I’m interested in information and decisions. Bad decisions come from failures in how information is gathered or processed. When and why do capabilities fail to convert to tangible real-world outcomes? Why do institutions fail at things that ought to be easy?
TerraSol is a solar site assessment app I built to turn about a month of initial assessment into a five-minute process. It combines geospatial analysis, grid-capacity estimates, project economics, and AI-powered zoning research. A solar developer adopted it as their primary site assessment tool across the mid-Atlantic.
Give an AI the information it needs and it can still make a bad business decision. I built ConstructSim to study why. Six firms have to finish a construction project, but each has its own costs, private information, and reasons to disagree.
Across three sampling temperatures, both setups with a supplier summary met every firm’s goals without backup steel in every run (18/18). Without that summary, 2/18 runs did.
36 fresh runs in one fixed scenario: three per condition at temperatures 0, 0.2, and 0.4. Both successes without a supplier summary occurred at 0.4. These are exploratory counts, not reliable success-rate estimates.
I write Signal-Noise Ratio and contribute to Palladium Magazine.
I write when I have nagging thoughts or ideas. I try to publish only when I’ve gained a better understanding through the writing, or when the process has produced a useful insight.
The gap between collected data and tacit reality is always the most important datum.
AI can accomplish tasks. Do tasks even describe jobs? How is responsibility assigned?
Our habits and priorities change. What do we owe the people we might become, and the people around us?
My career started in Coast Guard search and rescue, where information directs action. That experience seeded an interest in how information gets gathered, communicated, and used. Those interests are the common thread through data science, TerraSol, and my current work on AI.
I’m now the Principal Data Scientist at SteerBridge.
My independent research is mostly focused on AI agents making decisions together. We should be concerned about their ability to make good decisions on complex cooperative tasks. Do they make optimal decisions for their private interests and for the interests of the larger project?
I write about some of these problems, but the essays range further. A measure can keep improving while the thing it was meant to describe gets worse. Technology can save us time without leaving us any more of it. Those are some of the themes I return to.
I’m also a dad. A fair amount of my writing comes from trying to understand what I ought to do with that responsibility.
Search and rescue
Summit LLC
Deloitte
FTI
Data science
Founder
Principal Data Scientist
If you’re working on something related, or read an essay and think I’ve missed something, I’d like to hear from you.
Email Matt