The ramp video is fun, but i don't think it belongs in the same bin as other "optical illusions".
Yes, visual cortex failed, and it failed where a god would succeed (succeed in finding the simplest interpretation and making a sharp bet on that interpretation). But the most surprising optical illusions present a far lower bar which visual cortex then still fails to clear.
Often, they exploit what must be some specific implementation detail of visual cortex.
Often, they feel really weird from the inside. Sometimes they force competing interpretations, sometimes a single interpretation which turns out to be obviously wrong.
So i guess it's probably right to insist that an inference engine's optical illusions be a subset of its koans. At my level, the ramps are no koan, just a (funny and clever) lie.
But Feynman's point is: "Maybe not." We already know a computer-in-the-world can compute the fine-structure constant, in much the same way it computes its own temperature and fan speed. (I don't know enough physics to be at all sure the previous statement is true or even meaningful.) If you wanted to gauge how powerful a computer-in-the-world could be, or how powerful a computer-the-world's-in could be, you'd want to know whether a Turing machine could output the fine-structure constant, too, just as though it were pi or e or the first zero of a Bessel function.
Hire a third party (or set up a machine) to kill you if you don't drink the poison. If i were about to enter a marriage, i would indeed consider committing to it in this way.