AI Can Broaden Science If Institutions Reward Novelty

A commentary published online in Nature on September 29, 2026, posits that artificial intelligence (AI) has the potential to significantly expand the frontiers of scientific discovery. However, this potential can only be fully realized if scientific institutions, including funders, journals, and universities, fundamentally alter their reward structures. The core argument is that current systems predominantly incentivize the efficient exploration and exploitation of already established scientific knowledge, often referred to as 'mining old ground.' This approach, while yielding measurable results, inadvertently discourages the risk-taking and exploration necessary for true scientific innovation and the creation of entirely new fields of inquiry.

The commentary highlights that AI tools are becoming increasingly adept at analyzing vast datasets, identifying patterns, and accelerating research within existing paradigms. This efficiency can be beneficial for tasks like optimizing experimental designs, synthesizing existing literature, and even generating hypotheses based on current understanding. However, the authors contend that an overemphasis on these measurable outputs, which are easily quantifiable and publishable, can lead to a stagnation of truly novel research. The pressure to publish frequently in high-impact journals, which often favor incremental advances over groundbreaking, yet unproven, ideas, exacerbates this issue.

To foster genuine scientific progress and leverage AI's full potential for discovery, institutions need to actively cultivate an environment that rewards the exploration of uncharted scientific territory. This involves recognizing and supporting research that may not have immediate, quantifiable outcomes but holds the promise of opening up entirely new avenues of investigation. Such a shift would require a re-evaluation of grant review processes, journal editorial policies, and university tenure and promotion criteria. Funders might consider dedicated programs for high-risk, high-reward exploratory research, while journals could create special sections for speculative yet rigorously argued proposals for new research directions.

The commentary suggests that by reorienting incentives, scientific communities can encourage researchers to utilize AI not just for optimizing known processes but for venturing into the unknown. This could lead to the identification of entirely new scientific disciplines, the development of groundbreaking technologies, and a deeper understanding of complex phenomena that are currently beyond our grasp. The authors emphasize that the true power of AI in science lies not only in its ability to process information but in its capacity to help us ask entirely new questions and explore the fundamental nature of reality in ways previously unimaginable.