But what is cognition? People should reevaluate their views on how it works.
The brain is very complex and a very big ball of highly specialized circuitry (and the software running on those and managing various aspects and parts of it and them). The similarities are at the same time very striking but the differences are just as drastic in number of layers and other size related parameters our various brain components have.
Yes, AlphaGoZero is not going to learn to cook and sing and dance, but it beats our gaming component in a lot of areas. Of course a synapse is not really a memristor or a ReLU, but at the same time not that far off either. Similarly a biological neuron is not just a simple backpropagatin integrator like a perceptron, but it works very similarly.
And just as we don't really know how all the representations work in an ANN, we don't exactly know the biological aspects of memory/learning/seeing work in our brains, yet we have made enormous progress on both. See all the demos/visualizations on how various layers encode in deep nets and look at the data about the brain's visual cortex and the V1 circuit, look at the gene spliced mouse studies where memory encoding is studied (and sometimes only one neuron encodes for a memory/face/concept - just as with deep nets).
The brain is very complex and a very big ball of highly specialized circuitry (and the software running on those and managing various aspects and parts of it and them). The similarities are at the same time very striking but the differences are just as drastic in number of layers and other size related parameters our various brain components have.
Yes, AlphaGoZero is not going to learn to cook and sing and dance, but it beats our gaming component in a lot of areas. Of course a synapse is not really a memristor or a ReLU, but at the same time not that far off either. Similarly a biological neuron is not just a simple backpropagatin integrator like a perceptron, but it works very similarly.
And just as we don't really know how all the representations work in an ANN, we don't exactly know the biological aspects of memory/learning/seeing work in our brains, yet we have made enormous progress on both. See all the demos/visualizations on how various layers encode in deep nets and look at the data about the brain's visual cortex and the V1 circuit, look at the gene spliced mouse studies where memory encoding is studied (and sometimes only one neuron encodes for a memory/face/concept - just as with deep nets).