Computers changed the way people played chess:
We used to think that computers would play chess like we did, only “without the mistakes.” We now know that playing without the mistakes involves a very different style from what we had imagined. A lot of human positional intuitions are garbage, and the computer can make sense out of ugly-looking moves. A lot of the human progress since then has involved unlearning previous positional rules and realizing how contingent they are. Younger players, who grew up playing chess with computers, are especially good at this. For older players, it is a good way to learn how unreliable your intuitions can be.
We tend to find when doing baseball research, that the managers get strategy right as a group. They steal and bunt more when the run environment is low, less when it is high. While the lineups most managers use are not optimum, they’re very close to optimum. Much to my dismay, they get reliever strategy right. Baseball strategy is evolutionary, where if you don’t manage the game the right way over time, you lose. That gives us managerial moves close to what we’d get from a computer.
I have some issues with your analysis. I think it’s shrinking the gap considerably, because of the simplifications you made to do the analysis.
I think you need a weighting system. To me, a reliever who pitchers 4 innings and allows 2 runs (say, the 6th, 7th, 8th and 9th) did not have a “bad” outing. But in your analysis he does. Contrast that with a reliever who comes in for a tight ball game in the 8th and allows only 1 run. He is classified as “good”, when in reality that 1 run could be/likely is very costly. I think these factors are considerably skewing the results to bring the two together.
to use your baseball analogy ,
in chess , computers don’t “kick in”
until , about , the 4th or 5th inning .