The poster says the above is exact inference. I beg to disagree. The inference is plausible but not exact. How do we know that Mary has not given the book to someone else, or even how do we know that Mary hasn't eaten the book for dinner? We humans know that books are not tasty, but how would a machine know that if we don't tell it?
Of course we need a plausibility model. My point is only that exact inferences are extremely rare, and that almost all inferences we make on daily basis are plausibility inferences based on an extremely comprehensive model of the world.
The poster says the above is exact inference. I beg to disagree. The inference is plausible but not exact. How do we know that Mary has not given the book to someone else, or even how do we know that Mary hasn't eaten the book for dinner? We humans know that books are not tasty, but how would a machine know that if we don't tell it?
Of course we need a plausibility model. My point is only that exact inferences are extremely rare, and that almost all inferences we make on daily basis are plausibility inferences based on an extremely comprehensive model of the world.