Automation, productivity, and the Jobless Recovery

Many news outlets lately have been running stories on the mystery of the jobless recovery of the economy. That is, most economic indicators are rising, but the economy isn’t adding jobs, and unemployment is still high. In economic terms, this means that productivity has increased, and is probably still increasing. Economists describe productivity as the amount of economic product per worker. If the GDP goes up, but unemployment stays high, that means that productivity has increased. Increased productivity is supposed to be a sign of economic strength, but you have to wonder what the point of such strength is if it’s not producing more jobs. The assumption is that increased productivity will lead to growth, which will lead to more hiring. But what if it doesn’t? Suppose the new productivity we’re seeing is the payoff from all the investment in information technology and automation during the tech boom. Now that corporate leaders are seeing this increased productivity, are they more likely to spend their shareholders’ money hiring people, or investing in more automation? Back during the tech boom my brother predicted that we hadn’t even begun to see the productivity increases as a result of info tech and automation. Suppose he was right? How will society deal with these increases in productivity? Is there a way to turn the productivity increases in society’s favor? AI researchers have a duty to at least consider these questions.

Some might say that a goal of AI is to create machines that can do the work of humans. Perfect achievement of this goal would be equivalent to increasing productivity to infinity. I.e. we produce all our economic product without any human work or workers. In this light, the question becomes whether it’s possible to create a society that supports humans’ pursuit of happiness, but that is not dependent on human work. A collegue of mine has a vision of a kind of Athenian utopia, where machine-slaves do all our grunt work, and humans are freed to lie around eating grapes and chocolate bars and thinking about math and philosophy and stuff. The problem with this vision for me has always been: where do the grapes and chocolate bars come from? How do they get into the hands of the noble, yet indolent, citizens? One possibility is a soviet-style centrally-controlled economy, but central control of an economy the size of the United States seems likely to be highly inefficient. Would it be possible to retain an efficient, if not 100% free, market economy in such situations?

One clue of how to do this comes from agriculture. Thanks to technology, agricultural productivity in the US is so high that we no longer need all of our farms. Of course, the farm owners still need to eat. The solution? Take tax money and pay farmers not to plant or grow anything. Now suppose we apply this to all workers in all jobs, not just farmers and agriculture? The result, take tax money and pay people not to work. Since nobody would have any income, all tax revenue would have to come from the companies that own the machines that produce everything, which would be taken as taxes, and redistributed back to the non-working populace, who would live out their lives a happy consumers, tourists, artisans, philosphers, and dilettantes, spending their money as they see fit.

What I’m describing, of course, is welfare. Unlike some European countries, the U.S. with it’s protestant work ethic has been generally opposed to large amounts of welfare, and the Clinton-backed welfare reforms of the 90’s were intended to decrease, not increase, the number of people on welfare, demanding that people try and find work. But this can only work when the economy has lots of jobs. Imagine productivity keeps rising indefinitely. Who can say how 170 million unemployed workers would vote?

Back to work… building a new research infrastructure

So I passed my proposal last week, and sometime in the next week I should officially advance to candidacy. I now have a plan for my dissertation, finally, and it’s time to get to work. I took it relatively easy last week after the proposal cleaning my offices and investigating new robot simulators and software — more on that in another posting. Now it’s time to get back to work. Both on Topographica, and on rebuilding my dissertation research infrastructure and getting new experiments going.

The funny thing about research code is that the usual coder’s advice, Plan to throw one away, becomes plan to throw them all away. I.e. you never really know ahead of time which ideas will work and which won’t, and you can’t waste time building code to last until you’re really sure it will work. I learned this the hard way, spending a lot of time in the last couple years writing code that is now totally useless, and my dissertation work will grow out of a tiny kernel inside what I wrote.

This leads to a different idea about building software that what you might learn as an engineer. It’s the auxiliary parts of the code that need to be built to last, modular, easy to use — code for setting up experiments, tracking their progress, collecting data, and visualizing results needs to be well-built, modular, and bulletproof; this code forms your workbench, your virtual laboratory. A good infrastructure like this will allow you to quickly and easily evaluate new computational hypotheses. On the other hand, the actual algorithms under study need to be written and evaluated quickly, with the idea that any one may be thrown away tomorrow, and you shouldn’t invest too much in a particular idea or piece of code until you’re sure it’s going to be successful.

Now I think I’m on track, though