Robustness, not Reproducibility, as a Criterion for AI-Assisted Systematic Reviews

A major concern with integrating AI tools in systematic reviewing processes is that they compromise the reproducibility of the reviews. This paper aims to diminish the concern by showing that traditional human-led reviews also suffer from irreproducibility problems. It further argues for an additional criterion that reviews should satisfy: that they be subject to \textit{robustness analyses}. This is only now becoming possible thanks to these novel AI tools. If this analysis is correct, it should motivate a revision of existing guidelines for conducting AI-assisted systematic reviews.