I share your worries, as I too think that the abstractions used in the trained models in the talk are inherently leaky, meaning that they do not represent reali

I share your worries, as I too think that the abstractions used in the trained models in the talk are inherently leaky, meaning that they do not represent reality to a suffiently high degree (see corn flakes vs ketchup screen wiping action). As we do not (yet?) understand the underlying nature of those "2.0-ish" decision algorithms, we will not be able to make a judgement call when they will work, and when they will fail.