Deep Tech at Duke Funds Four AI for Metascience Projects Through OpenAI Partnership
Deep Tech @ Duke University is excited to announce the four research projects that we are funding through the Artificial Intelligence (AI) for Metascience research program. Duke has partnered with OpenAI to establish this program, which explores how AI can accelerate scientific discovery through multidisciplinary collaborations.
Deep Tech released a Request for Proposals (RFP) on March 27th and received a large number of strong proposals, all reflecting a commitment to advancing metascience. After careful review within our Deep Tech Team —David Hoffman, Buz Waitzkin, and Merritt Cahoon —along with our faculty advisory committee, which includes Nita Farahany, Ashish Anora, Evan Levine, and Brandon Garrett, we selected four projects to fund. Due to limited resources, many excellent proposals could not be supported. We are grateful to all who submitted and look forward to future opportunities to collaborate.
We also extend our sincere thanks to OpenAI for funding this research program and supporting the advancement of interdisciplinary scientific discovery.
The selected projects stood out for their potential to meaningfully advance metascience through innovative uses of AI, strong interdisciplinary collaboration, and clear plans for real-world testing and public benefit. The four projects represent a broad range of Duke departments and schools working on AI—including Pratt, Fuqua, Nicholas, Sanford, Medicine, Computer Science, Biostatistics and Bioinformatics, and Philosophy. We look forward to supporting these research teams as they advance their work.
Consilience: AI-Augmented Interdisciplinary Research
Principal Investigator: Brinnae Bent
Team Members: Chris Bail, Boyuan Chen, Walter Sinnott-Armstrong, David Johnston, Lee Tiedrich
Scientific progress often depends on insights that bridge multiple fields, yet collaboration across disciplines is frequently hindered by jargon, differing methods, and siloed knowledge. This project introduces Consilience, a voice-based AI system that actively supports cross-disciplinary collaboration by translating terminology, prompting reflection, and surfacing novel research connections in real time. Consilience will be evaluated during a university hackathon, where the system will engage with teams of graduate students from diverse fields, guiding their conversations and helping them synthesize interdisciplinary research proposals. Through a randomized controlled trial, we will evaluate the extent to which Consilience can facilitate deeper understanding, more effective communication, and the generation of innovative ideas across disciplines.
Metascience Discovery Accelerator for Interdisciplinary Research
Principal Investigator: David Carlson
Team Members: Karthikeyan K and Beth Hauser
AI tools like large language models (LLMs) are rapidly becoming part of scientific research, offering new ways to generate ideas, synthesize knowledge, and accelerate discovery. However, without the ability to critically assess these tools, researchers, especially early-career scientists, may rely on AI without fully understanding its limitations. This issue is particularly important in interdisciplinary research, where bridging fields like biology, statistics, and engineering requires complex reasoning and communication. This project explores how students can be trained not only to use AI, but also to evaluate it carefully, identify flawed reasoning, and apply it responsibly to support meaningful scientific progress.
AI to Manage Scientific Discovery and Translation
Principal Investigator: Sharique Hasan
Team Members: Wesley Cohen, Steven McClelland, Roger Masclans
Scientific research has the power to drive major social and economic benefits, but many promising discoveries are overlooked or fail to reach the real world. With thousands of new papers published daily, it's difficult to identify which ideas have the greatest potential and where valuable connections across disciplines could spark innovation. This project explores how AI like large language models can help bridge that gap. By analyzing vast amounts of scientific data, these tools can highlight research with high impact potential, uncover hidden links between fields, and translate complex findings into clear insights for policymakers, industry, and the public. In doing so, AI can accelerate the path from discovery to real-world impact, making science more connected, inclusive, and useful.
AI-Driven Mapping of Scientific Progress and Bottlenecks
Principal Investigator: Matthew Hirschey
Team Members: John Bradley, Pol Castellano-Escuder, Darren Stuart
Scientific breakthroughs often happen where disciplines meet, but the sheer volume of research makes it hard for scientists to spot connections, shared challenges, or potential collaborators beyond their field. This project explores how large language models can address this problem by analyzing vast scientific literature to identify emerging ideas, trace how concepts move across fields, and reveal common bottlenecks like limited data or computing power. By creating dynamic maps of scientific progress and obstacles, we aim to show how AI can foster interdisciplinary collaboration, guide funding decisions, and accelerate discovery across the research ecosystem.
Additional Information
This initiative aims to identify several leverage points for accelerating U.S. and broader scientific advancement. Successful proposals will contribute to long-term projects for Duke and possibly inbound investment for North Carolina, with some projects potentially pursued as public benefit applications.
All research outputs, publications, and findings associated with this initiative will be made publicly available in digital format and by publication where appropriate, aligned with the goal of ensuring that AI advances benefit everyone.