You may not be trying, but that doesn’t mean that’s not a potential outcome. GOFAI died for a reason, dude...https://twitter.com/garymarcus/status/1003281087974531072 …
-
Fin/ If nature can create intelligence w/o innate symbol proc, we probably can too. And until someone provides an example of an innate symbol proc system that outperforms NNs in a real-world application, I’ll remain convinced that it’s just another winter you’d be in for.
How about operating systems, word processors, spreadsheets; they are all written in languages that innately incorporate symbols. You got an OS that is end to end neural nets? I’d like to see how reliable it is.
-
-
I could only process this thread by using symbols in my head. Give me another 300 threads and I will try a NN.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
An OS that is end to end neural nets; wouldn't that BE an AGI?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
You don't have to go as far as OS or spreadsheets--frameworks like TF and Py-torch very much use "symbolic" approaches to track derivatives: graphs or "tapes". The programming via architectures isnt too dissimilar to logic-based; it's just that the search strategy uses gradients
-
I think the arguments are really about end to end which incidentally brings in continuous vs discrete spaces (since the latter necessarily force a higher level). It seems that many program types naturally fit more cleanly into 1 of the two. Eg Pattern matching vs Pattern proposal
- 3 more replies
New conversation -
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.