It's also a question of why not? i.e. Why not hardwire certain structural regularities of the world into the model instead of forcing it do discover them. This is presumably the benefit of cognitive heuristics: squeezing more useful development into a ~5yr maturation period
Yes, I think you need a strong basis from which to learn. Not full models of the world, but enough to get started, including right representational formats and some structuring of the world (eg representations of space, time, object). Blank slate is a lost cause.https://twitter.com/pwlot/status/951041150198173697 …
-
-
- 5 more replies
New conversation -
-
-
A concept of space ("where") emerges from having physical actuators. A concept of time ("when") emerges if the model can learn sequences. A concept of objects ("what") emerges from model optimization (which separates what & where when found independent).
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Does this imply that biological evolution itself didn't start with a blank slate? If not, then evolution searched (& selected w.r.t. its environments) from blank to get here, so why can't a machine?
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.