There's a whole massive underlying assumption here: that we will soon have affordable, deployable, non-finnicky machine learning models for enough problems to d

There's a whole massive underlying assumption here: that we will soon have affordable, deployable, non-finnicky machine learning models for enough problems to decimate the middle class. In my experience, ML models, even the best deep convnets and such, are still extremely "finicky" by human standards, in the sense that small or human-insignificant changes to the input vector can cause large-scale changes in the output classification/regression. Cool demos with video-games need to be run past some industrial automation experts before we go around proclaiming them to be the harbingers of industrial AI!