A PRODUCTIVITY BOOM LOOMS.

How correlated do you think productivity gains are from decade to decade? In other words, does productivity growth this decade mean more or less productivity growth next?

It’s a trick question: Turns out they aren’t correlated. In a new paper on “the modern productivity paradox,” three business- school profs ( two from MIT, one from Chicago) correlate U. S. productivity growth rates in adjacent, rolling 10- year periods over the last half century and find the graph is a random scatter of points, with the formal statistical tests revealing no significant correlation. That’s true both for labour productivity — total output divided by the number of workers — and “total factor productivity,” the overall gain in output after taking into account both greater labour and greater capital inputs.

The three economists ( Erik Brynjolfsson, Daniel Rock and Chad Syverson) read that lack of correlation from decade to decade as good news. We’ve just been through a decade of pretty lousy productivity growth. If decade results were positively correlated, this last down decade would augur another down decade, which in fact lots of people seem to be predicting. The paper’s authors are a lot more upbeat. The algorithms are coming, they argue. They believe artificial intelligence ( AI), which is just gathering steam, is going to have a transformative effect. ( AI is fine so far as it goes but I wonder if we’ll ever get “AW” or “AM”: artificial wisdom and artificial morality. The humanborne versions don’t seem to be doing that well these days.)

Like the first round of computerization in the 1970s and 1980s, AI isn’t yet juicing better productivity numbers. But we’ve seen this movie before. In 1987, MIT’s Robert Solow, who won that year’s Nobel Prize in economics, became even more famous with his quip that “We see computers everywhere except in the productivity statistics.” That spawned a lot of economic literature trying to figure out why. Was it a problem of incorrect measurement? Was it a quest i on of f alse hopes? Were the widespread, transformative applications that everyone assumed were coming simply not meant to be?

Brynjolfsson, Rock and Syverson argue there’s some truth in all these explanations. On mea s u r e ment , for instance, there’s a problem that much of the “output” of smartphones in terms of usefulness, enjoyment and diversion simply doesn’t get counted in GDP: Google Maps, for instance, is given away free despite its immense value. And there’s the complication that obsessive diversion smartphone and Internet users enjoy at work may actually reduce traditional GDP.

But mainly, the economists think, the problem is timing. In fact, the productivity paradox involving computers was solved not long after Solow discovered it. Productivity boomed through the 1990s as computers became commonplace in the workplace. That’s not unusual. Transformative technologies usually take time to diffuse through an economy. The Industrial Revolution, based on steam, took several decades to build up its own steam. As for electricity, one wellknown study noted that “half of US manufacturing establishments remained unelectrified until 1919, about 30 years after the shift to polyphase alternating current began.”

It seems likely the AI revolution is going through a similar latent phase before taking off. It’s even possible that in these early stages new technologies are a drag on productivity growth. Lots of money, brainpower and effort goes into them but with no shortterm payback. Driverless cars seem to really be coming, but they’re not here yet ( even if Uber’s said it’s buying 34,000 of them from Volvo starting in 2019). For the last few years, tens of thousands of well- paid people have been busy developing driverless cars but they’re still not developed quite yet.

When they do arrive, the productivity gains could be big. Suppose, the authors say, they eliminate the need for two million of the 3.5 million Americans who drive vehicles for a living. That’s 1.7 per cent of private employment in the U. S. If the transition takes place over 10 years, that’s a 0.17 percentage point boost to annual productivity growth: After the decade is over, it will be possible to produce the same total output, all driving included, with 1.7- per- cent fewer people. Similar calculations for call- centre jobs that could be replaced by increasingly sophisticated voice- recognition systems ( oh, joy!) add another 0.10 points per year for 10 years.

There are also possibilities for capital productivity growth. If it’s true the average car stands idle 95 per cent of the time, driverless cars will enable much greater efficiency in the use of the car capital stock.

These changes will present obvious difficulties for displaced drivers — and maybe even displaced emergency- room doctors, if driving no longer will be subject to human error — and that could well create social problems, even as new jobs do come along. But transformative change at least will explain, and eliminate, this latest version of the productivity paradox.

IT SEEMS LIKELY THE AI REVOLUTION IS GOING THROUGH A LATENT PHASE BEFORE TAKING OFF.