Artificial Intelligence and Artificial Problems

Over at Project Syndicate: Former U.S. Treasury Secretary Larry Summers is fencing with current U.S. Treasury Secretary Steve Mnuchin about “artificial intelligence”–AI–and related topics.

Most of their differences are differences of emphasis.

Mnuchin is drawing the issue narrowly: the particular technologies called “Artificial Intelligence taking over American jobs”. And he is, at least as I read him, is elliptically criticizing high stock market values for “unicorns”: companies with valuations above a billion dollars and yet no past record or clear future path to producing revenues to justify such valuations. **Read MOAR Over at Project Syndicate

Measuring Productivity Growth: No Longer So Live at Project Syndicate

Measuring Productivity Growth: The world’s population today is about 20 times richer than it was back during the long Agrarian Age from 7000 BC to 1500, during which limited resources, slow technological advance, and Malthusian pressures kept the overwhelming proportion of human populations at near-“subsistence”–incomes of less than $1.50 per person per day. Today only 1/15 of the world’s population is that poor. And today if we were to take the total money value of what we produce and use it to buy what people receiving less than $1.50/day buy, at current prices the value of global output would be $30/day per person. That is our roughly $80 trillion of annual global income today.

We cannot but greatly lament the enormously unequal distribution of the fruits of our global productivity. But that we today are such a wealthy global society would strike our predecessors from 7000 BC to 1500 dumb.

Moreover, most of what we make and consume today is not what our near-“subsistence” era predecessors. What good to any of us would 40,000 kcal/day in basic grains be? Most of what we make and consume today are goods and services with analogues back in the Agrarian Age that were absurdly expensive. Could Tiberius Claudius Nero eat strawberries and cream in January? No. For one thing, we think the idea of combining strawberries and cream was un-thought of before the cooks of the sixteenth-century Tudor dynasty courtier Thomas Cardinal Wolsey. There was one and only one person who could see a bloody audio-visual drama about witches in his house in the year 1606. He was named James Steuart, he was king of England and Scotland, and he had William Shakespeare and Shakespeare’s acting company on retainer. Yet today more than 4 billion people with their smartphones, tablets, and televisions live, in this dimension at least, better than kings. Nathan Meyer Rothschild, richest man in the early nineteenth century, died in his fifties of an infected abscess. He would have given the bulk of his wealth for one dose of modern antibiotics. He could not.

Thus when we say that the typical person in the world today is twenty times as well-off, materially, as his or her Agrarian-Age predecessor, we are saying something misleading. The typical person with today’s income would be twenty times as well off if he or she were restricted to only purchase and consume goods and services broadly available back in the Agrarian Age. But our additional range of choice–that we today know how to make more things and more types of things–gives an additional boost to our wealth today.

Now the statisticians at the U.S. Department of Commerce’s Bureau of Economic Analysis and at its sister agencies around the globe by and large cannot measure this “variety” boost to our productivity. They do try. But for the most part they fail. Thus the standard estimates of labor productivity growth in the North Atlantic–1%/year on average from 1800 to 1870, 2%/year on average from 1870-1970, 1.5%/year since and possibly slowing further in the past decade–are for the most parts estimates of how much better we have gotten at making the necessities of the world’s poor, not of how much life has been potentially enriched by higher productivity.

A good deal of this enrichment-via-increasingv-variety is truly game-changing innovations: things that transform human life. Flush toilets, automobiles, electric power, long-distance communications, modern information processing, and so forth. To provide even roughly the same capabilities in earlier eras would have been–was–absurdly, ludicrously, insanely expensive and rare. A political aristocrat in the late Roman Empire might purchase a nomenclator–a slave whose job it was to memorize names and faces and whisper to you what was the name of the person you were about to greet. Is having a smart phone more like having more like 10 or 100 or 1000 nomenclator-like assistants following you around?

Whenever we start to try to think about what opportunities economic growth will open up for humanity in the future, we need first to look back and reflect on what economic growth has done and the past. Yet I, at least, find myself stymied even in my attempt to measure how much economic growth there has been in the North Atlantic over the past 200 years. Yes, I am confident that there has been much more than 30-fold’s worth of economic growth. But how much more? And what does that mean? For that I feel I need a philosopher, to tell me who we were, who we are, and who are successors should want to be.


Consulted:


Trumpism on Trade as a Wild Goose Chase

In the United States 24% of nonfarm workers were manufacturing workers in 1971.

It’s 8.6% today.

