|
|
Post by faustus5 on Nov 15, 2017 17:46:52 GMT
It seems the same is true of neural networks. A machine learning network trained to recognize pictures of hippos could be run on specialized hardware or in a software simulator. It might run faster on the specialized hardware but they produce the same result: the details of how the network nodes are implemented simply don't matter. It doesn't matter so long as your goal is just to make a machine that satisfies your norms for what a conscious, intelligent being should behave like. If the idea is to fully replicate a human's consciousness, then it absolutely matters, and for all the reasons why functionalism is so appealing: what's floating around a single synaptic ion channel contributes to whether the neuron fires, which in turn contributes to the judgement and behavior of the neural networks it is a part of. Individual neurons are their own creatures with their own agenda's. They aren't simple mechanical circuits.
|
|