We must diversify our workforce bets
An all-of-the-above approach to retraining is the best way to prepare for the age of AI.
To best support unemployed workers in the age of artificial intelligence, we need to make two basic moves: Strengthen unemployment insurance so that when a paycheck evaporates, a higher percentage of it can be replaced, and improve our training systems so that a new job with a comparable paycheck will be within reach.
Getting the retraining pathways right will be a particularly important and difficult task, and this is an area where policymakers should avoid the lure of placing a single big bet.
Unemployment insurance is theoretically simple: It’s a cash benefit to eligible workers that the government should be able to deliver with minimal friction, assuming we reform the byzantine system that now prevails. Job training, and workforce development more broadly, are a different story. This type of support is not cash but in-kind, and it comes in different forms, from help crafting a resume to career guidance informed by labor-market information to a full retraining. (When direct funding is provided, it must be used for training[1] ). Such retraining can be provided at a community college, a government agency, or directly by an employer, and can result in a degree or a credential. Of course, different industries require different skills and expertise – which is to say that it is critical to offer a variety of training options.
Right now, I’m worried that policymakers will place sizable wagers on narrow training pathways with little flexibility to adjust course. This inflexibility could take two forms: an inability to shift funding according to which industries have growing demand for labor and which types of training are found to work best over time.
We hear a lot of worry that AI will wipe out white-collar jobs, but we only have the faintest understanding of what that might look like. There’s not even consensus on whether AI has driven recent employment trends. Exposure metrics merely indicate what share of tasks in a given occupation might be doable via AI, not necessarily whether the collection of tasks will be augmented or completely automated. In fact, exposure could translate into greater demand for that type of human labor. We do have projections on which sectors are expected to have employment gains – notably healthcare and social assistance – but where exactly labor demand shifts remains an open question. It would be a shame for policymakers to privilege job retraining for healthcare roles, only for the funds to be locked in place as larger growth sectors emerge elsewhere.
Likewise, we shouldn’t put all of our eggs in one type of training basket. Congress should set aside funding for a range of short and long-term training pathways, and ensure resources can be shifted when interventions are found not to work. Monica Prasad argues that apprenticeships should get pride of place. That is an option I would like to see us expand, but it’s no silver bullet: This type of intensive training might work for some workers, employers, and industries, but not for others. Apprenticeships can also be expensive, making them a difficult route to depend on for each job transition. A proven non-degree credential earned in a shorter period of time might be enough for a worker to secure horizontal transitions or wage gains in a new field.
We lack the data to readily evaluate what works and why. But once we figure it out, we also must be able to act on that information. Siloed programs and funding make that too hard.
Part of the problem is that we are often flying blind. Workforce programs receiving congressional appropriations may lack sufficiently rigorous evaluations, and portions of the labor force – such as workers in rural areas – are not properly tracked in major data collections. Even for the swaths of the labor market we do get more regular insights on, the picture is incomplete. In general, the quarterly wage records collected by unemployment agencies that inform key federal data sets do not include job title, job location, or hours worked. A small number of states have begun adding these categories to their collections, but a broader national effort is required. Those are the sorts of administrative data that would help analysts to assess different training programs and interventions.
As an example, recent research with more granular data on Trade Adjustment Assistance has demonstrated that the program did in fact generate positive returns for the workers who made it through the gauntlet of eligibility determination and received job training. This result runs counter to many years of conventional wisdom. But additional work remains to determine precisely why the training was useful. Workers receiving TAA training did not just end up in new industries with better wages than counterparts denied the same support; they were also much more likely to switch into new commuting zones. To unpack those sorts of dynamics for a broader set of workforce programs, we must add occupational and geographic data to our prominent administrative records.
When we figure out what works and what doesn’t, we also must be able to act on that information. Efforts to restore our workforce system, which has been starved for funding, must include measures to ensure those investments must go to their best possible use. Funding must be flexible enough for policymakers to continually shift resources away from efforts that fall short, reinforce what works, and test out new interventions. Our large set of siloed and rigid funding streams makes this difficult in practice. Dozens of labor programs are scattered across government agencies and can be quite hit or miss (emphasis on the latter). Eliminating one set of funds does not guarantee a similar increase for another workforce program. Congress should consolidate and simplify programs to enable those shifts and provide workers with straightforward access to proven training.
That access is most valuable when workers want the support provided. Reform efforts will amount to little if there is not enough freedom to choose. Individual workers may want roles in healthcare, but they might not. Apprenticeships may be intriguing to some, while others will prefer to take courses at their local community college. The government should open up a range of promising opportunities, provide detailed information on each pathway, and let workers pick for themselves.
Don’t force a square peg into a round hole. If policymakers double down on narrow, inflexible training pathways, they will only succeed in wasting resources and leaving communities behind. An all-of-the-above approach to retraining—grounded in rigorous data and driven by worker choice—is the best way for Congress to prepare for an economy increasingly driven by AI.
Will Raderman is Senior Policy Advisor at the Searchlight Institute.
 | A guest post by
|