This morning at the Fed’s annual Jackson Hole symposium David Autor presented a paper on robots and people. Officially titled Polanyi’s Paradox and the Shape of Employment Growth, the paper has gotten some interesting press (see the WSJ summary – “Autor Paper at Jackson Hole: Automation Is Polarizing the Labor Market” and the more provocative Bloomberg summary – “‘Robot Overlords’ Job-Stealing Exaggerated: Jackson Hole Paper”).
Autor’s main point of the current paper is that automation is a powerful force in economic restructuring, but has its limits. He references Michael Polanyi’s insight “that our tacit knowledge of how the world works often exceeds our explicit understanding.”
A key observation of the paper is that journalists and expert commentators overstate the extent of machine substitution for human labor and ignore the strong complementarities. The challenges to substituting machines for workers in tasks requiring adaptability, common sense, and creativity remain immense. Contemporary computer science seeks to overcome Polanyi’s paradox by building machines that learn from human examples, thus inferring the rules that we tacitly apply but do not explicitly understand.
Those complementarities, however, are not to be found at the low end, creating a polarization in the labor market. In earlier work he constructs a Routine Task-Intensity (RTI) measure (see David Autor and David Dorn “The Growth of Low-Skill Service Jobs and the Polarization of the US Labor Market”). As he points out in this earlier paper, there is a displacement effect:
Critically, automation of routine tasks neither directly substitutes for nor complements the core jobs tasks of low education occupations–service occupations in particular–that rely heavily on “manual” tasks such as physical dexterity and flexible interpersonal communication. Consequently, as computerization erodes the wage paid to routine tasks in the model, low-skill workers reallocate their labor supply to [low-skilled] service occupations.
But Autor believes that
employment polarization will not continue indefinitely. While many middle skill tasks are susceptible to automation, many middle skill jobs demand a mixture of tasks from across the skill spectrum. To take one prominent example, medical support occupations–radiology technicians, phlebotomists, nurse technicians, etc.–are a numerically significant and rapidly growing category of relatively well-‐‑remunerated, middle skill employment. While not all of these occupations require a college degree, they do at least demand two years of post-‐‑secondary vocational training. Significantly, mastery of “middle skill” mathematics, life sciences, and analytical reasoning is indispensable for success in this training.
Interestingly, he puts bus and taxi drivers in the low RTI category. One would think that Google cars (and others) have shown how easily that task can be automated. However, he argues that
the Google car, unlike a human vehicle operator, cannot pilot on an “unfamiliar” road; it lacks the capability to process, interpret and respond to an environment that has not been pre-processed by its human engineers. Instead, the Google car navigates through the road network primarily by comparing its real-time audio-visual sensor data (collected using LIDAR) against painstakingly hand-curated maps that specify the exact locations of all roads, signals, signage, obstacles, etc. The Google car adapts in real time to obstacles (cars, pedestrians, road hazards) by braking, turning and stopping. But if the car’s software determines that the environment in which it is operating differs from the key static features of its pre-specified map (e.g., an unexpected detour, a police officer directing traffic where a traffic signal is supposed to be), then the car signals for its human operator to take command. Thus, while the Google car appears outwardly to be as adaptive and flexible as a human driver, it is in reality more akin to a train running on invisible tracks.
He also points out the limitations of automated assembly lines and warehouses.
He uses these examples to highlight the adaptability and flexibility of humans. One thing he doesn’t point out is the inability of automated systems to innovate, as I pointed out in an earlier posting (“Benefits of human run factory”).
I do have to quibble a bit with him putting pharmacists in the category of high RTI scores. The pharmacist profession, at least, likes to tout the importance of their tacit knowledge, such as the knowledge of their patients drug interactions.
But Autor’s points about the role of tacit knowledge are worth remembering. Almost a decade ago, I noted in a posting that the changing nature of the economy made tacit knowledge more important:
The ability to innovate and to “design a compelling experience” are the important intangible assets. Routine activities — no matter how technically sophisticated or important — will gravitate to the cheapest workforce or be automated. Key to non-routine activities is a person’s tacit knowledge as well as problem solving abilities.
That being said, we should not underestimate the impact of these changes. As Autor notes in conclusion, adaptation “is frequently slow, costly, and disruptive.”