We failed workers on trade. These are the lessons for AI.
There is a reason why we have never offered truly universal re-training. But there is also an emerging path to doing so: Registered apprenticeships.
The discussion around artificial intelligence and job displacement has focused on how many jobs will be displaced. But even small numbers of job losses can have huge effects. Consider that between 1979 and 2019 the American economy lost 7 million manufacturing jobs but gained 60 million jobs overall. That 7 million should have been a blip in the labor market. Instead, it was an earthquake that shook the whole system. The job losses were geographically concentrated, the people who lost the jobs were not the same as the people who gained the jobs, and the new jobs were not always as high quality as the old jobs. As a result, those 7 million job losses led to communities hollowing out, a crisis of masculinity, and political extremism. Even if AI brings us net job gains, something similar or worse could happen, making the manufacturing crisis look like a dress rehearsal.
Luckily or unluckily, we did have that dress rehearsal, and this time around we ought to be ready for the performance. The experience of manufacturing’s decline has taught us how to help people adjust and retrain for new jobs, and how not to. Many analysts are now focusing on Trade Adjustment Assistance, the program that from 1962 to 2022 helped workers, firms, and communities adjust to manufacturing job losses.
Trade Adjustment Assistance provided income subsidies, retraining, relocation assistance, technical assistance, and the like. Although studies show that TAA was effective in helping workers adjust, very few workers ever got the full assistance. The criteria were so tough to meet that in the first seven years of the program no workers were helped at all. TAA was targeted at workers who lost their jobs because of imports. But how do you decide if a layoff was really because of imports, or because the company was weak to begin with, or because of change in tastes, or any of the other multitude of reasons why a company can go out of business? Even after the criteria were relaxed in 1974, they were still so strict that workers usually did not receive the first TAA check until months or years after the job loss. When they did receive benefits, it was mostly income replacement, and very little went to retraining. By the 1970s labor unions concluded they had been fooled, their support for free trade purchased with a program that was empty promises, and they swung hard toward protectionism. Conservatives, for their part, pointed to the delays and lack of retraining to label TAA yet another failed welfare program. TAA was reformed over the years, but the political coalition behind it frayed, and in 2022 the program stopped accepting new applications. That is where things will remain unless and until it is reauthorized by Congress.
The first lesson of Trade Adjustment Assistance is that we need a universal retraining program. The second lesson is that such a program always remains out of reach.
The lesson analysts draw is that any program to address AI dislocation should not be so narrowly targeted. It’s simply too difficult to identify job losses that occurred as a result of one specific factor. What is needed is a universal program that helps workers who are dislocated for any reason. This is the first lesson of the experience with TAA.
But the second lesson of the experience with TAA is that those universal programs always remain out of reach. Observers have understood the targeting problem since the 1970s (some even predicted it while the TAA legislation was being debated) and have proposed universal programs of retraining again and again. The problem is that universal programs always fall to the criticism that they are too expensive. Richard Nixon tried to develop a universal program in the 1970s that included training for all workers who needed it, not just those who had lost jobs to trade, but the idea did not get far because making it generous enough to satisfy progressives upset conservatives, and vice versa. Robert Reich tried to push Bill Clinton in this direction in the 1990s, but after the failure of his universal health care program Clinton drew the conclusion that he needed to move to the middle and focus on deficit reduction. Barack Obama proposed a universal training program in his State of the Union address in 2012, but it fell to the sequester.
The paradox is that in recessionary times universal programs do not seem possible, and in good economic times they do not seem necessary. European countries with universal programs of retraining implemented them or expanded them in the golden window immediately after the Second World War, when the economy was suddenly exploding but memories of the Depression were still fresh. The U.S. missed the window because during those years it seemed more important to develop a concerted program to try to get Americans to consume more.
A universal program of worker retraining that can actually make it through today’s political gauntlet has to be self-financing to meet the cost objections. Universal and self-financing? It seems like a tall order. But there’s an optimistic piece of news that does not receive a lot of attention these days: Under the surface of our political turmoil a startling bipartisan consensus has emerged around issues of workforce training, and a path forward is appearing around apprenticeships.
