"Nearly" passing the Turing Test?http://gizmodo.com/5062385/computer-nearly-passes-turing-test-for-artificial-intelligenceToday I realized just how "gray" the Turing Test really is. I've been aware for some time that a man by the name of Hugh Loebner set up an official Turing Test, but I didn't realize it would be this easy for developers to use smoke-and-mirror techniques to score well on it. It's all a little disappointing, really. If I were to define a Turing Test, it would involve producing a machine that attained something like a 90% fool rate after sessions with 10 highly trained interrogators which each asked the machine 100 questions. It's the "highly trained interrogators" that is key here. For example, I can guarantee you that none of the machines tested this weekend would have even a small hope of answering a statement/question such as this: Question: "I'm thinking of something that is big and round and 6 billion people live on it. What am I thinking of?" Trained interrogators would ask simple questions such as this, and today, the machines would likely score 0% on a series of 100 questions and attain a 0% fool rate. And that, I think, would demonstrate where we're at! On the other hand, I think it's good to have a contest that allows the current state of the art to show off and gives researchers something manageable to work towards. But I wish they wouldn't call it the Turing Test, and then start talking about sentient life forms... please... |