When discussing tomorrow’s labor market, the growth of artificial intelligence is a good place to start. Three dimensions of AI are relevant for American work and workers. The first two are AI’s depth and breadth. The third is the media picture of AI that shapes public perception.
Frank Levy is Daniel Rose professor emeritus at MIT and a research associate in Harvard Medical School’s Department of Health Care Policy. From 2012 through 2015, he was co-organizer of MIT’s CSAIL/Economics Seminar series to bring computer scientists and economists together, the source of some of the ideas in this piece.
By depth I mean the extent to which AI equals or surpasses human intelligence—the development that worries Ray Kurzweil and Stephen Hawking.
I am not much worried about this. Computers process information by applying mathematical and logical rules to data. As the late Seth Teller noted, this makes today’s computer cognition fairly rigid, limited to solving structured problems. Even deep learning image recognition software can be fooled by small rearrangements of pixels. Humans, by contrast, are highly flexible. They move easily from solving a problem to solving adjacent problems in part because they can draw correct inferences from very little data. Until computers achieve this flexibility, all-encompassing AI is not one of my concerns.
By breadth of AI, I mean the way that software with current levels of sophistication will increasingly penetrate the workplace and displace workers. I worry about this dimension. Consider automatic speech recognition/speech generation. Computers cannot yet hold intelligent conversations on randomly chosen topics—topics for which the computer has not been prepped—but their linguistic capacity exceeds that required in many mid-skilled jobs. One mid-skilled job in United States health care involves the person in a radiology department who listens to a radiologist’s dictated report and determines the Medicare reimbursement codes for the radiologist’s time. This is a structured problem with limited vocabulary and a largely fixed relationship between spoken words and reimbursement codes. Nuance, a speech recognition company, now offers software to do this job.
Most computer-related worker displacement has been in this structured, mid-skilled range: assembly line work, clerical work, autonomous tractors that plow straight rows. The extent of the displacement is uncertain since the offshoring of work eliminates many of the same jobs (this is no coincidence). Even if near-term computerization does not expand into higher- and lower-skilled jobs, a continued loss of mid-skill jobs to computers and offshoring has the potential for significant dislocation and makes individual mobility more dependent on a good education.
The third dimension of AI—the media portrayal—is wildly excessive and it comes at a bad political moment.
In the media telling, innovations like smartphones, robotics, and the growth of the Internet represent a new industrial revolution. In reality, the economy’s main index of industrial efficiency—output per hour of work (labor productivity)—has grown weakly for more than a decade. Similarly, the media blame computers for hollowing out the middle class. But with the exception of a growth spurt in the late 1990s, the income of the average family has largely stagnated since the 1970s, when computers were of little importance.
The gap between picture and data shows how the future of workers and work depends on more than just technology. Two other determinants are the education of the workforce and the labor market institutions that help define an employee’s market power—the minimum wage, eligibility for overtime, whether fast food workers can unionize, whether Uber drivers are employees or independent contractors. These determinants depend on political decisions.
In the years following World War II, the public sector focused on the future of work. We were a young population in a post-World War II economic boom with income inequality at historically moderate levels. A broad constituency supported public-provided educational opportunity including the GI Bill and more seats in public universities. With the experience of the Great Depression still fresh, the public also supported government’s role in protecting workers’ interests through actions ranging from a strong minimum wage to John Kennedy’s confrontation with U.S. Steel.
Today we are an older population in a moderately growing economy with a high level of income inequality. A byproduct of the environment is diminished government attention to the future of work. States have reduced contributions for public universities and in-state students in particular face rising tuitions. Government-promulgated labor standards—the value of the minimum wage, the fraction of workers covered by overtime regulations—had eroded substantially before recent attempts at revival.
At this moment, the media message that robots will soon be doing all the work takes on a political cast. It says the future of work and workers is already determined by technology. Policies to improve the labor market are futile and current economic arrangements—stagnant wages, a high level of inequality—are the only arrangements possible. For persons who are doing very well, this is a comforting story. For most of the workforce, the message is demoralizing.
In periods of technology diffusion including the current period, the future of work and workers depends as much on how we deal with the technology as on the technology itself. It is time we corrected AI’s third dimension so we can return to the job of building the future of work.
For the Future of Work, a special project from the Center for Advanced Study in the Behavioral Sciences at Stanford University, business and labor leaders, social scientists, technology visionaries, activists, and journalists weigh in on the most consequential changes in the workplace, and what anxieties and possibilities they might produce.