As AI capabilities grow, many scholars treat it as a universal scientific method. How accurate are these claims? Does AI outperform every technique or only some, and how is this changing? We compiled 2,507 head‑to‑head comparisons between AI and other scientific analysis methods across 27 disciplines from papers published between 2000 and early 2025. A clear dichotomy emerges. Compared with traditional statistics, AI often yields better results but at a much higher computational cost; roughly a quarter of cases show AI both more expensive and worse, a proportion that has remained stable for a decade. Compared with scientific computing, AI usually underperforms yet requires less computation. Since 2020, AI’s performance against scientific computing has improved markedly and now exceeds it in over half of the comparisons. These findings suggest AI is not a universal replacement for existing techniques but a valuable and improving component of a new AI‑enabled scientific frontier.
Review