Models that act on the physical world are only as good as the real-world data they are trained on.
Physical AI is only interesting when you train it on physical-world data. All models need real data, and the physical world models need real-world data. So in biology, you need a wet lab with pipettes and a vivarium with mice, and you have to generate high-quality data, because otherwise it’s garbage in, garbage out.
Source: Tomorrow’s Medicine, with Cyriac Roeding (eCornell Keynote)