“We can build these models, but we don’t know how they work.”

When talking about the simple loan models, etc, this seemed dubious to me. Are engineers with no knowledge of ML building the models?

It works by showing a probability of correlation between features and output after careful training, validation, feature selection, and weighting on a lot of data. Sure, you may be surprised sometimes, but it just means there were patterns in the data that you weren't aware of.

I did think the deep learning car that won't listen seemed cool, though.