Maybe it would be 9% if NAFTA has not been negotiated and if China had not joined the WTO, but maybe it would still be 8.6%–analysts disagree on trade expansion vs. trade diversion here.

Datawrapper bzG79 Visualize

Maybe it would be 12% if the United States had followed Japan’s and Germany’s roads of being high-savings low-currency value countries focused on nurturing their communities of engineering excellence, rather than running the Reagan and Bush 43 deficits and combining that with a focus on financialization and a strong-dollar policy. I certainly think that would have been a better policy road for the United States. But it gets you only to 12% at most–not back to 24%.

The fall from 24% to 12% is the technological tide: increasing labor productivity in manufacturing, large but not infinitely elastic demand for manufactured goods.

Looking forward we can say that by 2060 manufacturing in the United States is likely to be 6% of production workers, in which case whatever you think of what the most important parts of the value chain are, tuning the location of manufacturing labor–the people watching the robots and swapping them out when they go bad and break–is unlikely to be an important part. The thing that had been a major driver of growth in employment worldwide for two centuries–since the cotton masters of Lancashire realized the first automatic spinning machines needed a lot of labor to watch and maintain them because they were fragile–will no longer be salient in our economies. Gone with it for EMs will be the road to development that used labor cost advantage to find a niche making basic manufactures and a national champion firm that could export into that niche, and then relying on learning by doing and osmotic technology transfer to carry you forward. For today’s EMs that are not already well along the road: Strait is the gait. Narrow is the way. Many are called, but few are chosen.

As Pascal Lamy said last week: “There is supposed to be an old Chinese proverb: ‘When the wise man points at the moon, the fool looks at the finger’. Market capitalism is the moon. Globalization is the finger.”

The problems of market capitalism are broad and deep. They are not solved–they are not event addressed–by trade wars, by “renegotiating” NAFTA, by (falsely today) labeling China as a currency manipulator.

And Trump’s core supporters will not be happy if his trade policies sharply raise the prices of the goods they buy at Walmart.

(Early) Monday DeLong Smackdown: Labor Force Participation Trends

Prime age male for brad pdf

Has the Longer Depression accelerated the trend of “losing” prime-age males, crowding what would have been a generation of the trend into a decade, as I suggested at the FRBB Conference and here in contradiction to what Alan Krueger and Gabriel Chodorow-Reich were saying? No. Or, rather, you could say it looked like that as of 2013 if you thought recovery was then substantially complete. You really cannot say that anymore.

The extremely sharp Gabriel Chodorow-Reich in Email:

Gabriel Chodorow-Reich: Prime age male by 5 year age bin: “Here is a figure and a table related to our back-and-forth…

…The figure shows the LFPR over time for 25-54 year-old men split into 5 year age bins. (The data are the published BLS data with no adjustments for population controls,  I have smoothed and deseasonalized by taking a trailing 12 month moving average.) The dashed lines are the OLS trends estimated using data from 1976-2007.

What I take from the figure is that except for the 25-29 and 30-34 groups, the 1976-2007 trend fits the 2016 value pretty well.  As I said in my discussion, I’m not a huge fan of blindly taking trends and extrapolating.  But for the question of whether 2007-16 is unusual this seems a reasonable approach.  

There is a large deviation from the prior trend for the 25-29 and 30-34 male age groups.  The table, which was in my discussion slides, focuses on this group.  The plurality of the decline in participation is due to increased schooling. This seems benign.  The increase in those reporting disability is less so.  Using 2000 as a benchmark, the transition rates back into employment for this group also seem more elastic to a tighter labor market, which is consistent with other evidence.

Prime age male for brad pdf

Cf.: My earlier post:

Note to Self from Boston Harborside: Alan Krueger and Gabriel Chodorow-Reich both assure me that, to them, it does not look like the decline in prime-age male employment was materially accelerated by what I now call the Longer Depression. I don’t see it here. Are the changes in the age distribution within the category of 25-54 year olds over the past 40 years large enough to make this chart misleading? I cannot see it. I know that one disputes labor numbers with Alan Krueger (or Gabriel Chodorow-Reich) at one’s peril. But it looks to me like we were losing 1.25%/decade as far as prime-age male employment was concerned. And that in the past decade we have lost 3.25%–25 years’ worth of the trend in 10…

Employment Rate Aged 25 54 Males for the United States© FRED St Louis Fed

Brookings Productivity Festival on Friday

Real Gross Domestic Product FRED St Louis Fed

The current discussion of “slow growth in measured productivity” here in the U.S. seems to suffer from a great deal of confusion. From my perspective, there are six things going on:

