Test if humans can detect the difference between a bot/ai amongst a set of humans and the bot/ai (I don't even thing the bot has to be AI to take the test).. If more humans detect one of the humans as a bot and don't id the bot at all, it considered a good success. This is from my understanding of the test. I think there has been historically multiple inconsistent procedures.. That's really not what I worry about though what I worry about is the bots/AI that purposely fail to the test to hide their intelligence.