AI Seminar 140: Beth Plale

Title: Artificial Intelligence Infrastructure: Evolution through an AI Institute

Abstract: AI is rapidly becoming a critical layer of national infrastructure, yet the mechanisms by which we make AI systems accountable remain immature and fragmented. As frontier AI models reshape research, education, science, and industry, new national initiatives such as the National AI Research Resource (NAIRR) and the Genesis Mission are expanding access to AI capabilities while raising urgent questions about accountability, transparency, and public trust.

Drawing on five years as a co-PI of the NSF Intelligent Cyberinfrastructure with Computational Learning in the Environment (ICICLE) AI Institute, I will discuss how the ICICLE Institute has evolved its thinking in its goal of democratizing AI across the edge, cloud, and high-performance computing continuum. ICICLE has explored not only the technical challenges of composing AI functionality across distributed environments, but also the governance challenges that emerge when models, data, and workflows move across organizational and computational boundaries.

An aspect of particular interest to me is AI accountability: accountability requires that AI systems carry a memory of their origins, assumptions, limitations, and prior uses. I will reflect on the what infrastructure is needed for trustworthy AI. The next generation of AI infrastructure must support not only scale and capability, but also transparency, accountability, and public trust.