The Importance of Robustness in AI

Is it accurate to say kind of your learning or your algorithm, no matter how much it develops, is really only as good as what you put in? I mean, if you're looking at identifying pneumonia in a chest x-ray and you know, you feed it a bunch of stuff, but you never put in a patient that's had a new man activate. And then they see a patient with only one lung. And the thing has no idea what to do with that. So there are lots of humorous examples that have been published like models being 95% certain that a dog is actually a parrot or something like that because this AI algorithm is trained to recognize different bird species, but