Isn't the other side of the coin, though, that in these types of systems you can actually get a lot "wrong" and they still work? That is, in the example of a sa

Isn't the other side of the coin, though, that in these types of systems you can actually get a lot "wrong" and they still work? That is, in the example of a sand pile, if you get the rate of adding sand "wrong", it doesn't really matter: much the same thing happens. That's kind of the whole point. Whether that means that in simulating the brain it may turn out that you can get a lot wrong and still have a successful outcome is TBD, but I'm not sure the expectations are as solidly negative as the article suggests.