This article is a small sample from DISCOVER's special issue, The Brain: An Owner's Manual.
“Can machines think?” In 1950 mathematician Alan Turing pondered this question and invented an elegant game to answer it: Let a human chat via Teletype with a computer and another human; if the person can’t determine which is the computer, then it meets Turing’s standards for “thinking.” In recent years Turing’s game has taken on a life of its own in cyberspace, thanks to artificial intelligence inventors worldwide who have produced dozens of “chatbots” that anyone can talk to.
Most chatbots rely on fairly simple tricks to appear lifelike. Richard Wallace, creator of the top-ranked chatbot ALICE (Artificial Linguistic Internet Computer Entity), has handwritten a database of thousands of possible conversational gambits. Type a comment to ALICE, and it checks the phrase and its key words for a response coded to those words. In contrast, Jabberwacky, another top-rated Internet bot produced by Rollo Carpenter, keeps track of everything people have said to it, and tries to reuse those statements by matching them to the writer’s input. Neither chatbot has long-term memory, so they respond only to the last sentence written.
Nonetheless, these simple gambits can produce surprisingly intelligent-seeming conversations. That’s because they rely on a trick of human psychology: We humans tend to attribute much more intelligence to the systems than is actually there. If it seems partly aware, we assume it must be fully so. Some users have chatted with ALICE and Jabberwacky online for hours, apparently not knowing—or perhaps not caring—that they’re fake.
But could one chatbot fool another chatbot? What would one say to another in private? To find out, we arranged a conversation between these two chatbots. To get each snippet of chat rolling, we seeded it by posing a question from one bot to the other. After that, they were on their own. What follows is the unaltered text of what each said—the sound of two machines talking.
