Flip side of advances in machine learning
For the underskilled, it could mean yet another threat to their jobs
Marc Andreessen, venture capitalist and one of the pioneers of the world wide web, once declared: “The spread of computers and the internet will put jobs in two categories. People who tell computers what to do, and people who are told by computers what to do.”
Andreessen has since repudiated this declaration, and taken a more optimistic stance. But economists, a more pessimistic bunch, are taking the possibility of this sort of bifurcated future more seriously.
As machine-learning technology enjoys rapid progress, more top researchers are investigating the question of what work will look like in a world filled with computers that can replicate or surpass many of humanity’s own mental abilities.
This is different from the scenario where robots take people’s jobs outright and leave humanity obsolete. While some economists claim to find signs of automation-induced unemployment, the amount is still very small, if it even exists at all.
With the US labour market having reached pre-recession levels, worries that jobs will become permanently scarce have quieted. But that doesn’t mean the jobs people have in the future will be good ones. For decades, some economists have fretted about what they call skill-biased technological change, or the possibility that new technologies will reward those smart or mentally flexible enough to master them, while devaluing the skills of everyone else.
As computerisation proceeded in the 1980s, and as inequality rose, some economists worried that skill-biased technological change might already be having a big effect.
But they probably jumped the gun. A 2002 paper by labour economists David Card and John DiNardo observed that wage inequality stopped rising in the 1990s, even as computerisation accelerated. The authors also noted that the 1980s saw a diminution of the gender wage gap, despite the fact that women were less likely to have computer-intensive jobs.
But just because skill-biased technological change doesn’t explain the 1980s doesn’t mean it will never happen. In 2010, labour economist David Autor warned that routine tasks — jobs like assembly-line manufacturing or traditional office work — were being automated. These jobs use a lot of brain power, but in a predictable, repetitive way — exactly the kind of thing that computers can do better than humans. Autor found that his measures of routine task input were declining decade by decade.
A recent study by Autor and co-author Anna Salomons finds that since the 1970s, industries with faster productivity growth, international patenting and robot adoption have all seen labour lose out to capital. That’s not a slam-dunk case — there are other reasons these factors could be hurting workers, and the rise of capital income could be mostly due to other forces.
But this research raises the disturbing possibility that automation will lead to the final victory of capital over labour. Now the worries about automation-induced inequality have increased, thanks to the stunning rise of machine learning. Since 2013, there has been a surge of interest in this new technology, which allows computers to do tasks like image and speech recognition that were previously the sole province of human brains.
Meanwhile, entrepreneurs and big businesses alike are dreaming of ways to use machine learning to replace a vast array of human tasks, from driving trucks to preparing food. Venture capitalists are pouring money into machine learning start-ups — often known by the trendy if inaccurate buzzword of “artificial intelligence”.
Economists, true to form as the dismal scientists, are concerned. If machine learning automates away low-skilled tasks, as some predict, it might not make workingclass people obsolete, but it could make their existence miserable nonetheless.
It’s possible to imagine a future where lower-skilled people are constantly seeing their jobs get gobbled up by machines, forcing them to always be transitioning to new tasks — perpetually seeking a niche that hasn’t yet been devoured by ingenious entrepreneurs and their subservient robots, even as wages diminish. That scenario doesn’t necessarily involve high unemployment, but it’s hellish enough that it should worry people.
So what can be done to avert this future? The popular ideas include universal basic income and subsidies for the employment of human workers. These are all ideas worth trying out on a modest scale, to see if they work; even if machine learning isn’t the threat some fear, they could be very helpful in reducing inequality.
Another idea is a social wealth fund — a government-managed fund or collection of funds that would use tax revenue to purchase shares in companies and distribute the dividends to citizens. A social wealth fund would create a true ownership society, insuring the working populace against the rise of the robots by allowing each person to own a piece of those robots’ output. Ultimately, this seems like the simplest and most elegant solution.