For decades, automation has been a rough road for middle-skill workers. These jobs used to provide plentiful, family-supporting employment opportunities for those with only a high school education or even less. Robotics and, to a lesser extent, trade dramatically reduced the number of middle-skill jobs leading to what economists called a “polarized” labor market: many high-skilled and low-skilled jobs but relatively few that require a middle level of skill. The big question for economists now is how the job market will react to the approaching wave of AI-driven automation. Will AI continue to raise the premium on education and skills while moving those with middling skills down the wage scale?

According to a new analysis, the answer is more complex than the AI “substitution” (human replacement) or “augmentation” (human enhancement) camps would have us believe. For these authors—University of Toronto researchers Ajay Agrawal, Joshua Gans, and Avi Goldfarb—the focus is not on whether machines replace or augment human skills but on how increased computerization will reconfigure how human beings are deployed in the economy. They find that AI could democratize skills, unlocking job opportunities for lower-skilled workers, raising the “labor share” of the economy, and reducing the value of currently rare but narrow, more easily automatable skills. Playing on Erik Brynjolfsson’s “Turing trap” work, which posits the drive for human-level AI will result in mass disemployment, these authors argue we face a “Turing transformation” wherein automating specialized skills spreads capabilities more evenly across the population, benefiting the middle of the economic distribution mainly at the expense of the highly educated and skilled.

Agrawal et al. illustrate their findings with examples. They discuss how GPS technology has automated the highly skilled work of London cab drivers by putting an electronic version of their hard-earned “knowledge” (that’s the name of the test London cabbies have to take: The Knowledge) at the fingertips of everyone with a driver’s license. At the high end of the labor market, they speculate that AI could revolutionize medical diagnosis in much the same way: no need to go to medical school and work through a hospital residency to learn how to differentiate complex medical conditions when an algorithm can do that work by itself. Closer to home (for your humble correspondent), as large language models advance, workers who have spent decades honing writing skills face competition from LLMs that can produce and hone text in seconds. Lower-skilled writers can use these technologies to achieve far more than they could previously, and so narrow the gap between them and more skilled writers.

So, what’s the effect of this skill democratization on the labor market? By opening more types of jobs to more people, the study authors argue,

task automation, especially when driven by AI advances, can enhance job prospects and potentially widen the scope for employment of many workers . . . [thereby] improv[ing] the value of the skills of many workers, expand[ing] the pool of available workers to perform other tasks, and, in the process, increas[ing] labour income and potentially reduc[ing] inequality.

In other words, AI could deliver on a platter knowledge and skills that would otherwise take considerable time, effort, and money to acquire, potentially broadening and strengthening our human capital base and freeing up highly talented people to create even more. Who says we can’t have nice things?

There’s a bit of economic sleight of hand going on in the authors’ scenario. It supposes that highly skilled physicians who are replaced by, say, AI-supported nurse practitioners, will find other, higher-value work to do. It also suggests that the first-order effect would probably be to put downward pressure on physicians’ wages—since wages would be spread over a larger group of providers who can afford to work for less than a doctor with med-school debts. Like our London cabbie who used to spend four years studying the twists and turns of a medieval street system, physicians—workers who invest a decade or more of their lives to understand the intricacies of the human body—might face stiff competition from lower-skilled but AI-enabled competitors.

Or perhaps not. Given that we already have shortages of health care workers at just about every level, maybe AI will deliver a badly needed boost in productivity among health workers at the exact moment when aging populations and new treatments are flooding hospitals and doctors’ offices. Based on our relatively flat population-growth trends, the biggest problem our growing economy may face in the future isn’t too much automation but too little.

When it comes to projecting AI impacts on skills and jobs, we are in the arena of the highly speculative. Neither our fondest hopes nor our worst fears are likely to be realized. Predictions are hard, they say, especially about the future.

Read the full article here.