David Mumford

Archive for Reprints, Notes, Talks, and Blog

Professor Emeritus
Brown and Harvard Universities

skip to content

The Dismal Science and the future of work

July 20, 2015

Economics is an area that is built on mathematical models that simplify highly complex phenomena. People often forget that, as a result, economic models omit human and historical factors that are fundamentally non-mathematical and outside its scope. Thus the impossibility of building mathematical models of human psychology undermines that basic building block of economics, the "rational economic agent". But I also want to argue that advances in technology are transforming society in ways not dealt with in economic models, by altering the need for most human work, another foundation stone of economics. My thesis in this post is that, in addition to dealing with the Malthusian constraints caused by population growth, the next 50 years will see the growth of a nearly completely automated society that requires only minimal work from the large majority of its citizens. Such a development destroys the basic axioms on which economics is built, not to mention the basic structure of human lives. How in heaven's name will we adjust to such a "gift"?

As a general matter, it is vital to realize that all mathematical models are idealizations. Newton himself realized that modeling planetary motion by his law of gravitational attraction left out tidal effects. But every neat mathematical model can and usually is pursued by pure or applied mathematicians who push it to extremes. Thus some have studied what pure gravitational forces do to planets billions of years from now when this approximation ceases to be relevant. to actual events. I have done the same in my own research pursuing a beautiful mathematical model for comparing shapes of 3D objects to elicit its mathematical secrets, way beyond its relevance to detecting diseased organs in a human body. As for economics, Keynes himself wrote (in "The General Theory of Employment, Interest and Money", Ch. 21):

Too large a proportion of recent 'mathematical' economics are merely concoctions, as imprecise as the initial assumptions they rest on, which allow the authopr to lose sight of the complexities and interdependencies of the real world in a maze of pretentious and unhelpful symbols.

I am not an economist nor have I even taken basic courses in economics. My understanding is that of a bystander living in a sea of economic events and trying to make sense of them. So when I heard of the experiments of Kahneman and Twersky on the irrationality of economic decisions made by average people, it was an "aha" moment on the limitations of economics. Here is one of their experiments: subjects do some task leading to a reward. They are then asked to choose between two possible rewards: $10 or a good quality pen and pencil set. Most chose the money. A second set of subjects were also offered a reward but now they had three choices: $10, the previous pen and pencil set and a clearly inferior mass-produced pen and pencil set. Now most chose the good quality pen and pencil set! So much for acting rationally on the basis of fixed personal values attached to all goods, as "rational economic agents" should. Kahneman has gone on to show the pervasive effects of emotions on the simplest of tasks.

For some time I have been wondering, reading headlines and so on, about employment, off-shoring, robotics etc. I thought: why don't I try to find some data myself and see what's up. Through the wonders of google and the internet, I found a table put out by the Bureau of Labor Statistics on how many people in the US work at one of hundreds of categories of jobs: Employment by detailed occupation. Naturally, one first checks out ones own occupation: out of a total of 145,355,800 working people in the US in 2012, 63,300 are college math science teachers, that is one in 2300 or about 0.04%. Then I checked my wife's occupation, classified as a "fine artist" (note that the table includes self-employed people, presumably through their schedule Cs as well as those with an employer): only 28,800 (this includes painters sculptors and illustrators), one in 5000 or 0.02%. I'm sure if you include struggling artists with either no art related income or undeclared income (!), there would be easily as many as us college math teachers. But now the figure I had started looking for: agricultural workers. This would have been nearly 100% at the time when the USA came into existence. Now it is 815,500 that works out to be 0.56% of the entire working population! As for fishing, which is the main occupation in my town of Tenants Harbor, Maine, there are 31,300 workers, just about the same numbers as fine artists. My guess is that allowing for imported food, we are fed by around 1% of our workers. I suppose the main driver for this is the use of massive machines to plant, irrigate, fertilize and harvest huge fields. Wow -- who would have expected this two centuries ago. If we lived a simple life like that of the first settlers (but with machines), could most of us sit back contemplate nature all day long?

How about the machines? Do we need many people to manufacture these for us? There has been a precipitous decline in manufacturing jobs too, caused first by off-shoring but more recently by the jobs coming back to the US but now being carried out by robots. While I was doing research in computer vision, I used to answer questions about its usefulness by saying that computer driven cars would be coming in a decade or so. And now they are here. Both the needed analysis of perceptual data and the control of the vehicle can now be done at a human level by computer. This convinces me that full industrial automation is now possible and that at least all the necessities of life can in principle be supplied by a very small percentage of the population.

How about the non-necessities: the overhead of a capitalist economy, health, education and fun stuff? With online shopping, stores become a dispensable luxury and, I have to admit, I prefer Amazon to going to a mall. Hard also to admit but the guts of my job can be dispensed with too: professors can easily be supplanted by MOOCs (massively online courses, prerecorded from the best lecturer in the country). The full gory picture is detailed in Martin Ford's recent book The Rise of Robots. He conjectures that a large percentage of white-collar jobs can be carried out by AI-type programs. IBM's success in winning Jeopardy! by computers with access to big datasets would seem to confirm this. We seem to be facing long term structural unemployment and that this will get progessively worse as computers take over more and more tasks. Now all this flies in the face of economists' orthodoxy. Some years back, as we were struggling to emerge from the great recession, I asked the economist Roland Fryer whether he didn't feel there just weren't enough jobs to ever recover. His answer: new jobs have always turned up as the economy adapts though you can't predict where. The question, as I see it, is whether the present is really different from the past and whether now machines are taking over for good?

