Turing Test Near-Miss
I got all excited when I first saw this BBC article about a recent Turing Test involving five chatbots. One, Elbot, came close to passing. Then I read further and saw that the test requires that the ‘bot fool only 30% of the human audience. Elbot fooled 25% of the human testers. I think this says less about the quality of the ‘bot and more about the humans doing the testing.
You can chat with Elbot here, or read this Salon “interview” with it. I played around with the program for a while, and it really strikes me as ELIZA with a slightly better dictionary lookup. It tends to go off the rails after about three questions.
I dunno…I don’t think I’m going to be terribly excited about a Turing Test until the ‘bots are fooling 90% of the audience. Does 30% strike you as kind of a letdown, or am I expecting too much?