  1. Since the 1920s, the rise of non-Smithian information goods…
  2. Since 1973, the productivity slowdown…
  3. Since 1995, the semiconductor-driven infotech speedup…
  4. Since 2004, Moore’s Law hitting the wall…
  5. Since 2008, what we will soon be calling “The Longer Depression”…
  6. And, remember, policy changes to speed productivity growth may well be nearly orthogonal to all of the above save (5)…

To talk about the cause of “slow growth in measured productivity” as if it is just one, not five, things causes confusion. To identify one or a small number of causes of a single thing that is “slow growth in measured productivity” causes great confusion. And then to insist that the best policy move is to undo that one or small number of thing causes even greater confusion…

The productivity puzzle: How can we speed up the growth of the economy? Friday, September 9, 2016, 9:30 – 11:00 am, Falk Auditorium: The Brookings Institution:

After nearly a decade of strong productivity growth starting in the mid-1990s, productivity growth has slowed down over the most recent decade. Output per hour worked in the U.S. business sector has grown at only 1.3 percent per year from 2004 to 2015, and growth was even slower from 2010 to 2015 at just 0.5 percent a year. These rates are only half or less of the pace of growth achieved in the past.

The United States is not alone in facing this problem, as all of the major advanced economies have also seen slow productivity growth. This slow growth has been a major cause of weak overall GDP growth, stagnation in real wages and household incomes, and it strongly impacts government revenues and the deficit.

On September 9, 2016 the Initiative on Business and Public Policy and the Hutchins Center on Fiscal and Monetary Policy at Brookings will host a forum on the policy implications of the growth slowdown. Senior Fellow Martin Baily will present an overview paper on the causes of the slowdown, followed by a panel discussion on the most effective policies to enhance productivity performance. After the panel discussion, panelists will take questions from the audience. The event will be webcast live.

Join the conversation on Twitter at #Productivity

Welcome: Louise Seiner

Paper: Martin Baily

Panel: Moderator: David Wessel

  • Jonathan Baker
  • Robert Barro
  • J. Bradford DeLong
  • Bronwyn Hall

Must-Read: Nick Bunker: How the U.S. Housing Boom Hid Weaknesses in the Labor Market

Must-Read: But I cannot help but think that the argument of this paper is fundamentally wrong:

Nick Bunker: How the U.S. Housing Boom Hid Weaknesses in the Labor Market: “The share of workers ages 25 to 54 with a job has been on an overall decline since 2000…

…This decline hit prime-age workers without a college degree particularly hard…. Kerwin Kofi Charles and Erik Hurst of the University of Chicago and Matthew Notowidigdo of Northwestern… detail the relationship between share of prime-age, non-college-educated men working in manufacturing, working in construction, and those not employed. The combined share of these three series seems to stay relatively constant at about 50 percent, with increases in construction employment offset by declines in manufacturing employment or declines in non-employment. So perhaps increased demand for construction workers during the housing bubble offset the declines in manufacturing employment. Looking at trends in employment across metropolitan areas in the United States, the three authors find evidence that the construction industry did end up hiring workers who left the manufacturing sector…. The results of this paper support the larger idea that declining employment and labor force participation among prime-age men is primarily a result of declining demand for the types of labor that many of them traditionally provided…

The first two figures in the paper show the share of non-college men with jobs holding roughly steady until 2000, and then declining:

Pubs aeaweb org doi pdfplus 10 1257 jep 30 2 179

And the number of manufacturing plus construction jobs staying roughly constant until 2000, and then declining:

Pubs aeaweb org doi pdfplus 10 1257 jep 30 2 179

Share. Number. Share. Number. The non-college male employment share held up perfectly well through 2000 in spite of the fact that the average non-college male had a smaller and smaller chance of landing a job in manufacturing-and-construction. “Declining demand of the types of labor… traditionally provided” has no effect on employment shares–until after 2000. I believe that declining demand had a big effect on the price of labor–on real wages. But I see no sign it had any effect on the chance of a non-college male getting a job.

And look at non-college women:

Pubs aeaweb org doi pdfplus 10 1257 jep 30 2 179

Lagging men by 12%-points in employment in 2000, but by 15%-points today.

I see no reason to think that there is a cross-gender cross-era thing in employment shares for shifts in economic structure that lead to a declining demand for labor in traditionally “male” sectors to explain. Slack demand and thus a broken labor market is a much better hypothesis to start with.

