We think human nudity is obvious because we're biologically wired to recognize humans; and our culture expects clothing, making the lack of clothing all the more obvious to us!

But a computer has neither. It's only looking for commonalities between images

What I'm getting at is we're trusting these systems to find patterns that we think are super obvious, but they just can't because they DON'T THINK LIKE US. In fact, they don't think. At all.

we're training these systems to find patters that WE think are obvious

Like nudity. We throw a bunch of images at a bot and say "this is nudity". What the bot sees isn't naked humans, it sees an image. It has no cultural understanding of "nudity"

So it finds the most common element: skin color