The illusion of innovation: Why the exponential growth of artificial intelligence publications may not reflect epistemological progress in health sciences

Objective. The goal is to critically examine whether the exponential growth of artificial intelligence publications in health sciences reflects genuine epistemological and translational progress, or if it predominantly represents a process of bibliometric expansion characterized by inferential fragility, methodological heterogeneity, and limited clinical consolidation. Design/Methodology/Approach. A narrative review was conducted from meta-research, epistemological, and evidence-based medicine perspectives. The analysis focused on recurring structural issues in contemporary artificial intelligence literature in the health sciences. These issues include publication inflation, predictive reductionism, limitations in reproducibility, dataset fragility, constraints in external validity, and translational instability. Additionally, the review examined how current evaluative paradigms may conflate bibliometric visibility and computational performance with scientific robustness and clinical relevance. Results/Discussion. The contemporary expansion of artificial intelligence research in the health sciences is accompanied by significant methodological, inferential, and translational challenges. A review of the literature suggests that high predictive performance can coexist with heterogeneous datasets, limited external validation, constraints on reproducibility, and context-dependent analytical conditions. Additionally, the growing focus on predictive optimization may lead to the prioritization of computational performance over causal interpretation and mechanistic understanding. Together, these findings raise questions about whether the current growth in publications reflects proportional epistemological and clinical progress. Conclusions. The expansion of artificial intelligence research in the health sciences should not be uncritically interpreted as evidence of proportional scientific consolidation. Studies that integrate meta-research, including reproducibility, inferential robustness, interpretability, external validity, and translational applicability, are necessary to distinguish genuine innovation from quantitative growth without sufficient epistemological maturity.