I maintain that it isn't just hard, it is computationally impossible.
We should all know that given a belief about the world, and evidence, Bayes' Theorem describes how to update our beliefs.
But what if we have a network of interrelated beliefs? That's called a Bayesian net, and it turns out that Bayes' Theorem also prescribes a unique answer. However, unfortunately, it turns out that working out that answer is NP-hard.
OK, you say, we can come up with an approximate answer. Sorry, no, coming up with an approximate answer that gets within probability 0.5 - ε, for 0 Therefore our brains use a bunch of heuristics, with a bunch of known failure modes. You can read all the lesswrong you want. You can read Thinking, Fast and Slow and learn why we fail as we do. But the one thing that we cannot do, no matter how much work or effort we put into it, is have the sheer brainpower required to actually BE rational.
The effort of doing better is still worthwhile. But the goal itself is unachievable.