See my medium piece from earlier this week, for an entirely different view, that seeks to hybridize the two traditions, rather than dismissing either:https://link.medium.com/D2PA3HvurS
-
The man reasons with symbols and the brain works with neurons could be more accurate.
@GaryMarcus and@ylecun, you are not opposed, you are complementary, and maybe you can work together to answer that question : How symbols are represented in neural networks ?
How is symbol-manpulation represented in neural networks is the central topic in my 2001 book. My view was that “eliminative connectionism” — neural networks that try to replace symbol-manipulation altogether, as Hinton has advocated - was a mistake. @ylecun’s view unclear to me.
