Monday, February 12, 2018

Your job cannot be automated? Then you need to worry!


One of the greatest labor force changes of the 20th century was the movement of workers out of farming. In 1900, more than two out of five workers were in agriculture. Now it is less than two workers out of every 100.

It's not that people stopped eating. Rather, farm machinery and innovation increased the amount of food that could be produced per farm worker by more than a factor of 10. Food got cheaper and that got people to buy more food, but not 10 times as much. The end result has been fewer jobs in agriculture.

Automation is expected to come to other industries and occupations, and it is tempting to forecast less employment for them too. A variety of studies are using engineering information to determine which jobs will be automated next.

While automation may be a question of engineering, job loss is even more a question of economics. A key part of the agriculture story is that people were unwilling to purchase all of the food that farmers were capable of producing, even though food was getting cheaper. But not all industries share this with agriculture. 

Suppose that the automation in agriculture had only been for chicken farming and not for any other food production. Chicken would have gotten cheaper relative to beef, fish, vegetables, fruit, etc., and that would have caused people to buy more chicken and less of other types of food.

Many -- even most -- of the extra chickens produced would have been purchased by consumers, and there would have been less need to reduce employment in chicken farming. 

The most dramatic job losses would have occurred in the food industries like beef and fish that were not automated and that compete with chicken. In other words, jobs that are difficult to automate from an engineering perspective may be exactly the jobs pushed to extinction by automation because they cannot compete.

It all depends on the competitive landscape and how willing are consumers, encouraged by lower prices, to absorb the extra output made possible by automation.

Trucking is a modern example, because engineers are predicting that machines will soon do a lot of the driving formerly done by trucking employees. But the result may be more jobs for people in trucking and fewer jobs for people in railroads, airlines and shipping that compete with trucking (unless they also get more productive at the same time that trucking does).

Another example has occurred in my own profession: Two or three generations ago, a large fraction of economists were employed manually performing the arithmetic of statistical analysis. Then, computers came along to automate that arithmetic, without really automating the tasks done by theoretical economists.

The result was an increase in the fraction of economists doing statistical work, because universities, businesses and government wanted more statistical analysis when computers made it became cheaper and more accurate. The fraction of economists doing theoretical work fell, precisely because their tasks were not automated.

So the more interesting economic question for a worker is not whether his job can be automated but whether he or she will miss out on automation to occur in the workplace of his or her primary competitors.