AI, Job Loss, and Productivity Growth
It is really painful to see the regular flow of pieces debating whether AI will lead to mass unemployment. Invariably, these pieces are written as though the author has taken an oath that they have no knowledge of economics whatsoever.
The NYT gave us the latest example on Sunday, in a piece debating how many jobs will be affected by AI. As the piece itself indicates, it is not clear what “affected by AI” even means.
What percent of jobs were affected by computers? The answer would probably be pretty close to 100 percent, if by “affected” we mean in some way changed. If by affected, we mean eliminated, then we clearly are talking about a much smaller number.
Thinking of AI like we did about computers is likely a good place to start. First of all, we should remember that there were predictions of massive layoffs and unemployment from computers and robots for decades. This did not happen.
In fact, we have a measure of the extent to which computers, robots, and other technology are displacing workers. It’s called “productivity growth,” and the Labor Department gives us data on it every quarter.
Productivity is the measure of the value of output that a worker can produce an hour. We expect this to increase through time as we get better equipment and software, we learn how to do things better, and workers get more educated.
For the last two centuries, productivity growth has been a normal feature of the U.S. economy, and in fact, most normally functioning economies around the world. This is the basis for rising living standards through time. It is the reason that we can feed our whole population, and still export food, even with just around 1.0 percent of the workforce in agriculture, as opposed to more than 50 percent in the 19th century.
The big question is the rate at which productivity grows. Productivity growth has actually been pretty slow in recent years. It averaged just 1.3 percent annually since 2006. By contrast, it averaged close to 3.0 percent in the quarter century from 1947 to 1973.
Rather than being a period of mass unemployment and declining living standards, the rapid productivity growth in that period was associated with widespread improvements in living standards. We went from depression era living standards in 1947 to a prosperous middle-class society by the end, as ordinary workers were able to afford to buy houses and cars, and send their kids to college.
We should think of the promise of AI in the same way. The first paragraph in the NYT piece warns/promises:
“In 2013, researchers at Oxford University published a startling number about the future of work: 47 percent of all United States jobs, they estimated, were ‘at risk’ of automation ‘over some unspecified number of years, perhaps a decade or two.’”
That warning is pretty vague but let’s say that we could use AI to eliminate 47 percent of current jobs over two decades. If we held GDP constant over this period, that would roughly correspond to the 3.0 percent annual productivity growth we saw during the post-World War II boom. And, just as we saw high levels of employment through the post-war boom (unemployment got down to 3.0 percent in 1969), we could maintain high employment if the economy had the same sort of rapid growth that we had in that quarter century. That will be a policy choice not an issue determined by technology.
Will Prosperity be Shared?
In the post-war boom the benefits from productivity growth were widely shared. To be clear, not everyone was doing great. Blacks were openly discriminated against, and virtually excluded from many better-paying jobs. The same was true of women, as the barriers were just beginning to come down. But the gains from productivity growth went well beyond just a small elite at the top.
Whether that happens with AI and related technologies will depend on how we as a society choose to structure the rules around AI. One reason why Bill Gates and others in the tech industry became incredibly rich was that the government granted patent and copyright protection for computer software. That was a policy choice. If we did not have these government-granted monopolies, Bill Gates would probably still be working for a living. (Okay, maybe he would be collecting his Social Security by now.)
These monopolies serve a purpose, they provide an incentive to innovate, but it’s not clear they have to be as long and as strong as is currently the case. Also, there are other ways to provide incentives. For example, the government can pay for people to do the work, as it did when it paid Moderna roughly a $1 billion to develop and test its Covid vaccine. Of course, the government also gave Moderna control over the vaccine, allowing the company’s stock to generate five Moderna billionaires in a bit over a year.
It is not hard to envision routes through which AI can lead to widespread prosperity in a way comparable to what we saw in the post-war boom. Suppose that we don’t have government-granted monopolies restricted access to the technology, so that it can be freely used.
