[j2kun]

There are two contradictory claims:

1. The brain is like a neural network (which is purely logical) in the sense of ML.

2. Human brains cannot be explained by purely logical things.

The author also uses "concept," which is a technical term in computational learning theory with a specific meaning, as if it meant "intuition." How can you present "intuition" to a neural network? This distinction is swept under the rug. Not to mention all the recent work showing how easily neural networks can be fooled by slightly adversarially noisy inputs.

There are many grains of salt required for a useful discussion on neural networks. Instead of taking something we have no understanding of and making grand philosophical claims, we should be using the tools we have to understand that thing.

[eli_gottlieb]

Very much agreed. For one thing, comparing the human brain to a deep neural network leaves out the fact that the human brain mostly performs unsupervised perceptive learning, unsupervised causal induction, and reinforcement learning. None of these resemble the deep backpropagation done in most ML models.