Summary
Today’s AI systems cannot generate novel hypotheses to support scientific discovery for a variety of reasons. One primary factor is that the background knowledge and assumptions of a given scientific domain, called inductive biases, must be hand-crafted and incorporated into AI systems (e.g., the electrochemical properties of valid proteins). This bespoke process is time-consuming and limits the creativity of the system.
How we develop machine learning models to be creative is an open research question explored through the Foundation Models for Scientific Discovery (FoundSci) Artificial Intelligence Exploration (AIE) opportunity.
Truly intelligent AI for scientific discovery must be capable of automatically identifying relevant inductive biases or discovering these biases from a theory, system, or method based on that concept. FoundSci aims to develop and demonstrate an AI agent as an “autonomous scientist” capable of skeptical learning and reasoning to aid human scientists. The resulting AI will use scientific reasoning to generate creative hypotheses and experiments and refine those hypotheses to enable scientific discovery at speed and scale.
This effort envisions an autonomous scientist possessing the ability to characterize its uncertainty and skepticism and use them as drivers to systematically acquire and refine its scientific knowledge bases in a way that human scientist partners can trust. The resulting capability would augment Defense Department research and accelerate discovery in various disciplines, such as the discovery of new materials or new computing architectures.
Workshops informed the FoundSci AIE as part of DARPA’s AI Forward initiative.