In that world, I could likely go to a medical technician (someone trained in performing clinical tests and entering data), who could plug various test results into an AI system, and it would tell me if I have heart problem, kidney problem, or anything else. Rather than seeing a highly paid physician, I could have most of my health care needs met with this technology and a reasonably compensated medical professional, who may get less than one-third of the pay of a doctor.
There would be a similar story with legal assistance. Certainly, for standard legal processes, like preparing a will or even arranging a divorce, AI would likely be up to the task. Even in more complicated cases, AI could likely prepare a brief, which a lawyer could evaluate and edit in a fraction of the time it would take them if they were working from scratch.
People have pointed out that AI makes mistakes. There have been many instances where we have heard of AI systems inventing facts that are not true or citing sources that don’t exist. This is a real problem, but presumably one that will be largely fixed in the not distant future. We shouldn’t imagine that AI systems will ever be perfect, but the number of errors they make will surely be reduced as the technology is developed further.
In addition, it is important to remember that humans also make errors. There are few of us that cannot recall a serious mistake that a doctor made in diagnosing or treating our own condition or a close family member. A world without mistakes does not exist and cannot be the basis of comparison. We need AI to be at least as good as the workers it is displacing, but that doesn’t mean perfect.
AI and the Distribution of Income
We structured our economy over the last four decades so that most of the gains from the productivity growth over this period went to those at the top. Contrary to what is often asserted, most of the gains actually did not go to corporate profits, they went to workers at the top of the pay ladder, like CEOs and other top management, Wall Street types, highly paid tech workers, and doctors and lawyers and other highly paid professionals. These workers used their political power to ensure that the rules of the economy were designed to benefit them.
Whether or not that continues in the era of AI will depend on the power of these groups relative to less highly paid workers. Just to take an obvious example, doctors may use their political power to have licensing restrictions that prevent less highly trained medical professionals from making diagnoses and recommending treatments based on AI.
If that seems far-fetched, we already have laws that make it very difficult for even very well-trained foreign doctors from practicing in the United States. While the cry of “free-trade” was used to expose manufacturing workers to international competition, and thereby depress their pay, it almost never came up with doctors and other highly paid professionals.
Anyhow, we may well see a similar story with AI, where highly paid professionals use their political power to limit the uses of AI and ensure that it doesn’t depress their incomes. This also is an issue with ownership of the technology itself. If we don’t allow for strong patent/copyright monopolies in AI, and make non-disclosure agreements difficult to enforce, we can ensure the technology is more widely spread and cheap. This would mean that the gains are widely shared and not going to a relatively small group of Bill Gates types.
It is also important to understand how high incomes for a small group depress incomes for everyone else. Most of us don’t directly pay for our own health care. We have insurance provided by an employer or the government. However, insurers are not charities. (You knew that.)
If insurers have to pay out lots of money to doctors, then it will mean that our employers pay higher premiums, which they will look to take out of our paychecks. Alternatively, if the government is picking up the tab, there will be less money to pay for child tax credits, day care, and other good things.
Also, when the lawyers, doctors, tech workers and other would be beneficiaries from AI get high incomes, they buy bigger and more houses. That raises the cost of housing for everyone else. We can and should build more housing, but when you have a small segment of the population that has far money than everyone else, it is difficult to keep housing affordable for ordinary workers.
Anyhow, the point here is straightforward. Keeping down the pay for those at the top is not an issue of jealously. The more money that goes to the top, the less there is for everyone else, as long as we have not structured the rules in a way that takes away the incentive to be innovative and productive.
Fear the Rich, Not AI
The moral of the story is that there is nothing about AI technology that should lead to mass unemployment and inequality. If those are outcomes, it will be the result of how we structured the rules, not the technology itself. We need to keep our eyes on the ball and remember that structuring the rules is a policy choice.
And, one other point: those who want to structure the rules so that all the money goes to the top will want to say the problem is technology. It is much easier for them to tell the rest of us that they are rich and everyone else is not because of technology, rather than because they rigged the market. Keep that in mind, always.