One way to describe intelligence is as "the ability to understand something from a partial description". How minimalist this description can be is a measure of intelligence.
Show this thread
-
If your AI needs an excruciatingly exhaustive description of its task (as an explicit handcrafted program, or as a dense sampling of a static data manifold), it isn't actually intelligent. When you operate it, you're just querying the information you put into it. No autonomy.
Show this thread
Intelligence is the ratio with which your AI converts information about a task into skill at that task. Importantly, the task is typically not static, so "skill" here targets *future* situations, which may be very unlike past ones.
-
-
The conversion is from past operating space to future operating space.
Show this threadThanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
But in limited samples, there is no application of a "mirror concept" to be applied unless found in the examples given, so there is no skill building to other types of tasks.
-
Still limited to solving problems at the interface layer.
End of conversation
New conversation -
-
-
As seen from the perspective of a human.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
This is a nice subtle point with a small range of interpretations. To me this seems related to the concept of “affordances” in an agent’s environment. A handle suggests pickup-ability to an agent who has a pick-up skill, even if the object is novel otherwise.
-
In ARC, tasks often “suggest” the effect of such skills as “movement,” “copying,” “recoloring” etc to our human minds, which are anticipating information in the form of affordances—even though there is no literal, logical trace of such frames at work.
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.