An apprenticeship is a program in which the employer (not a third party such as a university or community college) trains the employee while he or she is actually working and receiving a salary. Even though the Obama administration could not get a universal retraining program in place, it did make investments in retraining, kicking off a decade-long learning process. Since then, we have learned that just starting training programs in universities or community colleges is not effective if there is no guarantee those programs will result in jobs, and programs raising awareness about job and training opportunities don’t accomplish much either. Apprenticeships are much more effective, as they are specifically implemented where there is a job already in place. An apprenticeship consists of on-the-job training as well as classroom instruction (often in partnership with universities or community colleges), resulting in a credential that is recognized by the industry. In America the system of registered apprenticeships insists on all three of these elements, to avoid turning an apprenticeship into a form of cheap labor that exploits workers, as internships can be.
Apprenticeships are, or could be, universal and mostly self-financing. They are no longer restricted to the trades where they originated; they are increasingly being used in nursing, education, finance, even in the tech industry, and studies suggest they pay for themselves in the long run, helping not just workers but also employers. The difficulty is to get employers unfamiliar with them to pay the initial costs, because firms understandably worry they will spend money to train a worker who might then leave. Once an apprenticeship system becomes established across an industry, this is less of a concern, because all employers share in the costs of training, and see the benefits for themselves. In countries where apprenticeships are widespread the government does defray the costs, but employers pay most of it. The question is how to get to that point, and here is where government can play a constructive role.
The number of registered apprenticeships in America has risen by 75 percent in the last 10 years through a series of subsidies. Apprenticeships have grown continuously through the Trump years and the Biden interregnum. Although Trump tried to push a free-market alternative to registered apprenticeships in his first term, Biden shut that down, and in his second term Trump seems to have accepted registered apprenticeships and even released an executive order to increase them.
Unlike universal basic income, apprenticeships don’t protect workers from rapid economic change, they prepare workers for change.
Do apprenticeships make sense for AI-related job dislocations? Apprenticeships work best in cases when jobs are available, but young people’s guesses about which jobs they should train for aren’t leading them to the right conclusions, that is, situations of skills mismatch. Apprenticeships solve the problem by removing the guessing game entirely: The training and the job for which it is designed come in a package. Apprenticeships also help where the worker has already made an incorrect guess and failed to find work or been laid off from a contracting industry. In this situation, an apprenticeship’s combination of training with a salary allows an older worker to meet the larger financial obligations that tend to come with age while training for a new career. AI is certain to create more of the job-search problems that apprenticeships solve, as students and workers who trained for one task discover it has been automated away.
Apprenticeships can help by redirecting labor into AI-resistant fields, not just the trades but also the “human jobs” in areas like teaching and therapy where embodied human attention is part of the job description. It’s also possible that apprenticeships could help more widely. Job loss because of AI is a result of the economy becoming more productive. Someone is reaping the benefits of that productivity, and will be spending the results of that productivity in some way that creates jobs—even if it’s just jobs in financial services to help that newfound wealth find higher returns. Because it’s impossible to predict the shape this new AI economy takes—because this economy itself is likely to shift every few years—no university or community college will be able to tell where the jobs will be, and workers themselves have no way of predicting correctly. We need to rethink from the ground up how we train workers. Attaching the worker to a firm, instead of to a specific set of tasks defined as an occupation, is a good place to start, because that firm could be doing something very different a year from now.
Unlike universal basic income, apprenticeships don’t protect workers from rapid economic change, they prepare workers for change—in the grand tradition of the most successful European welfare states—and in doing so, they enable companies and economies to change very quickly.
If job losses are severe enough, taxation of top incomes and attendant programs of adjustment will become politically easier than they have been to date. The real danger is if not enough jobs are lost because of AI—if we get a rump sector that is not big enough to support new political approaches to taxation or major new programs, but big enough to cause trouble, a white-collar equivalent of the Rust Belt—the Hoodie Belt?
So it makes sense to pay attention to, and move forward on, an approach that both parties have already ratified. Although apprenticeships have nearly doubled in the last 10 years, the number of apprentices per capita in the U.S. is still a fraction of what is found in other countries. We need to work out lots of things—what does an industry credential mean when entire industries are being threatened? How would apprenticeships work for the very small firms that are predicted to emerge in the AI era? But the need to move on this is clear. Having lived through one era of job dislocation, we don’t need AI to figure out how bad the alternatives could get.
Monica Prasad is Bloomberg Distinguished Professor of Economic and Political Sociology at the Johns Hopkins University and a Senior Fellow at the Niskanen Center.
 | A guest post by
|