Is it more intuitive to say that a convolutional neural network can map sets of pixels to scalars denoting categories of objects? I think not, but it's harder to misunderstand the less intuitive description, and much, much easier to overhype the more intuitive one. "OMG, neural nets can recognise cats and dogs!". Humans can recognise cats and dogs and also know what cats and dogs are, so it's very easy to misunderstand the ability of CNNs to map between sets of pixels and category labels as a general ability to understand the world, which is not there.