There’s an astonishingly simple fact about the world, the possibility of the world that is, and the possibility of knowledge (human or otherwise). A fact humans first identified thousands of years ago and codified it into immutable limiting fact in the 1800s.

A perfect, actual circle is not construct-able (in Euclidean space), not with a protractor, nor a computer. That is it is not possible to implement physically a circle, in any medium by any creature, machine nor material object of any kind. Every circular thing is at best a construct-able approximation.

It is possible to reason and imagine a circle or rather use properties of a circle to make things or talk about things. You can think of a circle as a limit, average, mean or ideal of circular things with circular properties.

(It’s even possible to invent spaces in which it is possible to do more (or less) with circles…. you decide at which point we’re not talking about a circle…)

My suspicion is this is what Plato meant by ideal forms. Some observations of the world only exist in-between actualizations. I suspect this is also why we will never pin down some definition of intelligence or morality or anything depending on absolute predictability. You may say “being able to have very good approximations is reliable enough!” But you just don’t know when having one more digit of pi in your circle or one more infinitesimally small line segment added to your n-gon would have averted disaster or made some model of the world seem so much better and so on.

The wobble around approximations of a circle implies that no machine or creature will ever really be able to go “alright that’s it! No more learning need be done!” In fact, often better and better approximations of an ideal like a circle can make a person or a machine less reliable. Knowledge has a cost.