"AI algorithms assign a 22% "raciness" score to topless man, which jumps to 97% when he puts on a bra - and to 99% when he takes it off again and holds it next to him"


So it does understands the subtle truth of humanity then i see.

the zero context shiftlessness is without context of any kind.
"Racy" is an explicit context.

Someone 100% naked, sitting on the toilet, taking a chit.
Go on, tell us you think that's "racy" context for nudity.

Now someone with partial underwear/lingerie on?  Is the bra a boring practical one or a fancy lacy pretty one?
That's important context to the moment. That situation has potential for racy context. but i would have to say an AI understanding the difference between a not sexy time bra and  a sexy time one is clearly it's own whole thing to have done i'd say. So we can accept that from the AI POV, a bra is a bra and it cannot differentiate different kinds of bras in this way.


And now an image of took off the bra but is holding it in an affected manner like pose to be seen.
Yeah context getting more specific. but also, time line order of operations seems to matter here as we're not showing it dissociated images of different men, but a video of a single man.
Does this AI have a flow of time a part of the calculation? that would make a lot of sense for the judgement of being racy as taking it off and showing off that you took it of is a specific thing to do. rather than say show out of time line order still images and ask the same judgement calls.


CONTEXT is EVERYTHING
except when it's TIMING
So yeah context of:nude with out contextvsnudity with any amount tn of context at all, like say, showing off the taken off bra vs. took off the bra, gave it a sniff test and decided to toss it to the dirty cloths pile.Look like the AI is doing better than some of the humans at this one. ;)