I would want the system to be trained by something like gradient descent or REINFORCE. The program should be written in the "deep learning style". So a tiny bit of gradient descent would definitely not qualify. Writing a DL lisp interpreter and coding in lisp would not qualify
what is excluded? and what about my point re distinguishing deep learning with and without symbol-manipulation? if you count anything with even tiny bit of gradient descent no matter how what else aren’t you just relabeling my point about “one tool among many”?https://twitter.com/tdietterich/status/952730726075572224 …
-
-
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
This Tweet is unavailable.
-
-
This is only partly correct. The goal of the programm in the chinese room i.e. Searle, has to pass the Turing Test. The state of art ML translation can't pass an Turing Test.
- 3 more replies
New conversation -
-
-
The heart of your argument is that Deep Learning cannot be of value for AGI without symbol manipulation. I contend that symbol manipulation isn't necessary. Machine language translation does not use symbol manipulation.
-
What is a symbol? Is a pixel a symbol? Hardly. But we can represent a text into an image using embeddings.
End of conversation
New conversation -
-
-
That's correct, it isn't conscious with what it does. Consciousness is an emergent cognitive capability and not all cognitive tasks require this. See: https://medium.com/intuitionmachine/deep-learning-system-zero-intuition-and-rationality-c07bd134dbfb … Where I spell out 3 axes for determining cognitive progress.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Perhaps understanding of a task is an overrated requirement. That's the nature of intuition based system, you don't have to be exact. Approximations are just good enough to get stuff done.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Yeah but thats because you can understand things or at least have semantic representation of objects and their relations.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
We are falling in the old fallacy of anthropocentric intelligence. Meaning is just relating sensory inputs with representations. Humans do it one way machines in other ways
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.