On May 19, 2023, Anima Anandkumar, Bren Professor of Computing and Mathematical Sciences, joined Demis Hassabis and Fei-Fei Li (PhD EE ’05) for a public discussion of Artificial Intelligence (AI) Enabling Science sponsored by the President’s Council of Advisors on Science and Technology (PCAST).
PCAST consists of individuals from outside of the federal government who advise the president on policy matters related to science, innovation, and technology, and currently includes 28 experts from a wide array of scientific fields.
Anandkumar described the current realm of generative AI, which can generate new pieces of text from scratch, new images given text prompts, and even new molecules given certain property specifications. “What’s different from the previous era of discriminative AI that was popular in the last decade is that back then, we could only predict a property given an image or given a molecule,” said Anandkumar. “Generative AI is the inverse process, which is much harder, and we are able to do that today.”
One example of generative AI in action is the prediction of SARS-CoV-2 variants such as Delta and Omicron. “The generative AI model generated unknown variants that haven’t yet emerged, and this is how we can be better prepared with better drugs and vaccines,” said Anandkumar.
In addition to using generative AI models in the domain of public health, Anandkumar described how AI can be used for extreme weather forecasting and climate change mitigation, which can lead to better risk assessment.
“In all these scenarios, what you saw is the enormous ability for AI to do fast scientific modeling, and as part of the process, also be able to do inverse problems and incorporate a larger set of hypotheses,” said Anandkumar.
Anandkumar concluded her presentation by making the case for a “foundation AI” model for science, a term that entails an AI neural network with broad capabilities that can be adapted to various specific scientific purposes. “What we’ve seen so far is AI already making a huge impact in specific domains on specific modeling tasks. By combining data and domain knowledge across multiple domains, I think we will be able to create a foundation model for science that will have a huge impact,” said Anandkumar.
The discussion on AI Enabling Science was moderated by PCAST co-chair Frances Arnold, Linus Pauling Professor of Chemical Engineering, Bioengineering and Biochemistry; Director, Donna and Benjamin M. Rosen Bioengineering Center. Arati Prabhakar (MS ‘80, PhD ‘85), the Director of the White House Office of Science and Technology Policy (OSTP) and Assistant to the President for Science and Technology, also serves as PCAST co-chair.
Current members of PCAST include John Dabiri (MS ’03, PhD ’05), Centennial Professor of Aeronautics and Mechanical Engineering, and Bill Dally (PhD ’86), Chief Scientist and Senior Vice President for Research at NVIDIA.
View Anandkumar’s talk and the full panel discussion via the White House’s official YouTube channel.