You are quite correct in your comments that neural networks will not predict the winning numbers everytime. They are however very powerful prediction tools that when used with other ai toolsets can reduce your odds and give you more winnings than other methods. The trick is to model the lottery you are playing correctly and then let the system work.
Here are the results of the first neural network we created in 1999. We have never changed the model and if you click on the link below you will see the number of winning categories and the dates for the winning numbers. The first tab gives you the summary of winners by category, the second tab gives you the dates and the time between the draws that a category won a particular prize.
http://www.ailotto.com/Cossack%20Model.htm
For those thinking the lottery is a waste of money we also track our neural network in similar fashion to how mutual funds and other stock portfolios gauge an investment return. Based on a $10 investment in 1982 and if all the winnings are keep in a bank account and only tapped when you run into a long period where you never win, the amount would have grown to over $16,911 by July 2006. In many ways this is a much better return than what you could have gotten investing $10 in the stock market!
For those interested if you send me an email to info@lottopirate.com I will send you the spreadsheet model used to track the success of the neural network.
So don't discount the power of neural networks, they work and are the next generation beyond simple wheeling systems that produce spectacular results when you properly model the lottery draw you ar playing.
Lotto Pirate