The Importance of Collaboration in Medical Machine Learning Applications

I'm struck by how nonlinear progress can be in the medical AI community. I think this is very common in the computer science and machine learning community. Another ingredient here, I think, is also very strong collaborations that you've built with clinicians and with radiologists specifically for checks. We have no public way of having the community transparently measure whether or not there are pitfalls in the way that we're saying models will generalize from site A to site B. But now let's approach it in a principled way. To do that, we need data. And we need data of the right kind.