If indeed it is going to be hard to find jobs for everyone who wants one, one need only look to France to see a simple near-term solution. In the US, many have been used to thinking of France as a decadent society, past its "gloire" and now falling behind the US and Germany who still believe in old fashioned hard work. Indeed, the official work week in France is 35 hours, not 40, and every worker is entitled to 5 weeks of paid vacation. I would contend that actually they know what's up better than us. Is it not rational if all the food and all the goods we need can be produced by fewer and fewer people that we should enjoy life more -- work less per week and take a decent vacation? We, the people of advanced countries in the 21st century, are wealthy -- or ought to be. It seems to me that the French have got it right. Of course, all this brings in the ridiculous level of inequality in our society. Again, it was a Frenchman, Piketty, who wrote the book making this utterly clear. If the number of working hours per year were reduced in the US, businesses would hire more people to do the needed work and this would decrease unemployment and be a first step in equalizing things as well.

Looking further ahead at the possibility of a society that is 90% mechanized and computerized, there is a bigger problem. Martin Ford proposes in the book mentioned above that a guaranteed minimum wage is the best long-term solution, allowing many to get by without work. But I would ask: what would this do to a person's self-image? A few would be creative, a few would take drugs, a few could fight wars but most people find their worth in doing a good job, earning their livelihood and raising a family. This pattern of life seems built in to our genes. It seems to require life-long silver spoon training to handle a life of leisure with any aplomb. There is a sci-fi literature on societies with no work and their imaginings are not pretty -- mainly the result is that people get into fights over nothing. And this challenge may be dwarfed by another: global warming that looks on track to generate intractable refugee problems. I wish my grandchildren were growing into a pleasanter world.

Thomas Riepe wrote me directly, noting how prescient an old article of Stanislaw Lem is:

Concerning the automatisation issue, I guess Lem's story turns out to become kind of correct: http://www.newyorker.com/magazine/1981/10/12/phools

My former colleague Alan Yuille wrote me also pointing out a startling even earlier precedent: John Maynard Keynes, no less, predicted in 1931 that the world was becoming so wealthy that a 15 hour work week would suffice to keep our economy going and should be expected to become the norm (see Essays in Persuasion, Norton, 1963, pp 358-373).

Moreover, he directed me to to the analysis of employment carried out by his colleague Uday Karmarkar in the UCLA School of Management. Summarizing the Foundations and Trends article entitled "The U.S. Information Economy", (vol.6,m 2012), Karmarkar, Apte and Nath divide the economy into 4 super-sectors:

  • Material Products (e.g. steel, cement, autos, consumer goods), a meagre 7.5% of employment
  • Material Services (e.g. Transportation, retail and restaurants, construction), 44.7% of employment
  • Information Products (e.g. computers, books and magazines, databases, music CDs), 2.8% of employment
  • Information Services (e.g. eduction, professional, financial, management), 45.0% of employment, now the biggest sector
Their main point is the steady growth of the last two, especially the last, as a percentage of the economy, so that we are now truly an "information service economy". When and if AI-style technology begins to out-perform humans in pieces of this sector, we can expect tremendous disruption.

My good friend Professor Shiva Shankar from the Chennai Math Institute wrote me a strong rebuttal, building on what is happening in India:

Dear David, in your latest blog 'The Dismal Science and the future of work' you say 'the next 50 years will see the growth of a nearly completely automated society that requires only minimal work from the large majority of its citizens'. I wish that were true, but I fear quite the opposite. As the environment deteriorates, humans are being increasingly involved in elementary labour, for instance in pollination! In the report below, it says 'For specific tasks mainly allocated to women and children -- especially the labour-intensive cross-pollination --wages are paid that are substantially below the official state or zonal minimum wages'.

It is no brave new world we are hurtling towards!

Best wishes. Shiva.

"Almost half a million Indian children are working to produce the cottonseed that is the basis for our garments and all the other textile products that we use. ... Children below 14 -- of which two-thirds are girls - are employed in the seed fields on a long-term contract basis through loans extended to their parents by local seed producers, who have agreements with the large national and multinational seed companies. Children are made to work 8 to 12 hours a day and are exposed to poisonous pesticides used in high quantities in cottonseed cultivation. Most of the children working in cottonseed farms belong to poor Dalit ('outcaste'), Adivasi (tribal) or Backward Castes families. ..." (Report by the India Committee of the Netherlands, www.indianet.nl/english)

Added in October: I came across a set of charts in the June 2015 Atlantic entitled "Are We Truly Overworked?" that gives quite relevant data on hours worked and seems to me to substantiate Keynes' views that we simply don't need to work so long every week to achieve our current level of prosperity. The most striking figure, for me, is that in the period 1950-2012, workers in Germany decreased their average total hours worked per year by 991 hours! In the US, by contrast, we decreased our annual total by merely 200 hours. In fact, in the US, men with college degrees worked 2.5 hours more per week in 2010 than in 1988. What a twisted effect of wealth and progress is this?

Added in November: Alan Yuille just sent me this link to an article in the Guardian newspaper soundng the same alarm.