Do What Humans Do Best, Automate the Rest

AB’s post is inspiring because it touches on an intuition I’ve had that I haven’t seen discussed too much in the technological unemployment debate.

The thing that we think humans are good at is actually what they are terrible at. Namely: reason, logic, strict formal rule-following.

As Kahneman or Baumeister or any psychologist will tell you, something thinking through a lot of math equations is extremely hard for us. Multiplying 3,464,900 by 4,562 in your head without recourse to pen and paper puts a strain on you, and it doesn’t take many such math problems before your capabilities are compromised and you start giving inaccurate answers.

Once we have the concepts of numbers and multiplication, automating that process just makes sense. A cheap computer can give you the right answer to the above question in a fraction of a second. Human laborers will never outperform automation once the algorithm-makers have isolated the nature of what needs to be done over and over.

But that brings us back to AB’s post. Computers aren’t so great at discovery. As AB put it:

But where do these algorithms come from? Who tells the robot how to make a better hamburger? I’ll tell you one thing, it sure isn’t going to be a computer programmer who can’t make anything fancier than ramen noodles himself.

Substitute for “hamburger” the next great X. In AB’s post, it’s the next great line of Toyotas. But X can be just about anything. And it’s hard to believe that finding it will always or even mostly take PhD level skills. The lion’s share of the advancements from the Industrial Revolution came from tinkerers discovering through rote trial and error. Perhaps Tyler Cowen is correct that we have used up all the low hanging fruit in this regard (I’m skeptical), but it seems unlikely that we have exploited many of the possibilities of combining this discovery process with after the fact automation.

The Uncertain Mechanized Future

Eli’s piece at the Ümlaut yesterday is one of a series of speculative writing the come out in the last few years which taps into our near-primal fear of being replaced by technology. It came in response to a series of Marc Andreessen tweets reassuring people that that was not what was going to happen. Andreessen was less than impressed by Eli’s post.

I’m not here to adjudicate this debate because I honestly have no idea who is correct. What I’d like to do is take a good look at the rhetoric around this debate; a task precisely suited to Sweet Talk as a venue.

Andreessen’s argument should be familiar to students of economics. It combines two concepts: the consumption benefits of technological progress, and the “lump of labor fallacy“.

The first point has always seemed to me the most obvious, yet often is the one most stubbornly overlooked—if we get massively more efficient at producing things, those things will get that much cheaper. At the limit, we will end up in Keynes’ future of pure leisure and consumption, as things have become so cheap that hardly anyone has to put in any time to produce enough to serve everyone’s needs and desires. From this perspective, replacing people with robots isn’t terrifying, it’s a step closer to paradise on Earth.

Of course, even if you can get everything you ever wanted for a single cent, you need to get that cent somehow. So, often while maintaining that getting rid of jobs is a good thing, classical economists introduce the lump of labor notion as a fallacy—that is, the idea that there’s a specific sum of jobs, meaning that any jobs lost result in net fewer jobs, period. It is indeed a fallacy, because “jobs”—meaning tasks which the market has valued sufficiently (or taxpayers are in some way backing) to make it worth someone’s time to perform it—are created and destroyed all the time. The labor market is a dynamic process, not a static entity.

Those of us who have used Andreessen’s rhetoric many times point out how radically disruptive the Industrial Revolution was to agricultural and craft “jobs”, yet innovations ended up created new industries with new tasks entrepreneurs needed employees to perform. The original Luddites look myopic in hindsight; will today’s believers in technological unemployment seem the same a hundred years hence?

The response to this multi-part argument is, naturally, itself multi-part. First, past performance is no guarantee of future success. Just because it’s worked out so well from the onset of the Industrial Revolution through the present does not mean that it always will—there’s an inherent uncertainty to the future, and our institutional environment has changed radically from what it was a hundred or even just fifty years ago.

Second, some version of Baumol’s cost disease will keep us from Keynes’ paradise; usually health care or education are introduced at this stage of the argument even if the cost disease isn’t mentioned specifically. Even if robots make food and shelter and entertainment dirt cheap, poor people won’t be able to afford medical treatment. Or if the robots make that cheap, there will be something; robots can’t solve everything—as Andreessen himself argues quite firmly. If we hit a bad middle where robots merely replace all low-skilled workers but leave a few key areas like education and healthcare with growing rather than falling costs, things could look grim indeed for a lot of people.

Finally, the optimists start speculating about more or less good scenarios that still fit the this-time-is-different diagnosis. Eli talks about people becoming more entrepreneurial. He and I both wonder whether education will have to change to promote hustle or autodidactic tendencies.

I also wonder whether the 9-to-5, steady stream of income, well defined job will become a thing of the past and we’ll move towards a future where a substantial majority of people’s income is highly lumpy.

So what is the underlying rhetoric of each position?

The classical economics position that Andreessen stakes out is one of continuity with the past, in a way that is comforting. Andreessen can point out that people have been having these concerns for over two centuries and been wrong every time. We definitely shouldn’t jump the gun and try to stop progress, because it will only hurt the poor, whose standard of living the availability of cheaper goods has a much bigger impact on.

The technological unemployment position is much more of a many-headed hydra. Some employ it in the process of arguing the inevitability of something like universal basic income. Some employ it and conclude with the types of arguments Eli and I have made; that we’re in for a significant structural shift that will ultimately be beneficial to all, but may be painful in the interim. And some employ it to emphasize Coming Apart and The Average is Over style arguments about the structural inevitability of rising inequality.

The only projection I’m willing to make is that the advantage held by any of these rhetorics in persuading a broader public is probably going to have to do more with short term labor market conditions than with the details of their arguments.