Who am I to say?

I used to get paid to make database vendors unhappy

In that same era, I learned agile development from Ward Cunningham, who invented it, which is a little like learning guitar from some guy at a party who turns out to be Django Reinhardt. Both experiences left me with the same twitch: when someone shows me a benchmark, my first question is who ran it and what they were hoping to find.

I question, therefore I am.

Before all that, math degree at Northwestern and a stretch as an actuary (never saw this), where I learned that “the model says so” and “so what” are fully compatible statements.

Today: American passport, Belgian mailbox. Four years in four Latin American countries and a Schengen visa clock that makes me the only person in the class whose study schedule is partly set by an immigration lawyer’s spreadsheet. I am doing research in emotion recognition and NLP in David Stuckler’s research training group. Many years of software engineering underneath it all, which mostly means I have watched this field rediscover the same three ideas four times each.

Next few years: Finish self-funded ALM in AI and grad cert in psychology. The plan after this is a master’s in global mental health at London School of Hygiene & Tropical Medicine and then a PhD at Cambridge, entirely on someone else’s dime, ideally with Fulbright’s face on it.

Academic clickbait: Emotion recognition is largely trained on English text written by people having a reasonably good decade, then sold to users who already have a therapist, a wearable, and a copay. I work the other end. The most populous nation on earth has 2.9 psychiatrists per million people, and nobody there needs an app that falsely claims to replace a clinician. Existing clinicians need to be faster. My second opinion, related: our leaderboards are Certainty Theater, single-point estimates with the random seed variance and the search budget quietly filed off under “future research”. I will die on this hill and I have the systematic lit review receipts to prove it.

Fun facts, all variations on a theme: I edited a comedy magazine, improvised jazz, and performed slam poetry. All three to find out in public, in real time, whether an idea works. For Stanford’s Creativity MOOC, I recruited a team that finished first out of 1,220 teams. Much more creative teammates than me included one who had been voted Britain’s cartoonist of the year. My job was Chief Cat Herder.

If you are working on anything multilingual, anything mental health adjacent, or anything where you suspect the evaluation metric is lying to you politely, reach out: Jeff_Winchell@g.Harvard.edu.

Tell me the most confidently reported number you have ever personally watched get made up.