About
I am a researcher focused on technical AI governance, with a particular interest in applying systems-thinking based safety methodolgies to the challenge of developing and deploying advanced AI systems responsibly.
My current work centers on STAMP (Systems-Theoretic Accident Model and Processes) and how this accident causality model can surface categories of risk that conventional failure-based models tends to miss. Most AI safety risk analysis focuses on what happens when a model produces an error or acts contrary to its intended objectives. However, unsafe behavior can emerge from the interactions between components in an advanced AI system even when no individual component has failed. Understanding and governing those systemic risks is what I am currently focused on.
I am attempting to write more regularly about systems thinking, hazard analysis, and risk characterization as they apply to advanced AI, particularly agentic systems. You can find my writing on this site and on my Substack.
Background
Before moving into technical AI governance, I was an astrophysicist studying galaxy evolution. I earned my PhD in Physics at UC Irvine, where I studied environmental quenching mechanisms in dwarf satellite galaxies. I then did postdoctoral research at the University of Washington measuring the cold gas content in nearby galaxies with radio telescopes.
This background gave me experience in quantitative modeling, working with complex interacting systems, and constructing explanations for emergent behavior.
Get in Touch
spfillingham [at] gmail [dot] com