I recall google engineers complaining that their clever insightful carefully engineered code was soundly beaten by a statistical approach.

The current approaches aren't so much AI as having really, really, ridiculously large datasets.