For decades, researchers have used the Turing test to evaluate how well a machine can think like a human. But this gauge of artificial intelligence is 60 years old, and is in dire need of an update, experts say.