CSE DSI Machine Learning Seminar with Jia Liu (SEAS, Harvard)

Agentic and Physical AI for Scientific Discovery and Intelligent Manufacturing

Agentic and physical AI are opening a new frontier in scientific dicovery and intelligent manufacturing by enabling AI systems to move beyond passive data processing toward active reasoning, decision-making, experiment execution, and closed-loop optimization in real-world environments. In this talk, I will present our recent progress in developing agentic and physical AI for scientific research and intelligent manufacturing, with an emphasis on building systems that can analyze complex multimodal data, coordinate specialized tasks, interact with human operators, and bridge digital intelligence with physical experimentation as well as chip and bio-manufacturing. As a representative application domain, I will first discuss our work on flexible electronics for brain-computer interfaces and the large-scale datasets they enable across neural activity, spatial gene expression, and behavior. I will then examine the limitations of conventional deep learning frameworks in analyzing large-scale, heterogeneous scientific data. 

Next, I will discuss the opportunities created by generative AI, particularly its reasoning capabilities, as well as its current limitations in reliability, scientific grounding, and sustained multi-step problem solving. I will further explain why large language models alone are insufficient to address these challenges. To overcome these limitations, we are developing agentic AI systems that integrate reasoning, memory, tool use, and iterative planning. I will present three examples: a spike sorting AI agent for neural electrophysiology, a spatial transcriptomics AI agent for spatial gene expression analysis, and a behavior AI agent for animal behavior analysis. By integrating these agents across modalities, we aim to construct a large-scale functional brain atlas linking neural dynamics, molecular states, and behavior. Finally, to close the loop between AI-enabled data analysis and real-world experimentation and manufacturing, I will discuss our efforts on Agentic Lab and APEX, which combine AI agents, wearable mixed-reality systems, and human-in-the-loop frameworks to enable closed-loop scientific discovery and intelligent manufacturing. I will conclude with perspectives on the future of agentic and physical AI as a foundation for next-generation science and engineering.

Professor Liu received his Ph.D. in Chemistry from Harvard University in 2014 and completed postdoctoral training at Stanford University in 2018. He joined Harvard School of Engineering and Applied Sciences as an Assistant Professor in 2019. At Harvard, his lab develops tissue-integrated intelligent bioelectronic systems that merge flexible and soft electronics with living tissues, and combine multimodal in situ characterization with deep learning and agentic AI to decode and control biological processes. His research spans tissue-like bioelectronics and cyborg organoids to AI-driven multimodal analysis and tissue functional control for understanding neural dynamics, organ function, and disease mechanisms. Professor Liu has pioneered new paradigms in bioelectronics, establishing foundations for soft electronic materials, nanoarchitectures for tissue-like electronics, and AI-integrated bioelectronic systems. His work was recognized as a milestone in bioelectronics by Science (2013, 2017) and was selected among Top 10 World-Changing Ideas and Most Notable Chemistry Research (2015). He has received numerous honors, including MIT Technology Review’s “Innovators Under 35” (Global List), the AFOSR Young Investigator Program (YIP) Award, the NIH/NIDDK Catalyst Award (from the NIH Director’s Pioneer Award Program), the William F. Milton Award, and the Aramont Award for Emerging Science Research Fellowship. Professor Liu is also a cofounder and scientific advisor of several deep-tech startups translating his lab’s innovations into practice, including Axoft, Elastro, MorphMind (AIScientist), and NanoRythmics.