If this last week's episode does one good thing it will be to convince programmers that it's not that hard to get started doing ML stuff. Not to say that it's all trivial or straightforward, if you look at the linked image below you can see that a simple NN will miss lots of stuff. In this example the image is being examined for hotdogness as a whole, not being segmented or searching for a bounding box of hotdogness.