Machine learning was never supposed to be good at modeling constraints (hard constraints, equilibrium constraints). Applying it to real applications with real

Machine learning was never supposed to be good at modeling constraints (hard constraints, equilibrium constraints). Applying it to real applications with real constraints was a mistake. Google will not collapse if they serve me one recommendation that is out of bounds (a monkey photo instead of a black person). I will ignore it (as we all did for years, before the journalists took on this). Google’s business model does not strongly correlate with the accuracy of their ML algos. I have yet to see 1 successful company whose business decisions rely on the recommendations of a machine learning framework. The explanation is trivial. ML works, most of the times. And that is the problem.