[Pharaoh]

A good programmer can write an EA that's perfectly optimized for backtests. In the real world, market conditions change all the time, and a program that was perfect for last year may fail terribly this year.

I believe that it is possible to create a good EA, but that EA would have to be "smart" enough to avoid big news events and determine if market conditions are optimal for its strategy(ies).

There are a lot of EAs that will happily follow a trend and make you some money, but then will wipe you out when the trend reverses. A lot of them run with no stoploss with the idea that any reversal of a trend is temporary (a recipe for a margin call unless each trade is a tiny fraction of a percent of your account) and also to be able to show a higher percent of winning trades, even if the losers are devastating.

I'll second Ernest's advice. If you want to test EAs with real money (AFTER demo testing them), use an account that allows nanolots. Then a 1000 pip loss is only $10. Only scale up trade size if you are convinced that the EA can handle itself well under a variety of market conditions.