I was reading about the proposed SDK, and I was wondering, how does uncertainty drive exploration? For instance, suppose you are blindfolded and feeling the contours of an object with your finger. You are uncertain what it is. How does a learning module decide where to move your finger so that the uncertainty is dispelled? I suppose you could move your finger to every spot on the object, but I’m assuming that real minds are more efficient.
This question could apply to higher level problems too, not just sensory.
2 Likes
You move it till you can predict how the sensory input changes when you move it
1 Like