I am irrationally excited about this. @fchollet has asked a set of fundamentally interesting questions with ARC and I'll be playing for fun! It's like Sudoku for AI addicts ;)
(n.b. I personally got many examples wrong - huzzah, a dataset where you question your own cognition 
)https://twitter.com/benhamner/status/1228070207400202240 …
-
None of the challenges I've found in ARC so far require importing a concept such as Fibonacci. Part of the aim of ARC is that it's self descriptive. ARC does involve counting, sorting, etc, but I would argue those are "built-ins" when it comes to human cognition at this level.
Counting/numbers was identified by Spelke as an innate Core Knowledge system, shared by humans and non-human primates. That's why it is featured in ARC
-
-
was it subitizing or counting? They are different.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
This is the same Spelke we are talking about?pic.twitter.com/lj5J3BdBEA
-
I notice that it is a common bias among analytics types that assume many learned skills are innate. What's innate among humans is an approximate number system. I agree here that counting is cultural evolution artifact.
- Show replies
New conversation -
-
-
I think our innate core knowledge system is primarily visual. Words/numbers are secondary and they are dependent on and conjure up a visual representation. The name given to a number enables us to be sure that we are talking about the same thing when we communicate together...
-
In sighted people, yes. This is also why we called our architecture Visual Cognitive Computer. How to represent (and learn) the core knowledge is part of the problem to be solved. The innate biases are not a direct hard-coding of core knowledge.
End of conversation
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.