Andrew Beam, PhD, Chief Technology Officer at Flagship Labs 97, Inc., will deliver the Fifth Annual Gilbert S. Omenn Lecture, presented by the Department of Biomedical Informatics at Harvard Medical School. A reception will follow.

Artificial Intelligence has catalyzed remarkable advances in biomedical science, with example breakthroughs in protein structure prediction and generative modeling of biological systems. Drawing from recent developments and his own research in protein diffusion models, Dr. Beam will examine the complex relationship between AI and scientific discovery. While the success of these specialized models is undeniable, they present an intriguing paradox: unlike large language models, biomedical AI systems have not demonstrated comparable benefits from increased scale. Instead, their effectiveness often stems from carefully crafted inductive biases that enable efficient learning from limited datasets, with attempts at scaling showing diminishing returns.

Building on these insights, the lecture will examine current claims about large language models' potential to revolutionize scientific discovery. While these models excel at synthesizing vast amounts of scientific literature, their ability to discover new scientific knowledge will be bottlenecked by access to an experimental platform that will let them design and test new hypotheses. Beam will argue that without the capacity to design and control real experiments, AI systems will remain stuck at the forefront of scientific knowledge, rather than advancing it. This limitation points to a necessary shift in how we develop AI for scientific applications, emphasizing the need for systems that can actively participate in the experimental process.

Registration is only required for those without a Harvard ID. If you are not a Harvard ID holder, please email [email protected] to add your name to the guest list for entry to Countway Library.