I'm building AIs and intelligent botnets for over 13 years now. One thing I learned is that I now think humans can't understand mathematics.
No matter how good you think you are, a simple AI algorithm will proof you wrong. From physics, to algorithms, to simulations, to automation, to analysis - and in evolutionary AIs also the solution generation for a problem.
Every single scientific mathematical model that occured in my life was in an "imperfect" state where humans think that our current "model of things" in form of mathematics is correct while it's totally incomplete and a few days of AI simulation figured out a better way to do it.
I somehow see mathematics not as a knowledge pool; more like a serialization format for transferring knowledge. And I think that serialization format often is too complicated for other humans to be understood and therefore has bias in interpretation and resulting conclusion errors.
And I personally think that this is a bad thing and a problem that needs to be fixed desperately.