Artificial intelligence is rapidly changing how scientific research is conducted—and according to researchers at the University of Cambridge, the shift is just beginning. In a new article, they highlight how AI is set to take over routine tasks and unlock entirely new forms of cross-disciplinary discovery.
Image Credit: Anggalih Prasetya/Shutterstock.com
At the forefront of this is the Infosys-Cambridge AI Centre, where teams are building intelligent systems that can enhance simulations, sift through complex datasets, and even conduct research autonomously. With AI managing the execution, researchers are freed to focus on creativity and conceptual thinking, setting the stage for faster, more connected scientific breakthroughs.
The Computational Shift Driving AI in Science
What’s fueling this change isn’t a sudden leap in theory—it’s the explosive growth in computational power. Cambridge researchers argue that this is what’s enabling AI to play a much larger role in scientific discovery. The Infosys-Cambridge AI Centre, launched in 2024, serves as a research hub focused on three key areas: AI-enhanced simulations, the mathematical underpinnings of neural networks, and autonomous research systems.
Dr. James Fergusson explains that current AI mostly mirrors human work, but faster. The next frontier, he says, will be AI that can also explain why it does what it does. Meanwhile, Dr. Boris Bolliet is working on multi-agent systems that act like digital research teams.
These AIs can plan experiments, cross-check results, and execute complex projects, essentially managing the research process end to end. This evolving model allows researchers to shift from doing the technical heavy lifting to focusing on high-level problem-solving, with implications across fields as diverse as cosmology and medicine.
Redesigning the Research Workflow
At the heart of Cambridge’s approach is a fundamental rethink of how research gets done. Instead of relying on single AI models, the Centre is developing multi-agent systems like CMBAgent and DENARIO. These platforms split big scientific questions into manageable tasks, assigning different agents to generate ideas, verify outcomes, and iterate toward solutions. The result is a system that mimics the dynamic of effective research teams; one agent proposes, another challenges, and a third verifies.
These systems are already being used in cosmology, where they analyze vast datasets and simulate the behavior of galaxies, all under human supervision. What makes them especially powerful is their potential to span disciplines. Unlike most researchers, who are trained within specific fields, AI agents can fluidly connect insights from astronomy, biology, and beyond, breaking down long-standing academic silos.
And transparency is built in. Every computational step is logged and traceable, helping to address one of the biggest concerns with AI in science: the “black box” effect. Researchers can see how conclusions were reached, which is critical for reproducibility and trust.
The Polymathic AI project takes things a step further by teaching AI systems to recognize patterns across multiple areas of physics. The idea isn’t to replace scientists, but to support them by taking over repetitive, data-heavy tasks. This allows human researchers to focus on conceptual breakthroughs while AI handles the grunt work.
Early results suggest this could radically speed up the pace of discovery while improving the accuracy and reproducibility of research. And because the systems are designed to self-check using multiple agents, the risk of error or misinformation—like hallucinated facts—is significantly reduced.
New Roles for Researchers, New Demands on AI
As AI takes on more responsibility, the role of the human scientist is evolving. Tasks like coding, data processing, and literature reviews are becoming automated. This frees researchers to do what machines can’t: ask the right questions, make creative leaps, and interpret results with nuance.
Dr. Fergusson likens this shift to Edison’s old adage about invention: AI delivers the 99 % perspiration, while humans bring the 1 % inspiration. But this shift isn’t confined to the lab. Businesses are already demanding AI that can explain itself, especially in decision-making contexts, and want to automate knowledge work more broadly.
The Infosys-Cambridge collaboration is helping bridge that gap, turning academic advancements into real-world tools. Yet, challenges remain. AI systems consume enormous amounts of energy, and even the best models can occasionally produce errors or unreliable outputs. That’s why Cambridge researchers are focused not on general AI, but on building reliable, transparent tools that work with human intelligence, not around it.
Redefining What it Means to Do Science
This shift has big implications for education and future scientists as well. Students with bold ideas but limited technical skills can now use AI to bring their concepts to life. It’s no longer just about how well you can code—it’s about how creatively you can think.
Ultimately, Cambridge researchers aren’t trying to build machines that replace scientists. They’re building tools that amplify human capabilities, making research faster, more interdisciplinary, and more accessible. And while challenges like energy use and reliability need to be addressed, the benefits are already starting to show.
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.