AI versus Human Grant Review in 2025: Structural Limits of Human Peer Review and the Rationality of AI-based Validity Screening

This preprint analyzes the structural limitations of human peer review in competitive research funding systems and evaluates the rationality of introducing AI-based validity screening as of 2025. The paper demonstrates that human reviewers face inherent constraints—such as cognitive load, inconsistency, COI-driven exclusion of domain experts, and limited capacity to detect structural or logical defects in proposals—that cannot be resolved through procedural reforms alone. Using a systems-level framework, the study argues that AI models available in late 2025 already surpass human reviewers in the domain of validity checking: detecting logical inconsistencies, missing methodological components, numerical incoherence, and discrepancies with applicants’ past research records. The paper proposes a hybrid evaluation architecture consisting of AI-based validity screening, dual editorial oversight, and human expert judgment restricted to value-based assessment. The analysis shows that this hybrid system increases transparency, fairness, and reproducibility while reducing reviewer burden and noise. It also explains why such a system remains robust under future advances in AI capability. A policy appendix discusses the implications for national research funding agencies, including the rational rebalancing of competitive and baseline funding once AI-based quality assurance is implemented.