clipped from: technology.newscientist.com Eduardo Miranda shuts the door of his study, leaving two "warbling" robots to their own devices. He has programmed them to blurt out sequences of random notes, and two weeks later, he returns to find that the robots are still cooing in their eerily human voices, but they have now "evolved" to sing a repertoire of 20 sounds together.
Miranda equips each robot with software that mimics the human voice, and gives each a microphone that acts as its ears and a camera for its eyes. One robot starts by babbling a random sequence of about six notes. When the second robot hears this, it responds with a babble of its own. The first robot then compares the two strings of notes. If it deems them to be similar, it nods. The second robot detects this and "memorises" the settings that created the sequence. If the noises are dissimilar, the first robot shakes its head, causing its partner to discard, or "forget", that sequence.