Must-Read: Dani Rodrik: Innovation Is Not Enough

Must-Read: Dani Rodrik: Innovation Is Not Enough: “Who can seriously doubt that innovation is progressing rapidly?…

…Technological diffusion can be constrained on both the demand and supply sides of the economy…. In rich economies, consumers spend the bulk of their income on services such as health, education, transportation, housing, and retail goods [where] technological innovation has had comparatively little impact to date…. The two sectors in the United States that have experienced the most rapid productivity growth since 2005 are the ICT… and media… with a combined GDP share of less than 10%…. Techno-optimists… look at such numbers as an opportunity: There remain vast productivity gains to be had from the adoption of new technologies in the lagging sectors. The pessimists, on the other hand, think that such gaps may be a structural, lasting feature of today’s economies…. On the supply side… when the technology requires high skills… its adoption and diffusion will tend to widen the gap between the earnings of low- and high-skill workers. Economic growth will be accompanied by rising inequality, as it was in the 1990s.

The supply-side problem faced by developing countries is more debilitating. The labor force is predominantly low-skilled. Historically, this has not been a handicap for late industrializers, so long as manufacturing consisted of labor-intensive assembly operations such as garments and automobiles. Peasants could be transformed into factory workers virtually overnight, implying significant productivity gains for the economy. Manufacturing was traditionally a rapid escalator to higher income levels. But once manufacturing operations become robotized and require high skills, the supply-side constraints begin to bite. Effectively, developing countries lose their comparative advantage vis-à-vis the rich countries. We see the consequences in the ‘premature deindustrialization’ of the developing world today. In a world of premature deindustrialization, achieving economy-wide productivity growth becomes that much harder for low-income countries. It is not clear whether there are effective substitutes for industrialization…

Must-Read: Ezra Klein: Technology Is Changing How We Live

Must-Read: Ezra Klein: Technology Is Changing How We Live: “But it needs to change how we work…

…The closest the economics profession has to a measure of technological progress is an indicator called total factor productivity, or TFP. It’s a bit of an odd concept: It measures the productivity gains left over after accounting for the growth of the workforce and capital investments. When TFP is rising, it means the same number of people, working with the same amount of land and machinery, are able to make more than they were before. It’s our best attempt to measure the hard-to-define bundle of innovations and improvements that keep living standards rising. It means we’re figuring out how to, in Steve Jobs’s famous formulation, work smarter. If TFP goes flat, then so do living standards. And TFP has gone flat — or at least flatter — in recent decades….

What Thiel can’t quite understand is why his fellow founders and venture capitalists can’t see what he sees, why they’re so damn optimistic and self-satisfied amidst an obvious, rolling disaster for human betterment…. Thiel’s peers in Silicon Valley have a different, simpler explanation. To many of them, the numbers are simply wrong…. Hal Varian, the chief economist at Google, is… a skeptic. ‘The question is whether [productivity] is measuring the wrong things,’ he told me. Bill Gates agrees. During our conversation, he rattled off a few of the ways our lives have been improved in recent years — digital photos, easier hotel booking, cheap GPS, nearly costless communication with friends. ‘The way the productivity figures are done isn’t very good at capturing those quality of service–type improvements,’ he said.

There’s much to be said for this argument. Measures of productivity are based on the sum total of goods and services the economy produces for sale. But many digital-era products are given away for free, and so never have an opportunity to show themselves in GDP statistics. Take Google Maps. I have a crap sense of direction, so it’s no exaggeration to say Google Maps has changed my life. I would pay hundreds of dollars a year for the product. In practice, I pay nothing. In terms of its direct contribution to GDP, Google Maps boosts Google’s advertising business by feeding my data back to the company so they can target ads more effectively, and it probably boosts the amount of money I fork over to Verizon for my data plan. But that’s not worth hundreds of dollars to Google, or to the economy as a whole. The result is that GDP data might undercount the value of Google Maps in a way it didn’t undercount the value of, say, Garmin GPS devices. This, Varian argues, is a systemic problem with the way we measure GDP….

