Is AI Actually Taking Away Jobs?

Image on Unsplash

When policymakers and technology executives discuss the impact of artificial intelligence on employment, they have developed a particular narrative. It goes like this: yes, jobs will be displaced by automation, but new jobs will be created and workers can retrain for those positions. The net effect over time will be roughly neutral or even positive.

However, the data slightly contradicts this narrative. Eighty-five million jobs will be displaced by artificial intelligence between 2025 and 2030, and ninety-seven million new jobs will be created during the same period. While there will be a net gain of twelve million positions globally, there is a more troubling reality about what those new jobs actually entail.

Seventy-seven percent of the new jobs created will require master's degrees, so the workers displaced from routine roles like customer service and data entry will need not just retraining but advanced degree-level education to qualify for the replacement positions (SSRN, 2025).

A customer service representative earning roughly forty thousand dollars per year who loses their job to a chatbot cannot realistically acquire a master's degree in computer science and transition to a machine learning role, yet this is the implicit assumption in the optimistic narrative.

What makes the situation worse is that historical evidence shows retraining programs do not actually work. Evidence from the Trade Adjustment Assistance program found that workers who participated in retraining had lower employment rates two years after job loss compared to similar workers who did not participate (AI Policy Perspectives, 2025).

Four years after losing their jobs, TAA participants were underemployed and earned slightly less than non-participants. The opportunity cost of engaging in retraining actually made them worse off financially and professionally than if they had simply attempted to find new employment immediately.

The Workforce Innovation and Opportunity Act reported that seventy percent of participants were employed in the second and fourth quarters after completing retraining, but there is no control group comparison, so it is impossible to know whether these people would have been employed anyway without the program.

Retraining displaced workers is substantially harder than creating the technology that displaces them in the first place. The assumption that market forces and government programs can absorb this transition has been disproven repeatedly across multiple decades of automation waves.

By 2027, seven point five million data entry clerks will be unemployed due to AI automation, and sixty-five percent of retail positions will be automated. The jobs that replace them will require qualifications and skills that do not exist in the current workforce (Final Round AI, 2025).

Women comprise a disproportionate share of the vulnerable workforce, with fifty-eight point eight-seven million women in the US workforce occupying positions highly exposed to AI automation compared to forty-eight point six-two million men. Any retraining failure disproportionately harms women (SSRN, 2025).

The timeline for this disruption is not some distant future scenario. It is happening now, with 342 tech company layoffs eliminating 77,999 jobs already in 2025 alone. This means 491 people are losing their jobs to AI every single day (Final Round AI, 2025).

The gap between job displacement and job creation is not a technical problem that can be solved with better retraining programs, because the fundamental issue is that we are automating jobs faster than we are creating new ones. The new ones we are creating require skills we are failing to teach.