User blog comment:Alemas2005/SauceBot Busted! - Chat Blog/@comment-24057825-20120825090111/@comment-4052371-20120827040406

Simple roulette wheel? How about a random number generator? Also, just because it makes mistakes doesn't mean it has to correct them. Maybe if it detects a message with its own username, it just enters it into the log and ignores it.

And also, bots that accept an information input and return an output are very common (For example, the patrol bots on Wikipedia read recent changes and tell the server to revert them if they think they're vandalism). So there's no reason it couldn't simulate a human in a conversation.