The main point of Sabine's video is the complexity and inability of current science to predict how a protein will fold based on its molecular structure, which AI has now got closer to.
It goes to show how complex biology and in turn medicine is, which speaks to how far we are from being able abandon randomised controlled trials and replace them by figuring out new treatments in silico.
Still AI needs training data curated and verified by humans before being able to extrapolate new findings. AI still fails badly outside the training data, but exciting where we might get to in the next few decades and beyond.