Here’s what I have been saying since 2001: “Neural networks have to be supplemented with the key components of symbol manipulation in order to capture the compositionality of language and thought.” Agree or disagree, @ylecun? At our debate you said “diagree”. Your stance now?https://twitter.com/ylecun/status/1065298686413078528 …
-
Perhaps
@GaryMarcus is claiming that we will need to explicitly add "symbolic components", specifically designed to power symbolic reasoning in a targeted way; whereas@ylecun thinks it might be possible for symbolic reasoning to develop emergently in a more agnostic substrate?
my claim is slightly more subtle: any successful general AI will have components that map directly onto the key machinery of symbol manipulation, viz variables and operations over those variables, including machinery for binding and instances. see chapter 2-3 Algebraic Mind.
-
-
Still trying to understand - when you say "components" / "machinery" do you refer only to mechanisms placed there by the designers? Or does your claim entail mechanisms that arise emergently: hard for human observers to find / point to, but demonstrably real if we did find them?
-
What matters is not designer intention, but whether there is a perfect mapping. This is something I go over in detail in Chapter 3 of The Algebraic Mind, showing examples of cases where some models map perfectly and some don’t, indepedently of what creators assert.
End of conversation
New conversation -
-
-
Is this symbolic machinery a virtual simulation or manifested as real machinery?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Do you think that insects meet those criteria?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.