The AI research revolution: Why making progress could mean breaking the system

For most of modern history, the pace of discovery has been constrained by how quickly humans can digest, analyze and build upon what is already known. Artificial intelligence (AI) is changing that.

Whether in academia, government or industry, researchers spend significant time reading, experimenting, analyzing data, publishing their findings, and applying for funding support. Once the results are disseminated, other researchers must digest them before building upon them, or identify new opportunities inspired by the findings. AI has the potential to accelerate every step of this process.

Generative AI systems such as Claude , Chat GPT and Gemini can rapidly match patterns, identify connections and even generate ideas. They can give the illusion of human intelligence , though that impression is misleading. AI does not replace human scientific judgment. What it does possess is the ability to process information at a scale and speed that humans cannot match.

Knowledge discovery is constrained not only by human attention, but also by the money needed to investigate promising ideas. Research funding is secured through competitive grants from government agencies like the National Institutes of Health and the National Science Foundation , yielding research results and products that have transformed society . AI’s most valuable role here may be as a digital adviser and collaborator.

A researcher can ask AI to summarize literature, suggest ways to design an experiment, challenge core assumptions, and even propose extensions to research ideas. AI reduces the cost of exploring an idea before deciding whether it deserves serious investment. Indeed, the benefit to society may come not from AI producing a particular answer, but from enabling researchers to efficiently ask and investigate many more questions.

There is an important distinction between generating possibilities and deciding which possibilities are worth pursuing.

For now, humans remain responsible for determining which questions to tackle and whether an AI-generated response is even correct. AI can hallucinate by fabricating data, giving false citations and providing incorrect explanations that can all appear convincing because they are accompanied by seemingly credible sources. A researcher who lacks a strong understanding of the underlying science may not recognize such errors.

This is why AI in its current form will not eliminate the need for researchers. Instead, it will change how researchers spend their time and how they train the next generation to use it. For example, chess grandmasters now train against and learn from AI chess systems . This does not eliminate these highly skilled players; it provides a mechanism to improve their skills. AI is poised to do the same for researchers and knowledge discovery, but with the potential to have greater impact.

The consequences will extend far beyond universities and research labs.

If AI can increase research capacity, it may help speed up the discovery of new medicines, materials and engineering solutions. It could help researchers analyze massive datasets that would otherwise take years to examine. For example, AI is already accelerating pancreatic disease research , uncovering methods to detect cancer for as much as three years prior to diagnosis. AI was also used to create a counterexample for an unsolved mathematical conjecture proposed by mathematician Paul Erdos in 1946. Most recently, AI proposed a solution to the Navier-Stokes existence and smoothness problem .

But this productivity creates a problem of its own.

When AI makes it significantly easier to produce research papers and grant proposals and to perform technical analyses, sheer output becomes less meaningful. If everyone becomes hyper-productive, then no one is.

This will force universities to recalibrate how they evaluate researchers . When AI dramatically increases the amount of research that can be produced, institutions can no longer rely on quantity as a proxy for quality. The emphasis will need to shift toward the originality of ideas, the importance of discoveries and the impact of the resulting work.

Federal research agencies will face a similar challenge. If AI makes it possible to quickly produce polished research proposals, government agencies charged to invest taxpayer dollars could face a flood of applications that cannot be considered by simply adding more reviewers. Funding processes will need to place greater emphasis on novel, high-risk ideas rather than incremental research that AI is well suited to formulate.

Scientific journals face their own set of challenges. An explosion in AI-assisted manuscripts could overwhelm traditional peer review. AI must become part of the solution by helping editors screen submissions. But human judgment will remain essential for detecting subtle errors and determining whether research represents a meaningful contribution.

Perhaps the most consequential change will come from AI’s ability to amplify knowledge. First, AI helps researchers identify new knowledge. That knowledge then becomes part of the expanding scientific literature ecosystem. Future AI can then use that knowledge to assist researchers in making additional discoveries.

This transformation will not happen overnight. It will also not eliminate the need for human researchers. It will, however, change the relationship between humans and knowledge discovery. As AI becomes better at processing what is already known, researchers may be able to spend more time exploring what is not known. The scarce resource may increasingly become not just information, but knowing which questions are worth asking.

Sheldon H. Jacobson, Ph.D., is a professor of Computer Science at the University of Illinois Urbana-Champaign. Daniel Solow, Ph.D., is a professor of Operations at Case Western Reserve University in Cleveland, Ohio.