The six-year industry-academia-federal project will combine several emerging technologies to develop the tools and playbook needed to master the next evolution of electronic warfare.
As AI data centers become larger and more powerful, they consume vast amounts of electricity and water, raising critical questions: Can AI become more sustainable?...
Every day, AI tools get more capable, and deployment is outpacing our understanding of what they get wrong. Duke Engineering Dean Jerry Lynch connects with...
Duke Engineering researchers are working with local Durham households to study the compounding effects of heat and air pollution to help people worldwide.
Featuring David Smith
Featuring Amanda Randles
Featured conference speakers include General Charles Q. Brown (Ret), NSA Deputy Director Tim Kosiba, Congresswoman Deborah Ross and Congresswoman Valerie Foushee.
Sep 18 8:00 am-8:00 pm
Sep 4
Chill out with us for the first Power Trip of the 2026-2027 school year! Join us for a tour of Duke University’s chilled water plant #2, which provides cooling services […]
1:30 pm
Sep 7
Sep 8
Learning in Networks Reading Group and Seminar Series: Tuesdays, 12pm in 330 Gross Hall; lunch included The institute focuses on fundamental questions in statistical inference, machine learning, causal inference, and […]
12:00 pm Gross Hall, Ahmadieh Family Grand Hall, Room 330
![Every day, AI tools get more capable, and deployment is outpacing our understanding of what they get wrong. In Episode 1 of Pratt Blueprint, Duke Engineering Dean Jerry Lynch connects with three Duke University experts to break down where that gap is already showing up: in cybersecurity, in the everyday AI tools people use to get advice, and in health care.Charley Kneifel, AVP and Chief Technology Officer for Duke's Office of Information Technology, explains how generative AI has compressed the time between a discovered vulnerability and a working exploit from months to under a day, and breaks down how deepfaked identity documents have become sophisticated enough to fool routine identity verification processes across higher ed and beyond.Brinnae Bent, Executive in Residence at Duke's Pratt School of Engineering and Director of the Duke TRUST Lab, explains why the same human preference data that makes chatbots feel helpful also makes them sycophantic, and points to a widely reported case of a pregnant woman wrongfully arrested after a facial recognition misidentification, to show what happens when a black box system fails and no one can explain why.Jessilyn Dunn, Theodore Kennedy Associate Professor of Biomedical Engineering at Duke, walks through how bias baked into training data gets perpetuated in clinical AI tools, and why clinicians are more likely to double-check an alarming AI output than a reassuring one, even when the reassuring one is wrong. The overarching narrative is AI earns trust by appearing competent, and that appearance of competence is exactly when human verification tends to collapse. Dean Lynch closes the episode on why judgment, not just technical skill, has to be part of how the next generation of engineers is trained.Chapters:0:00 AI Risks in 20261:01 The judgement gap2:05 Cybersecurity4:05 Why AI is trained to please you7:02 Using AI in healthcare9:23 The takeawayLearn more about our guests: Charley Kneifel: [https://oit.duke.edu/profile/charley-kneifel/]Brinnae Bent: [https://scholars.duke.edu/person/brinnae.bent]Jessilyn Dunn: [https://scholars.duke.edu/person/Jessilyn.Dunn]Duke University | Pratt School of Engineering | https://pratt.duke.eduSubscribe for future episodes of Pratt Blueprint.](https://pratt.duke.edu/wp-content/plugins/youtube-feed-pro/img/placeholder.png)