2026 Fall Colloquium

Event details Date: Tuesday, September 29, 2026 Time: 12:30 PM to 2:00 PM Location: AHF B57J Allan Hancock Foundation Campus: University Park Campus Department: Spatial Sciences Institute About this event When geographers share responsibility for analytic reasoning with artificial intelligence, it often becomes unclear where the warrant for their claims resides. A claim has warrant when it is produced by reliable methods, rests on tested assumptions, survives defensible alternative analyses, and can be inspected by others. In traditional spatial analyses, these conditions are satisfied through a chain of analytic work performed by a researcher. That researcher translates a substantive question into a formal spatial problem, selects an analytical path from a garden of forking analytical alternatives, adapts that approach when the particulars of a place demand it, executes a workflow, and appraises the result. When some or all of these tasks are delegated to AI, any warrant for the resulting claims rests on a division of epistemic labor that spatial scientists have yet to extensively examine. That uncertainty has implications for how researchers conduct and evaluate spatial science, and how educators train the next generation. This work examines whether current approaches to evaluating the capacity of generative AI to autonomously design, execute, and interpret spatial analyses risk overstating the performance of these systems. I present a set of approaches to making this type of evaluation, and link this work to the challenge of pedagogy, curriculum design, and class planning in the spatial sciences in presence of generative AI. Key words: Geographic Analysis, GeoAI, Foundation Models, Agentic AI, Multi-Agent Systems, Reproducibility, GIScience Official event details: https://calendar.usc.edu/event/2026-fall-colloquium