From Nuclear Physics to Psychometrics: A Career in Measurement
My first doctorate was in experimental nuclear physics. My second, earned summa cum laude at Rutgers, was in statistical psychometric methods. The common thread is measurement: learning what you can honestly claim from data, and what you can't.
Physics trained me to take systematic error seriously and to treat a well-designed experiment as a form of argument. I published in Physical Review Letters, and then went looking for harder measurement problems: the human ones.
Psychometrics is where I found them. My dissertation applied Monte Carlo methods to the problem of measuring things you can never directly observe, like cognition, personality, and intelligence, and was published in Statistics in Medicine. Latent variable modeling, Item Response Theory, and Bayesian methods are the working tools of that trade.
Everything I teach about AI runs on that same discipline, because most confusion about AI is, at bottom, confusion about measurement: mistaking fluency for competence, benchmarks for understanding, confidence for accuracy.