From prediction to demographic plausibility: a critical review of machine and deep learning in longevity modeling

Abstract Longevity forecasting lies at the intersection of demography, public policy, and actuarial science, yet the recent expansion of machine and deep learning methods has been markedly uneven across these fields. This article offers a selective and critical review of machine learning (ML) and deep learning (DL) applications to mortality and longevity modeling, with particular attention to the contrast between demographic and actuarial uses. On this basis, the paper asks whether a productive collaboration between demography and deep learning remains possible, and argues that it does—provided that applications are problem-driven, interpretable by design, uncertainty-aware, and validated against demographic criteria rather than predictive accuracy alone. The most promising path for demography lies not in the wholesale adoption of black-box predictors, but in the development of hybrid and demographically grounded models capable of combining flexibility, interpretability, and substantive relevance.