At Duke University, artificial intelligence and advanced computing are no longer experimental add-ons. They are becoming the default tools for asking questions that were previously impossible to answer. The shift is changing how researchers work across biology, medicine, and the broader sciences. [1] [2]
Foundation Models Act as Microscopes for Biology
Rohit Singh, a computational biologist at Duke School of Medicine, treats foundation models like new kinds of microscopes. These machine learning systems are trained on hundreds of millions of data points, letting researchers see patterns across millions of gene expression profiles and biological sequences. Each model reveals a different aspect of life that can be studied.
Training these models requires massive computational effort, but access to graphics processing units (GPUs) has cut what once took weeks into a matter of days. The Duke Computer Cluster provides the high-performance computing backbone for these tasks, and a small GPU center planned to open in 2027 will expand capacity further.
The payoff is direct. By learning abstract representations of proteins, genes, and cells, Singh’s models can compare healthy cells to cancer cells and ask whether a drug can zero out the difference. Insights from this work feed into projects ranging from gene therapy development to drug discovery for multiple diseases.
Interpretable AI Builds Trust in High-Stakes Decisions
While faster hardware expands what is possible, Cynthia Rudin focuses on making AI systems people can understand and trust. For years, most AI models operated as black boxes, hiding the reasoning behind their predictions. That opacity is a problem in fields like healthcare and criminal justice, where decisions carry serious consequences.
Rudin’s lab designs interpretable models that let experts see how decisions are made. One such model, built with Duke radiologist Joseph Lo, analyzes subtle imaging patterns in mammograms to predict a patient’s risk of developing breast cancer over the next one to five years. Because the model is transparent, clinicians can troubleshoot and evaluate the data and reasoning behind each prediction.
These interpretable systems have already been applied across healthcare settings, from predicting seizures to analyzing medical images. The goal is not just accuracy, but accountability in domains where trust matters as much as performance.
AI Bridges Disciplines to Accelerate Discovery
Duke’s Deep Tech initiative has funded four projects that explore how AI can accelerate scientific discovery through multidisciplinary collaboration. Each project tackles a different barrier that slows progress across fields.
Consilience, led by Brinnae Bent, is a voice-based AI system designed to support cross-disciplinary collaboration. It translates terminology, prompts reflection, and surfaces novel research connections in real time. The system is being tested during a university hackathon, where it guides graduate student teams from diverse fields as they synthesize interdisciplinary research proposals.
David Carlson’s project focuses on training students to use AI critically. Rather than simply adopting large language models, researchers learn to evaluate these tools, identify flawed reasoning, and apply them responsibly. This matters most in interdisciplinary work, where bridging biology, statistics, and engineering demands complex reasoning.
Sharique Hasan’s team explores how AI can translate complex findings for policymakers, industry, and the public. By highlighting research with high impact potential and uncovering hidden links between fields, these tools can shorten the path from discovery to real-world benefit.
Matthew Hirschey’s project maps scientific progress and bottlenecks by analyzing the literature. Large language models trace how concepts move across fields and reveal shared challenges like limited data or computing power. The resulting maps aim to guide funding decisions and foster collaboration beyond traditional silos.
Infrastructure Keeps Pace with Ambition
None of this would be possible without sustained investment in computing infrastructure. The Duke Computer Cluster already supports large-scale scientific applications, and the upcoming GPU center is designed to minimize power and water consumption and carbon emissions through energy-efficient practices.
Singh notes that access to more GPUs directly accelerates his computational biology research. The investments Duke has made position the university as a leading site for this kind of work, attracting talent and enabling projects that depend on both scale and sustainability.
As AI becomes embedded in the research process, the challenge is not just building faster models, but building models that are useful, trustworthy, and aligned with the needs of the people who depend on them. Duke’s approach suggests that the future of scientific discovery will be collaborative, interpretable, and deeply computational.
