Some light discussion with some interesting examples on the use of neural networks to solve problems. Pollard describes their use to predict the success of popular songs and movies. While this is typical of the current use of neural networks, it is pobably a bit much to expect specific recommendations - such as "how to design a health care system." They don't work that way. The phenomenon described part way through the article - 'clustering' - is the result of a separate analyis. And it is very much a gloss to say, 'change this variable and you get the result you want'. That's a pojection, made by a human, based on the clusters produced by the analysis.