Causal Mechanisms of the Gender Pay Gap

AI systems increasingly participate in cultural interpretation, not only in prediction or automation. This paper treats neural causal modeling as an interpretive technology for studying the gender pay gap: a way to ask how an administrative category such as gender becomes connected to wages through culturally patterned institutions such as occupations and industries. Using $1{,}057{,}573$ full-time workers from CPS-MORG, we compare a neural structural causal model (NSCM) with OLS and Oaxaca--Blinder baselines. The most robust finding is that the direct wage penalty remains near 30\% across methods, while measured indirect pathways through occupation and industry are smaller. We deliberately do not frame the largest NSCM total effect, 52.6\%, as the central claim because it is sensitive to a strong graph constraint. Instead, the contribution is a culturally attentive framing of causal AI: variables such as occupation should not be treated as neutral controls, binary gender categories should not be mistaken for lived gender, and high-capacity causal models should be read alongside social theory about sorting, norms, and institutions.