Theory: In 90% of real world Machine Learning use-cases, the ML could be replaced with alternative techniques which are 90% simpler (to understand, deploy, maintain, etc.) and 90%+ as effective. The results of design driven by techno-fetishism, instead of real human needs.