We just have a different definition of understanding. I say something has understanding if it exhibits behavior we typically associate with understanding. If you don't like that definition that's fine, but it doesn't disprove the fact that today's AIs can exhibit behavior which we would associate with understanding if exhibited by a human or animal. And that's a marked improvement on AIs from a decade ago. So if AIs continue to progress they will be able to conduct scientific experiments and eventually be able to wipe out humanity.