Machines image, people see
This is a distinction without a difference. If the sensors and software are competent enough, it can surely accomplish the same predictive performance.
Yes and no. Yes, because there's no fundamental reason this shouldn't be true. No, because software isn't anywhere near that advanced yet.
Visual algorithms are getting pretty good at accomplishing specific tasks, such as navigation, 3D reconstruction, tracking, object identification, etc. The problem is doing all of those things in a unified way on the same scene, fast enough to be useful, *and* figuring out what to make of new things that are unlike anything that's been trained for recognition.