As a first approximation, why not? Behavior is generally all we have in front of us, plus any other assorted social signals. Internal mental states are invisible, as is the personal history of the individual. We might note a man beating a kitten on the sidewalk, and believe this behavior sufficient grounds to reduce this person to the category "dick", even if we remained unaware of his high intelligence, his doctoral paper on gender-inequality, and the fact his mother hates him.
> But when we assess a machine, we do know everything about it based on its actions.
This is obviously false. Every developer has had hour-long debugging sessions to track down a mysterious behavior. Sometimes entire teams are stumped by a technical glitch. Until the bug is found, nobody knows everything about the machine.
Or whether it’s only a small subset who do.