Economists’ working paper examines AI’s implications

In times like these the Federal Reserve chairman has to think like an entrepreneur. The chairman finds himself in a situation like many entrepreneurs: a decision must be made and the information available isn’t adequate or conclusive.

Just like an entrepreneur, the chairman reviews the data and other information, seeks out the often-conflicting opinions of trusted advisers, seeks other sources, then checks out what others — central banks in this case — are doing, and, then ultimately make the decision and trust his or her own judgment.

Politicians and financial markets are awaiting the Federal Reserve’s monetary policy decision on interest rates. At issue is whether our economy is faltering and needs stimulation through lower interest rates or is humming along and adding stimulus would then just stoke price inflation.

One example of inconclusive information is found in the jobs data, including the employment and unemployment numbers.

The jobs numbers will undoubtedly play an important role in the Fed’s decision since they reflect the current health of the economy and the confidence that both businesses and individuals have in the future. Given the lack of precision in the unemployment numbers, though, their importance as a decision factor is a good reason why human wisdom, experience and entrepreneurial skills are needed in economic policy decisions.

The jobs picture has never been a simple matter. The primary reason is that the size of the labor force is such a flexible number. As a practical matter, the government had to define that number to include only those with jobs and those actively seeking jobs; the employed and unemployed. There are ongoing efforts to come up with solid definitions of those without jobs who are not actively seeking employment but that remains an elusive goal. Some of them are discouraged job seekers, surely, but measuring discouragement, let alone estimate its duration, is neither easy nor practical.

Another complicating factor is the impact of automation — including machines, artificial intelligence — on our economy and employment.

The relationship between automation and jobs is both simple and complicated. For the displaced worker it is painfully simple: “The robot took my job.” At the macroeconomic level where everything is tallied up into our Gross National Product, it is a more complicated matter.

The complications arise because automation changes worker productivity, wage levels, employment levels and labor’s share of income in different and often interactive, ways.

Economists do not yet have a clear picture of automation’s role or its effects on jobs. As just one example of that, we’ve known that automation creates jobs as well as eliminates them ever since the days of huge IBM computers and punch cards. One of the first things to be automated was payroll. The initial result was that roomfuls of payroll clerks were laid off, but then platoons of technician and programmers — new jobs — were necessary to keep the computer happy. The net effect, though, on the economy and on jobs, was and is, uncertain.

Two researchers, Silvain Leduc and Zheng Liu, at the San Francisco Federal Reserve Bank, decided to investigate whether the data they had on automation could be integrated into standard economic forecasting models.

Their report is entitled, “Robots or Workers? A Macro Analysis of Automation and Labor Markets,” One of the findings was that, “…automation has contributed to lowering real wages since the early 2000’s, with particularly pronounced effects after the Great Recession.” This provides a much-needed explanation of why the recovery, particularly during the past few years, has not been accompanied by wage growth or wage-driven inflation.

Another key finding was that automation increased the volatility and exaggerated the up and down swings in employment and unemployment.

The findings of this research team could be a breakthrough in economic policy decision making. Instead of having scattered, fragmentary and sometimes conflicting jobs reports that require an entrepreneurial agility to interpret, we could have an integrated economic forecast model.

An integrated forecast model is particularly useful in examining contingency policy choices. As AI and automation usage expands that would allow the Federal Reserve to anticipate economic changes rather than reacting to them.

Working papers in economics do not usually arrive with a brass band. They call them working papers because they are the first drafts of accepted economic theory. While the San Francisco researchers carefully documented their work and backed up their conclusions, there is still a lot of vetting to go before their breakthrough work finds its way into the economic policy decision rooms.

Maybe there should be a brass band, though. Economists have been struggling for years to get a handle on automation’s impact on our economy — not just anecdotes but hard data. This working paper is a good first step.