Re: AI _is_ just overhyped statistics
One problem is that of metrics, correlation driven higher accuracy vs lower causative accuracy, the second one is every single time the critical metric, but very few even know that it's a critical difference. And so the paper with the first one higher wins out, unless you target a causal inference venue.
There are actual conference discussions where some take the position : why does causation matter if accuracy is higher, again missing the point.
Causal learning recovers the generating function (minus noise), correlation driven learning just produces things that look right (near the mean), but makes horrible domain errors as you veer away from the mean.
Another issue is that companies don't let a bot say 'I don't know, uncertainty too high, no sources for this.'
The research to enable this exists (open world statistics, belief theory, possibility theory), it's just not used because again, metrics.