A persistent strain of thought assumes static robotics models are useless unless they're trained on the exact environment they operate in, and I'm not sure why.
That aside, arguments from handwaving about biology are extremely weak - this is the same proof structure used to explain that because we only have two eyes, a machine only needs 2 cameras to drive a car safely. It wants you to forget everything that happens after the light hits the retina, and difference between an always-wet lens that's cleared every second or two and a camera. i.e. it's a noble goal to have some model that never saw a kitchen also be an expert chef, but, there's no logical reason to claim that'd ever be the case.
That aside, arguments from handwaving about biology are extremely weak - this is the same proof structure used to explain that because we only have two eyes, a machine only needs 2 cameras to drive a car safely. It wants you to forget everything that happens after the light hits the retina, and difference between an always-wet lens that's cleared every second or two and a camera. i.e. it's a noble goal to have some model that never saw a kitchen also be an expert chef, but, there's no logical reason to claim that'd ever be the case.