Usual ML models take static slices of data and process them. Human brain, on the order hand, used to work with dynamic data, even within a single perceptual frame, and timing of events is a major variable there. No surprise that those ML models or what they call AIs are so bad at causality: they operate in different mode.