> The issue is that ML has no conceptual and causal understanding of anything Practitioners work with much more tightly defined objectives and methods for impr

> The issue is that ML has no conceptual and causal understanding of anything Practitioners work with much more tightly defined objectives and methods for improving their systems than "conceptual and causal understanding of anything". From the perspective of folks in the industry we've been making remarkable progress, well beyond what we could have expected. We're saturating major benchmarks, sometimes in one to two years. Back in the day, during the "AI Winter", it was often 10 years before a new breakthrough happened.