Hasn't that horse bolted?
As a researcher working on image processing (including remote sensing) I must really wonder what they think this will achieve. The tools I develop for remote sensing aren't fundamentally different from those I work on in astronomical or medical applications, or document processing. After all, AI methods are supposed to be generic. In CNNs the real slog is getting enough high-quality ground-truth data. Thus, an export ban on a trained neural network for an application might just work (not likely), but a ban on the generic code itself is hardly going to help, especially if you can buy it for e.g. document processing. And of course, there isn't an absolute shitload of code for these tasks available elsewhere, for free (<cough> GitHub <cough>).
Biting the hand that feeds IT
