I don’t think he’s saying that error is good. He’s saying that ignoring error is bad. And there’s a valley of bad rationality in which the error level is low enough that people ignore it, but it still matters. Specifically, he says that black box models that don’t include error are bad. You should model error of components and error propagation and design circuits that are robust to error. I’m not sure what this would mean in practice, though. Today we design systems that recover from computers crashing, but recovering from wrong computation is harder.
But this is 1948. In 1956, he showed how to take digital gates with known error rates and build from them digital gates with lower error rates. This opens up the possibility of getting the error low enough that you can ignore it and embrace digital logic. I’m guessing that he changed his mind and rejected the earlier paper, but it would be nice to have an explicit statement.
I don’t think he’s saying that error is good. He’s saying that ignoring error is bad. And there’s a valley of bad rationality in which the error level is low enough that people ignore it, but it still matters. Specifically, he says that black box models that don’t include error are bad. You should model error of components and error propagation and design circuits that are robust to error. I’m not sure what this would mean in practice, though. Today we design systems that recover from computers crashing, but recovering from wrong computation is harder.
But this is 1948. In 1956, he showed how to take digital gates with known error rates and build from them digital gates with lower error rates. This opens up the possibility of getting the error low enough that you can ignore it and embrace digital logic. I’m guessing that he changed his mind and rejected the earlier paper, but it would be nice to have an explicit statement.