AI has taken your employment — can retraining be beneficial?

A multitude of individuals fret that AI will assume their roles. However, a recent poll performed by the Federal Reserve Bank of New York revealed that instead of terminating employees, numerous firms implementing AI are retraining their staff to leverage the new technology. Nonetheless, there remains minimal investigation into whether current job-training initiatives are effectively assisting workers in adapting to a changing labor environment.

A fresh research paper begins to address that void. A group of scholars, including doctoral candidate Karen Ni from the Harvard Kennedy School, examined worker outcomes following their involvement in job-training initiatives under the U.S. government’s Workforce Innovation and Opportunity Act. Researchers scrutinized administrative salary records covering the periods before and after workers finalized their training. Subsequently, they evaluated workers’ earnings when shifting from or into a profession that was significantly “AI-exposed” — a term referring to the degree of tasks that have the potential to be automated, both in traditional computerization terms and via generative AI technology.

Overall, the training initiatives exhibited a favorable impact, with displaced laborers experiencing heightened earnings after transitioning to a new field. Still, the income for individuals targeting a high AI-exposed field was lower compared to those aiming for a low AI-exposed field.

In this revised dialogue, Ni elaborates on the significance of job-training initiatives as AI adoption reshapes the labor market.

---

With all the debate surrounding job displacement and AI, what prompted you to concentrate on retraining specifically?

When considering the interruptions that a significant technological advancement may cause within the labor market, it’s crucial to comprehend whether we can assist workers who might be displaced by these innovations in transitioning to alternative roles. Therefore, we focused on understanding, OK, we recognize that some of these individuals are being displaced. Now, what can job training services offer them? Can they enhance their employment prospects? Can they facilitate their advancement in terms of earnings? Is it feasible to retrain some of these workers for roles that are highly exposed to AI?

We aimed to document the transition and flexibility for these displaced workers, particularly those from lower income backgrounds. This way, we can explore how to support these individuals, whether through better investment in such workforce development or training initiatives, or modifying those programs to meet the shifting demand in the labor market.

What insights can we gain by examining data from government workforce development initiatives?

A major benefit of utilizing these trainees is that it offers nationwide representation, thus permitting a comprehensive examination of trainees across the country and capturing a significant diversity in terms of their professions and backgrounds. For the most part, our sample includes displaced workers who generally earn lower incomes, averaging around $40,000 annually. Some individuals are making significant shifts from one job to a drastically different one. We also observe a considerable number of those who end up returning to the same types of positions they previously held. We believe those workers are likely attempting to acquire new skills or credentials that could aid them in re-entering similar occupations. Some of these individuals may have been displaced from their roles due to AI. However, the job dislocation in this sample could arise from various causes, such as a regional office closure.

Could you give instances of high AI-exposed occupations versus low AI-exposed occupations?

AI exposure pertains to the degree of tasks within a profession that machines or large language models could feasibly undertake. Among our cohort of job trainees, some of the prevalent high AI-exposed professions included customer service representatives, cashiers, and office clerks. In contrast, the professions with the least AI exposure were predominantly manual laborers, such as movers, industrial truck drivers, or packers.

AI retrainability by occupation

What were your primary conclusions?

Initially, we examined the division before participants entered job training: whether they were displaced from a low AI-exposed or high AI-exposed occupation. Following training, we observed quite favorable returns in earnings universally. However, workers transitioning from high AI-exposed positions experienced, on average, 25 percent lower earnings increases post-training compared to individuals originally coming from low AI-exposed occupations.

Then we looked
“““html

at the division following job training, if they were concentrating on high AI-related occupations or low AI-related ones. When analyzed in this manner, it appears that employees generally fare better when they pursue positions characterized by lower AI exposure in contrast to those aiming for jobs that are more significantly AI-exposed. Individuals targeting high AI-exposed sectors typically encounter a disadvantage of 29 percent in terms of earnings, compared to those focusing on more generalized skills training.

Are there any suggestions that displaced employees could derive from these insights?

I would tentatively assert that our findings appear to indicate that, for these AI-exposed workers engaging in job training programs, striving for positions with lower AI exposure generally yields a more favorable outcome. Nevertheless, the observation of positive returns across all groups implies that there are likely additional elements that ought to be examined. For instance, what are the specific kinds of training they are receiving? What types of skills do they aim to develop? There exists considerable variability among the numerous job training centers nationwide, in terms of quality, intensity, and the kinds of occupations for which they can provide services. There is ample opportunity for future research to explore how these factors may influence outcomes.

Furthermore, in this scenario, the training initiative mainly caters to displaced workers from lower income brackets. Therefore, I believe we cannot make broad claims suggesting that “everyone should pursue a job training program.” Our focus was on this particular demographic.

You also developed an AI Retrainability Index to evaluate occupations that both adequately prepare workers for jobs with higher AI exposure and also provide greater earnings than their previous roles. What did the index uncover regarding which occupations are deemed most “retrainable”?

We aimed to establish a method for assessing by occupation how retrainable workers are in the event of displacement. Our ranking index reveals that, depending on their starting point, individuals may possess varying capacities for retraining into high AI-exposed roles. The only three occupational categories that exhibited a positive index value — indicating that we categorize these as highly AI-retrainable occupations — were legal, computation and mathematics, and arts, design, and media. Hence, an individual transitioning from a legal profession is more retrainable for lucrative, high AI-exposed positions compared to someone from a customer service background.

In summary, we discovered that 25 to 40 percent of occupations are AI retrainable, which, to us, is unexpectedly high. One might assume that transitioning from a lower-wage role would present significant challenges in retraining for a position with higher AI exposure. However, our findings suggest that there may indeed be substantial potential for retraining.

“`