I'd have expected the opposite, unless I am misreading after too few or too many coffees
The real question for AI is this: can we do reasoning and causality without symbol-manipulation? Deep learning founders like @YLeCun (see our debate) and Hinton (eg https://sites.google.com/site/krr2015/home/schedule …) have said no; I have argued yes.https://twitter.com/GaryMarcus/status/1065280340669816832 …
-
-
-
Oops! Good catch!
- 1 more reply
New conversation -
-
-
Neural networks have sufficient expressive power, so the main difference between symbolic AI and ML is whether we use hand crafted solutions or automatic function approximation. Could you state in which way your idea of symbol manipulation differs from what a neural TM could do?
-
No, this isn't true in a practical sense. Having the system work out how to 1) Find composable symbols 2) perform constraint propagation 3) perform backtracking search 4) work out look-ahead optimizations is extra work that the NTM won't be able to do. And can be skipped
- 12 more replies
New conversation -
-
-
That's funny! Delete and repost. This is an important tweet.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Yes -> No, No -> Yes, ~(Yes ^ No) ■ ;)
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Oops you mean Lecun et al have argued Yes.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Since we are talking about doing things with computers, the key question is "can we do it well and do it efficiently". It seems we can do some perception with symbols, but not well, and not efficiently. I think causality can probably be done with both, but which is best?
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.