How is agentic AI transforming university research?

What is agentic AI, and how can it accelerate scientific discovery?

What if AI could help scientists discover something they weren't even looking for?

This is one of the possibilities of agentic AI. AI agents can work through a series of steps, analyze information and identify findings worth investigating further, rather than just answering questions or completing individual tasks,

During the webinar, Varshney shared an example from Stony Brook University, where researchers are developing a “self-curious microscope”. Instead of only examining what scientists tell it to look for, the microscope is being designed to spot unusual or interesting observations on its own and bring them to researchers' attention. This matters because researchers cannot examine every piece of information or anticipate every possible discovery, and AI could help them explore things they might otherwise miss.

Elliott also explained how AI agents are being used to support everyday research tasks, from analyzing data to developing experimental procedures.

This opportunity gives researchers more time to investigate new ideas and potentially uncover findings that wouldn't have been possible through traditional methods alone.

How can AI support interdisciplinary research?

Another opportunity lies in making collaboration between academic disciplines easier. Traditionally, university research has often been organized around specialized departments and areas of expertise. While this supports academic depth, it can make interdisciplinary collaboration difficult.

AI tools can help researchers understand concepts, methods and findings outside their own specialisms. Elliott suggested that this could encourage new collaborations between researchers in fields such as physics, chemistry and mathematics.

At Stony Brook University, the AI Innovation Institute takes an approach that combines developing new AI technologies with applying them across different academic disciplines. Varshney argued that the most valuable discoveries may emerge where these activities intersect, with real-world research challenges informing the development of new AI capabilities.

However, universities may also need to reconsider how they support interdisciplinary work, including academic incentives, promotion and research evaluation.

What are the risks of using agentic AI in research?

Despite its potential, agentic AI does introduce important questions around accuracy, security, ethics and academic integrity. One challenge is ensuring that AI-generated findings can be verified, particularly in scientific fields where inaccurate conclusions could affect subsequent research.

The panelists discussed the importance of safeguards, including systems that restrict what AI agents can do and methods for checking their outputs. Varshney also emphasized the importance of maintaining human expertise and critical thinking.

As AI becomes increasingly embedded in research and teaching, universities must ensure students and researchers understand the information they are working with rather than relying entirely on automated outputs.

The discussion also explored global representation in AI research. AI models trained predominantly on research from certain regions may overlook questions, languages and perspectives from elsewhere.

Varshney highlighted federated learning as one possible approach, allowing institutions to collaborate and benefit from shared learning while keeping sensitive data within their own systems.

How should universities prepare for AI-driven research?

As AI becomes a bigger part of university research, institutions face a practical question: how can they make the most of these tools while managing the costs and risks?

One challenge is funding. Traditionally, research grants have covered costs such as staff, equipment and laboratory materials, but as researchers increasingly rely on AI, universities may also need to budget for access to AI models and the computing power needed to run them.

Varshney suggested that AI could eventually account for a substantial share of some research budgets, although how institutions should fund this remains an open question.

Choosing the right technology is important – Elliott encouraged universities to think about what they want AI to achieve, rather than automatically investing in the most advanced tools. A smaller, less expensive AI model may be enough for some tasks, while more complex research could require greater resources.

"Staying close to your mission is probably the most important piece of advice I could give."

Charles Elliot, Field CTO at Google Public Sector

However, Varshney encouraged universities to look beyond using AI to save time. Its greatest potential may lie in helping researchers tackle more ambitious questions, explore new areas of research and make discoveries that were previously out of reach.

"I think a lot of people are thinking about 10% efficiency improvements, but what I would encourage everyone to think about is unlocking 10x new capabilities."

Lav Varshney, Infinity Professor and AI Institute Director at Stony Brook University

What's next for AI in university research?

Agentic AI could change not only how quickly research is conducted, but also who can participate, how disciplines collaborate and which questions universities are able to explore.

However, realizing these opportunities will depend on institutions combining technological investment with research expertise, responsible governance and a willingness to rethink established practices.

Watch the full QS and Google webinar to hear more examples and insights on preparing universities for AI-driven scientific discovery.