Does generative AI inevitably worsen or reduce inequality?

The strong claim — that generative AI will inevitably widen or inevitably narrow inequality — is what the evidence does not support. A state-of-the-art interdisciplinary review across four information-intensive domains finds the same two-sided structure in each. In information, AI can democratize content creation and access yet dramatically expand misinformation. In the workplace, it can boost productivity and create jobs yet distribute the benefits unevenly. In education, it offers personalized learning yet may widen the digital divide. In healthcare, it can improve diagnostics and accessibility yet deepen pre-existing disparities. Every domain carries an explicit trade-off that complicates any a priori hypothesis about net effect.

The takeaway is that the inequality outcome is deployment-contingent, not predestined by the technology. The same capability that democratizes can also concentrate, depending on who gets access, how the tool is integrated, and which incentives govern its rollout. This cuts against both techno-optimist and techno-pessimist framings, which each pick one branch of the trade-off and treat it as the whole story. The practical consequence is that the locus of control sits with deployment choices and policy, not with the model. It also reframes the productivity findings elsewhere: a tool that lifts immediate output but devalues foundational learning can raise inequality precisely by helping the already-skilled more than the learning novice. If outcomes are contingent, then the question is not "what will AI do to inequality" but "what are we choosing to do with it."

— "The impact of generative artificial intelligence on socioeconomic inequalities and policy making", https://doi.org/10.1093/pnasnexus/pgae191