Can AI skills help older or less-educated job candidates?

The same experiment asks whether AI skills can repair a candidate's disadvantage, not only add to an advantage. Each pair set a candidate with a built-in penalty against one without: an older candidate (approximately 60) against a younger one (approximately 32), or an Associate's Degree against a Bachelor's Degree. The disadvantaged candidate, or one of the pair when no disadvantage occurred, was randomly given one of five AI skill treatments. The abstract reports that AI skills "partially or fully offset disadvantages related to age and lower education," with effects "strongest for office assistants, for whom formal AI certificates play a significant additional compensatory role." The conclusion calls this a "substitution effect" and frames AI upskilling as "an equalizer in the labor market."

The excerpt's explanation is a signal argument. The "technological obsolescence" stereotype that often attaches to older workers (Hudomiet and Willis, 2022) is countered when an applicant shows AI skills, which the conclusion says signal "that the applicant is technically current and adaptable." The paper separates two kinds of signal: traditional education "signals long-term persistence and general cognitive ability," while AI skills "appear to signal immediate readiness for technological disruption." This is the authors' interpretation. The excerpt reports the offsets but no test that separates the signaling explanation from other reasons a recruiter might favor these candidates, and it contains no mediation analysis.

The closest existing note, Can AI narrow the education performance gap?, covers the education side on a different outcome: in a randomized experiment with 1,174 adults, AI narrowed the higher-education advantage on performance, and lower-education users kept part of the gain without AI. This paper moves the same question to the hiring screen. Both point the same way for lower-education workers, but they measure different things, performance and interview invitation, so neither confirms the other. The equalizer reading also rests on a condition the authors state themselves: AI could support social mobility "provided that access to training is equitable," and the excerpt does not examine access. The sibling note on the aggregate premium gives the baseline these offsets are measured against.

The excerpt supports the offset for two penalties, one age contrast and one education contrast, in hypothetical résumés for three occupations. It does not give offset sizes or intervals, and it does not show whether the office-assistant strength holds once the cited tables are included. "Effectively neutralizes ageism" is the authors' phrase for a stated-preference result, and it claims more than a single age contrast in a survey can carry. The implication, at the strength the evidence allows, is that AI skills can change who gets an interview on a screen. That is a claim about screening, not about pay, hiring decisions or on-the-job results.