entirely wrong that over the last three decades Chinese manufacturing replaced a big chunk of U.S. manufacturing due to cheaper labor costs. However, "not entirely wrong" is not the same as "right."

For one thing, even when Trump is identifying a real problem, his "solutions" are either hopelessly vague or likely counterproductive (or both). But Trump isn't wrong only in the sense that he has no effective policy prescription. He also misapprehends what's actually happening.

Even during the last presidential election four years ago, it was already evident that wage pressure and other factors were leading some global firms to leave China in pursuit of even cheaper labor in Vietnam, Pakistan, Bangladesh, and elsewhere. Of course, jobs leaving China for other developing countries doesn't do anything for displaced U.S. workers in the industrial heartland, but it does show that there is a limit to the logic of outsourcing: Countries that succeed in attracting manufacturing because of cheap labor will eventually have higher wages that undercut their chief advantage. I'm not saying that cheap foreign labor isn't a threat to U.S. manufacturing jobs; I am saying that it's not the whole story.

At least half the story is automation. Even as U.S. manufacturing output has increased despite outsourcing, manufacturing jobs have decreased in number. This phenomenon will likely account for a larger chunk of job losses over time. If robots are cheaper than workers in Detroit today, they will be cheaper than workers in Dhaka soon enough. In the long run, Trump's xenophobia is misplaced. A more visionary rabble-rouser would promote robophobia instead.

This is not simply a story of transition from a manufacturing economy to a service economy. Services can be--and are being--automated too. Twenty years ago, a displaced factory worker might find work, albeit less lucrative work, as a taxi driver. In another decade (or much less), that avenue will be foreclosed, as self-driving taxis make taxi driver as unavailable a career choice as elevator operator. In many restaurants, iPads have begun to reduce the need for wait staff. These trends will continue as technology improves because robots do not form unions, take sick days, or require health insurance.

There is a robust debate raging about how much work that requires abstract thinking (such as practicing law) can be automated (nicely summarized by Frank Pasquale and Glyn Cashwell in this 2015 article), but even if one takes a pretty skeptical view about the capacity for automation of some of the more complex mental tasks, the future still looks highly automated. Absent some substantial change, there will not be enough jobs for people who want them.

But wait, you say. Hasn't the employment situation recently improved? Maybe the impact of automation is overblown. Maybe, but probably not, as I can explain with the following graph of labor force participation (LFP) over the last forty years (which I generated using a handy tool available on the Bureau of Labor Statistics website):

Note that nearly all of the increase in LFP between 1976 and 2000 is the result of women entering the workforce in large numbers. As another BLS publication notes, "women reached the peak of their labor force participation in 1999." That peak is very close to the overall LFP peak in 2000, after which LFP declines substantially. The decline begins in the early 2000s, well before the Great Recession. That timing is just about right as a marker for when computer technology would have begun to yield substantial increases in productivity and concomitant job losses.

We do see a modest uptick in LFP over the last year. Although I would like to be proven wrong, I see no reason to think that the recent LFP increase will bring us back to the 2000 peak. It is possible that new technologies will create all sorts of new jobs that we have not imagined yet, but that is not likely the way that automation will work, at least over the long run. (Go to Postscript 1, below, for a defense of LFP rather than other measures.)

So, assuming I'm right that the future holds many fewer jobs than people available to fill them, what happens then? The answer depends on politics. A rational society would treat the end of the need for work as liberating by ensuring that everyone had reasonable access to the fruits of the robots' labor. People would be free to develop their interpersonal relationships and pursue their passions.

True, there are dystopic versions of this future. Even assuming the robots do not turn against us as in The Matrix or Terminator films, we might squander our leisure on sloth, as in Wall-E.

But will everyone even have access to the benefits of automation in order to squander them? Standard analyses of this problem (such as this one) assume that in the absence of either socialism or massive philanthropy from future tech multi-billionaires, our existing capitalist system will lead to a society in which the benefits of automation are distributed very unevenly. Capitalists who own the means of production and a small cadre of super-skilled knowledge workers will reap the benefits, while a gigantic underclass lives in misery.

That's unlikely. Think about Henry Ford's insight that if he paid his workers a decent wage, he would have not only satisfied workers but customers to buy his cars. If the benefits of technology are beyond the means of the vast majority of ordinary people, that severely limits the ability of capitalists and super-skilled knowledge workers to profit from the mass manufacture of the robotic gizmos.

Accordingly, it is not that difficult to imagine that one path to a redistributionist future runs through capitalism rather than around it. Enlightened capitalists would understand that they need customers and that, with automation severely limiting the number of jobs available, customers can only be ensured through generous government-provided payments to individuals and families.

Of course, capitalists have not always been enlightened. In prior periods, reforms that ultimately benefited owners of capital were adopted over their objections. That could happen again. My point is simply that a future of highly concentrated wealth is limited by the fact that the very wealth in question depends on the existence of customers, which in turn depends on government action to redistribute wealth. Such redistribution--to facilitate not only the comforts of the robotic future but the other necessities of a meaningful 21st-century life--would go a long way towards addressing the legitimate concerns of the people in the non-deplorable basket.


Postscript 1: It could be objected that LFP is not a good measure of the economy's ability to produce jobs. After all, LFP can decline for all sorts of reasons that have nothing to do with job availability. More people going to college and getting graduate degrees means fewer people in the work force. So do demographic shifts that lead to more retirees. Indeed, even if we focus on a decline in LFP among prime-working-age men, it is possible to identify other factors that could explain what's going on. E.g., people shifting to the underground economy. I agree that LFP is at best a crude measure of the economy's ability to create good jobs. After all, even unemployed people count as part of the labor force for LFP purposes, so LFP can overstate the health of the jobs picture. Nonetheless, I would defend LFP as a reasonable gauge of jobs over the long run because nearly all of the other factors that contribute to LFP are in turn influenced by the availability of good jobs. E.g., people are unlikely to take early retirement when there is good-paying rewarding work available.

Postscript 2: Anyway, even if I'm wrong about the story until now, I could still be right about the future. Eventually, automation and AI could leave very few jobs available for humans. And that is really what I wanted to talk about.