Just the fact that it has been trained by playing against people can sometimes be used against it. Just played 20 times, and won 10, tied 5 … although may not be statistically significant, I think it can be beaten often by thinking along the lines of “what would most people do” and then countering it.
…, which means, just by permuting the R, P, S action labels to P, S, R ones and prompting the permuted interface to a human, then the computer will be beaten.
Just the fact that it has been trained by playing against people can sometimes be used against it. Just played 20 times, and won 10, tied 5 … although may not be statistically significant, I think it can be beaten often by thinking along the lines of “what would most people do” and then countering it.
I expected it to play paper first, to combat the expected rookie move of playing rock, and surely enough it did. So you have a point on some level.
…, which means, just by permuting the R, P, S action labels to P, S, R ones and prompting the permuted interface to a human, then the computer will be beaten.
Am I right?