Who they are
A digital twin is not a robot. It is a forecast with a conscience.
Unlearn.AI is a San Francisco company with a deliberately contrarian name. Its product is the digital twin - and the first thing to know is what it is not. It is not a 3D avatar. It is not a chatbot wearing a lab coat. It is a statistical prediction: for every participant who enrolls in a trial, Unlearn generates a comprehensive, longitudinal forecast of how that exact person would most likely have progressed had they been handed the placebo.
Feed that forecast into the analysis and something quietly radical happens. You no longer need as many real human beings sitting in the control group, because you already have a credible, patient-by-patient model of the counterfactual. Unlearn folds this into two unglamorous-sounding inventions: PROCOVA, the statistical method, and TwinRCT, the trial design. Together they let sponsors cut control arms by up to a third without breaking randomization and without smuggling in bias.
The math is old-fashioned in the best way - it is prognostic covariate adjustment, a technique statisticians already trust - supercharged by machine-learning models trained on more than 300,000 patient histories and a million-plus interactions. The result is thirteen disease-specific Digital Twin Generators, each one a specialist that has, in effect, read every trial in its field and remembered how the placebo patients fared.
It is a tidy bit of judo. Rather than ask regulators to trust a black box, Unlearn handed them an old, respectable statistic wearing new, faster shoes.
"Replace siloed workflows and black-box predictions with a unified, upstream workspace for trial design, planning, and analysis."
— Unlearn.AI, on what it actually sells
