THRUST 4

The overall goal of Thrust 4 is to pressure test the proposed paradigm of critical-thinking AI guided closed-loop experimentation using materials discovery and understanding as a testbed to drive further development of foundational AI tools outlined in Thrust 1.

In Phase I, we established a Closed-Loop Transfer (CLT) paradigm that transfers from the closed-loop discovery regime into the hypothesis-driven discovery regime to yield new knowledge of how molecular structure encodes photostability in light harvesting small molecules – a crucial bottleneck impeding solar cell technologies. Key to the success we achieved in our first five years is a modular chemical synthesis platform that is friendly to both automation and AI. By putting synthesis considerations at the beginning, rather than at the end, of the AI-guided discovery process we have eliminated the longstanding synthesis bottleneck that previously precluded the robust integration of AI with C-C bond-based materials discovery. Although some recent work has demonstrated closed loop paradigms in molecular discovery, including work from some of our labs partially funded by NSF-MMLI, none, prior to our recently published work, has discovered new chemical knowledge. Central to the success of CLT in making this leap to new knowledge discovery is the fusion of physical modeling with AI.

In Phase II, CLT will take a radically new dimension by pairing up with a multimodal AI agent (Thrust 1) that can autonomously propose hypotheses, request data from the closed-loop experimentation to test hypotheses and learn and ultimately deliver new knowledge non-existent in literature (CLT 2.0). In this way, our work will advance multimodal language models from the “undergraduate” to “graduate” level by infusing critical thinking and learning through closed-loop experimentation.