Reskilling
Reskilling is the process of teaching workers substantially new skills so they can move into a different role, occupation or industry, as distinct from upskilling, which deepens skills for the job someone already holds. The topic gains urgency whenever a technology such as AI changes what employers need, and it is often presented as the obvious remedy for displacement. The accurate picture is more qualified: reskilling programs exist at large scale and some design types work well, but the evidence that training alone moves people into better jobs is mixed.
Programs fall into several families, with different histories and funders. Government schemes include the United States' Trade Adjustment Assistance, created in 1962 to help workers hurt by import competition, and later job-training legislation such as the Workforce Innovation and Opportunity Act of 2014. Singapore's SkillsFuture, launched in 2015, gives citizens credits to spend on approved courses. Germany's dual apprenticeship system combines classroom and workplace learning, and short-time work schemes keep people attached to employers during downturns. Employers run internal academies and tuition benefits, universities and online platforms offer certificates, and bootcamps teach narrow skills like programming in months. The funding, duration and employer involvement differ widely, which is one reason a statement that reskilling works or does not work is almost always too coarse.
The evaluation literature suggests modest average effects with large variation. Card, Kluve and Weber's meta-analyses of active labor market policies across many countries found that job-search assistance tends to help quickly, while classroom training often shows small or even negative effects in the short run — participants are busy training instead of working — and larger gains after two or three years. Evaluations of displaced-worker programs have often found disappointing results, partly because older workers with long tenure lose firm-specific pay that retraining does not restore. More encouraging evidence comes from sectoral programs that design curricula with employers in a specific industry, provide support services, and place participants into jobs in that sector; randomized evaluations of programs of this type have reported earnings gains. The pattern is that training works best when it is tied to real local demand.
Task-level change is why occupation labels mislead. A job is a bundle of tasks, and AI usually changes the bundle before it eliminates the job: drafting becomes reviewing, searching becomes verifying, and the worker's value shifts toward judgment, domain knowledge and handling exceptions. That makes adjacent moves — to occupations sharing most skills with the current one — more feasible than leaps between unrelated fields, and it argues for training that is short, frequent and embedded in the work rather than a one-time course. Workplace learning is also hard to measure, since much of it takes place informally, and published program statistics tend to capture formal courses.
Real constraints go beyond curriculum. Workers need income and time to retrain, which low-wage and caregiving workers rarely have; employers must actually hire graduates; and geography matters, because the jobs may be in a different city. Credentials matter less when hiring is skills-based, but many employers still screen on degrees. Self-selection further complicates evaluation: people who enroll in training are often more motivated than those who do not, so participants' good outcomes do not prove the program caused them. Early AI-literacy initiatives from companies, governments and schools exist, but rigorous outcome evidence for them is still sparse.
A frequent misconception is that anyone displaced can simply learn a high-demand skill such as coding. Skill demand shifts, entry-level openings in a hyped field can dry up, and learners differ in background and in what they can sustain. Another pitfall is blaming individuals for gaps that are structural, such as a shuttered local employer. Good practice pairs training with placement support, employer commitments, wage insurance or income support, and honest labor-market information about which roles are hiring.
Read reskilling as a bet that workers can be moved along the task map from shrinking activities to growing ones, a bet that pays off when training is specific, employer-backed and supported by income, and falls short when it is offered as a generic fix.