Rewriting the rules of discovery: Why negative results are essential for AI-driven science and global equity

Abstract

Scientific progress depends not only on successful experiments but also on the rigorous documentation of those that fail or produce null or inconclusive outcomes. However, the current publishing system continues to favour positive findings, creating publication bias, limiting reproducibility, and wasting valuable scientific effort. In the era of artificial intelligence (AI), this imbalance has acquired new significance, as machine-learning models trained predominantly on successful outcomes risk learning an incomplete and systematically biased representation of scientific reality. This opinion article argues that negative results should be recognized as an essential component of the scientific record and as a critical resource for AI-driven discovery. We discuss the cultural, methodological, and structural barriers that discourage their dissemination, examine their implications for scientific integrity and global equity, and propose practical measures to promote their systematic reporting, including dedicated publication formats, standardized reporting practices, curated databases, and transparent integration of AI into scientific workflows. We argue that valuing rigor over outcome will not only improve reproducibility and accelerate discovery but also contribute to a more inclusive, ethical, and reliable scientific ecosystem