Twenty years ago I asked a friend, who happened to be a professor at Harvard Medical School, for a recommendation for a general practitioner doctor. His immediate response was, “Well you need a great diagnostician!” That’s because all healthcare starts with figuring out what’s going on, increasingly accomplished by looking at patterns in a bunch of data, images, old doctor notes. Some doctors are terrific at it. 50% are below average.
Erik Duhaime knew this too, and he and his co-founders — Zach Rausnitz and Tom Gellatly — went a few steps further. Like everyone awake today, they knew that machine learning algorithms can be superb at looking at patterns. And computers don’t get fatigued by a long day. Or distracted by a fight with their spouse. Or bored after seeing many many negative results. And unlike top doctors, top algorithms can serve everyone. Erik, Zach, and Tom believe — and I agree — that over time ML will transform healthcare, a gigantic market that desperately needs improving.
Of course these algorithms need to be trained, on a lot of data, and this data needs to be properly labeled. By people who know what they’re doing and at a reasonable cost. It turns out Erik’s Ph.D. research was focused on combining many people’s opinions of complex issues to get to the right answer — a perfect match. So Centaur Labs was born, to help the multitude of ML projects targeted at healthcare, in big and small companies, by properly, accurately, and cost-effectively labeling the data their algorithms desperately need in order to learn.
The team has accomplished a lot and still has a lot left to do. Medical labeling is complicated, far more difficult than identifying stop signs. First, the expertise is specialized, and the right people need to label the right data. Centaur needs to find them and match them correctly. Second, the task can be repetitive, and the labelers — who often work really hard at their day jobs — can get fatigued. Centaur’s software has to be able to tell, and assign a lesser, or zero, weight to the tired medical student who’s lost focus or the doctor who opened up his Chateau Lafite Rothschild a little early. Add to that the fact that health data typically carries security, access control, and provenance requirements that stop signs just don’t, and the fact that the data is sometimes an image, sometimes text, sometimes an audio file… A lot to do.
Still, when the team finishes their jobs and succeeds, not only will they build a massive business but they will improve healthcare for patients around the world. That’s a mission we at Matrix find it very easy to fully get behind.