Mainly: casting the end goal of intelligence as the optimization of an extrinsic, scalar reward function. But also:
Show this thread
Posit: the conceptual framework of reinforcement learning brings misconceptions about intelligence that have & will set back the field of AI
-
-
-
casting agent(s), environment, and reward, as separate, relatively static entities.
Show this thread -
I'd say intelligence is not an optimization process, and these objects are not the key concepts.
Show this thread
End of conversation
New conversation -
-
-
What are the misconceptions?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
What is/are the reasons for this and how can those working in field of AI alter their perspective to avoid these set backs?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
On every test set, right back to first DARPA speech it blows the alternatives away. Maybe we need new benchmark problems.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Take a step back. Look at the funding and hardware advances which have appeared because of these “misconceptions”.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
I don't know. My control policy has been reinforced by intrinsic and extrinsic rewards (emotions, hormones, money, etc).
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
I do agree. Humans learn in a much more efficient way, and use rationality and logic on top of all
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
How can it be that RL, which attempts and succeeds to achieve more than SL/UL, is setting back the field of AI ?
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.