This has nothing to do with perception, because the difference between A and B can't be detected by science and measurement devices, perception is vastly more lax method of measurement and has no chance to perform better than science.

Yes, detection is an unresolved problem and a critical point from modern causal inference that purely statistical methods may not able to differentiate direction; even though there are quite a few algorithms using asymmetric distribution argument, i.e., KL divergence being asymmetric.