AI narrows skill gaps in test tasks — but does that actually shrink pay gaps between workers?

This explores whether AI is shrinking the gap between lower- and higher-skilled workers right now. The short version is that the strongest evidence in the collection shows AI closing *performance* gaps on specific tasks. None of the notes here directly measure *wages* closing. That distinction matters, because a skill gap that narrows in an experiment doesn't automatically become a pay gap that narrows in the labor market.

The performance evidence is real and fairly striking. In a study of 758 BCG consultants, those in the bottom half gained the most from GPT-4, while top performers improved less Does AI help lower-skilled workers catch up to top performers?. A randomized experiment with 1,174 adults found generative AI erased about three-quarters of the advantage that higher-educated participants had on a business problem-solving task. Some of that gain stuck around even after the AI was taken away Can AI narrow the education performance gap?. There's a catch, though. The BCG study *required* AI use, and the authors warn that when people choose for themselves whether to use it, the equalizing effect may reverse. A related note finds that productivity gains show up when workers apply skills they already have, and disappear when they use AI to learn new ones When does AI actually boost worker productivity?.

On wages specifically, the corpus offers models, not measurements. Acemoglu's model predicts AI's wage effects will be spread more evenly across groups than past waves of automation. The same model predicts capital income will pull ahead of labor income overall, so even an even split among workers could sit inside a bigger split between workers and owners Will AI increase inequality less than past automation?. Agrawal, Gans and Goldfarb go further. In their model, early AI helps weaker workers by covering for poor execution. As the tools improve, the value shifts to workers with better judgment, and the equality gain bends back into a U-shape Does AI's wage-equalizing effect actually reverse as it improves?. If they're right, today's catch-up effect could be temporary.

The labor-market data points the other way in places. Payroll records show no economy-wide job losses, but young workers in AI-exposed jobs are being hired at 19% lower rates, while experienced workers see no such gap Is generative AI displacing workers at economy-wide scale?. That pattern favors experienced workers, the opposite of what the experiments suggest. Demand for AI skills is clustering around a technical core (Python, SQL, machine learning) instead of spreading across occupations Is AI creating common skills across jobs or deepening divisions?. Firms with more internal AI capability are also replacing contract workers faster and more cheaply Do firms substitute labor for AI at different rates?. Exposure differs by gender too: in female-dominated fields it reaches lower-paid workers just as much as higher-paid ones Does AI exposure hit low-wage workers harder in some fields?.

The conclusion that ties this together is that the outcome isn't built into the technology. A cross-domain review finds that generative AI can widen or narrow inequality depending on who has access, how it's built into work, and what incentives surround it Does generative AI inevitably worsen or reduce inequality?. Brynjolfsson argues the deciding choice is whether AI is designed to replace workers or to help them, because replacement shifts bargaining power, and so wages, toward whoever controls the tools Does automating labor concentrate power more than augmenting it?. So an honest answer to the question is that AI *can* level skill differences inside a task. Whether that ever shows up in paychecks depends on choices about how it's deployed, and the corpus doesn't yet have the wage data to say it has.