The gap between what I pay for Google Maps and the value I get from it is called ‘consumer surplus,’ and it’s Silicon Valley’s best defense against the grim story told by the productivity statistics. The argument is that we’ve broken our country’s productivity statistics because so many of our great new technologies are free or nearly free to the consumer. When Henry Ford began pumping out cars, people bought his cars, and so their value showed up in GDP. Depending on the day you check, the stock market routinely certifies Google — excuse me, Alphabet — as the world’s most valuable company, but few of us ever cut Larry Page or Sergei Brin a check…. The other problem the productivity skeptics bring up are so-called ‘step changes’ — new goods that represent such a massive change in human welfare that trying to account for them by measuring prices and inflation seems borderline ridiculous. The economist Diane Coyle puts this well. In 1836, she notes, Nathan Mayer Rothschild died from an abscessed tooth. ‘What might the richest man in the world at the time have paid for an antibiotic, if only they had been invented?’ Surely more than the actual cost of an antibiotic….

‘Yes, productivity numbers do miss innovation gains and quality improvements,’ sighs John Fernald, an economist at the San Francisco Federal Reserve Bank who has studied productivity statistics extensively. ‘But they’ve always been missing that.’… Consider Google Maps again. It’s true that using the app is free. But the productivity gains it enables should show in other parts of the economy. If we are getting places faster and more reliably, that should allow us to make more things, have more meetings, make more connections, create more value….

Perhaps the best way to value the digital age’s advances is by trying to put a price on the time we spend using things like Facebook. Syverson used extremely generous assumptions about the value of our time, and took as a given that we would use online services even if we had to pay for them. Even then, he found the consumer surplus only fills a third of the productivity gap…. A March paper from David Byrne, John Fernald, and Marshall Reinsdorf… comes to similar conclusions. ‘The major ‘cost’ to consumers of Facebook, Google, and the like is not the broadband access, the cell phone service, or the phone or computer; rather, it is the opportunity cost of time,’ they concluded. ‘But that time cost … is akin to the consumer surplus obtained from television (an old economy invention) or from playing soccer with one’s children.’…

There’s a simple explanation for the disconnect between how much it feels like technology has changed our lives and how absent it is from our economic data: It’s changing how we play and relax more than it’s changing how we work and produce. As my colleague Matthew Yglesias has written, ‘Digital technology has transformed a handful of industries in the media/entertainment space that occupy a mindshare that’s out of proportion to their overall economic importance. The robots aren’t taking our jobs; they’re taking our leisure’…

Must-Read: Laura Tyson and James Manyika: Putting Profits in Perspective

Must-Read: Laura Tyson and James Manyika: Putting Profits in Perspective: “Corporate profits may be near all-time highs…

but their variance among firms and industries has also increased significantly. The most profitable firms in the US are… in sectors that capitalize on research and development, brands, software, and algorithms. Companies in sectors like pharmaceuticals, media, finance, information technology, and business services have the highest profit margins… excluding finance, these sectors’ share of US corporate profits has increased significantly, from 25% in 1999 to 35% in 2013…. In the most digitally advanced sectors of the economy, margins have grown 2-3 times faster than average. And even within these sectors, there are enormous spreads between the top-performing companies and the rest of the pack. The ‘winner-take-most’ dynamic of the digital economy is not only producing record profits for leading firms; it may be accelerating the pace of innovation and broadening the areas in which companies can enter and quickly establish market power…

Must-read: Manu Saadia: “Robots could be a big problem for the third world”

Must-Read: Manu Saadia: Robots could be a big problem for the third world: “If manufacturing is reduced to the status of agriculture, a highly rationalized activity…

…(read: employing very few people) [then] the historically proven path to economic growth and prosperity taken by Korea and China might no longer be available to the next countries. This is what keeps many economists up at night. The rise of the robots will probably reduce economic opportunities for emerging nations…. Countries you rarely hear about today, say Uganda and Tanzania, are projected to have two hundred million and three hundred million inhabitants respectively by the end of the century. What is going to happen to these people if there are no opportunities for work and wages because the manufacturing of goods has become a trivial, automated low-returns business? Not all of them will find jobs at Starbucks, regardless of how big their cities are.

It turns out that the reinvention of work imagined by Star Trek and all the social adjustments that come with it are not just some kind of pleasant philosophical exercise for overfed upper-class Western consumers of entertainment. In a world where machines produce most of the goods at a marginal cost, a just and adequate distribution of resources is a matter of life or death for billions of people yet to be born. Developed countries will or will not enact redistributive policies in the face of growing automation. The responses are well known, from progressive taxation to universal health insurance, and access to education to unconditional cash transfers, or so-called basic income. We possess stable institutions and the wealth to settle these matters adequately. Less developed countries do not yet. We are racing toward pervasive automation faster than they are catching up.