|
AI policy and governance advocate dedicated to a safe post-AGI future.
I'm a MATS research fellow (9.0, extending into the 9.1 independent research program) studying public attitudes toward AI risk. My current project — The 5 Americas of AI Concern — segments 2,735 U.S. adults into distinct concern profiles using latent class analysis, then tests how each group responds to different informational frames about AI governance.
Before the research pivot, I spent a decade building startups and doing government affairs. I have degrees in political science and law, neither of which I use in the way my parents expected. I've completed AI governance courses through BlueDot Impact.
I've done art direction for The Intelligence Curse and contributed AI-generated video to the Species YouTube channel. I write a Substack that was featured in Jack Clark's Import AI — and wrote a post every day in November for Inkhaven.
A nationally representative study (n=2,735) segmenting Americans into five distinct AI concern profiles via latent class analysis. Interactive scrollytelling visualization testing how nine different policy framings resonate with each segment.
fiveamericas.ai →
An AI agent benchmark where LLMs play Blood on the Clocktower — a social deduction game that tests deception, persuasion, and theory of mind. Supports both closed and open-weight models with persistent memory, live spectator mode, and a pixel-art frontend you can watch in real time. Also a tool for AI safety monitors — can you spot which agent is scheming before the town does?
bloodbench.com →
Art direction for a series of AI policy essays by Luke Drago and Rudolf Laine exploring "the intelligence curse" — what happens when powerful actors can derive wealth from AI instead of human labor. Visual identity, AI-assisted imagery, and launch collateral.
intelligence-curse.ai →
AI-generated video content for an educational AI communication channel making complex concepts accessible through visual storytelling.
youtube.com →
When people ask how I started working in AI safety, I tell them it was by posting synthetic photos of telephone poles on X. This is not the response they were expecting.
I spent years deep in "post-photography" — using generative AI to create images that feel photographic but depict scenes that never happened. Liminal spaces, suburban architecture, lonely sentinels of communication. It turns out that in a field where everyone optimizes for the same signals, being the person who makes haunting post-photographic telephone poles is surprisingly effective differentiation.
"The thing that makes you slightly odd might be exactly what makes you valuable."
Read the full essay →For research collaborations, speaking, policy roles, or to argue about AI governance over coffee.
justrmil@gmail.com Or grab the resume →