If EY says that starting the game with P-KN4 is a move which the real-life Kasparov would never make, this means that (either he’s lying or) he gives zero probability to Kasparov making that move, and so by construction his bot will never make that move either.
Together with the construction of RYK, the sentence I quoted amounts to “Kasparov will never open with that move, but I don’t believe he never will”.
Given the EY here in both scenarios is the active agent (EY, in flesh, and the RYK box, consisting of a synthetic EY guessing at what K does, and a system choosingly randomly based on EY-over-K predictions what the box does next)...
Yes. “never” is a strong word here. Assume when he says “never” he means, “EY thinks it is really unlikely for K to do”.
In other words, when EY sees EY-playing-K-but-Selecting-Moves-Randomly-in-proportion-with-predictions-of-K, and observes RYK has played an unexpected/unlikely move, EY concludes that RYK has probably picked a move that is evaluated to be less fit than some other move. RYK is a worse player than EY because RYK is not picking its best options. RYK is the gambler that sometimes takes the sucker bet.
Yes. “never” is a strong word here. Assume when he says “never” he means, “EY thinks it is really unlikely for K to do”.
Yes, the sane meaning for “never” is ‘negligibly often’… but then again the sane meaning for “sometimes” is ‘non-negligibly often’.
In other words, when EY sees EY-playing-K-but-Selecting-Moves-Randomly-in-proportion-with-predictions-of-K, and observes RYK has played an unexpected/unlikely move, EY concludes that RYK has probably picked a move that is evaluated to be less fit than some other move. RYK is a worse player than EY because RYK is not picking its best options. RYK is the gambler that sometimes takes the sucker bet.
I do get his point, but I think that sentence in particular is paradoxical. Had he said “rarely” and “rarely” it’d be OK, but “sometimes” and “never”...
If EY says that starting the game with P-KN4 is a move which the real-life Kasparov would never make, this means that (either he’s lying or) he gives zero probability to Kasparov making that move, and so by construction his bot will never make that move either.
Together with the construction of RYK, the sentence I quoted amounts to “Kasparov will never open with that move, but I don’t believe he never will”.
OK, I’ll play.
Given the EY here in both scenarios is the active agent (EY, in flesh, and the RYK box, consisting of a synthetic EY guessing at what K does, and a system choosingly randomly based on EY-over-K predictions what the box does next)...
Yes. “never” is a strong word here. Assume when he says “never” he means, “EY thinks it is really unlikely for K to do”.
In other words, when EY sees EY-playing-K-but-Selecting-Moves-Randomly-in-proportion-with-predictions-of-K, and observes RYK has played an unexpected/unlikely move, EY concludes that RYK has probably picked a move that is evaluated to be less fit than some other move. RYK is a worse player than EY because RYK is not picking its best options. RYK is the gambler that sometimes takes the sucker bet.
Yes, the sane meaning for “never” is ‘negligibly often’… but then again the sane meaning for “sometimes” is ‘non-negligibly often’.
I do get his point, but I think that sentence in particular is paradoxical. Had he said “rarely” and “rarely” it’d be OK, but “sometimes” and “never”...