An often overlooked comment is one from Dr. Said Esquivel who headed the team that built the first computer program called an AI – Eve. In an interview marking the one year anniversary of Eve “passing” a Turing Test, Dr. Esquivel stated that:
We set up special chat rooms where we let groups of people talk. Most of the participants were general members of the public with one or two team members and every fourth chat involved Eve. The group would chat for an hour or two with only a few limits on conversation topics (emphasis added). At the end of the conversation, the administrators would ask the participants if they thought any in the group were AIs. After nearly 400 such chats, Eve was correctly identified as an AI less often than a human was misidentified as one.What were these limits to conversation? Dr. Esquivel doesn’t say. We can assume questions like “Are you an AI?” were discouraged, but what other questions would give an AI away? How old are you? (Eve had been built a few months earlier.) What do your parents do? Who was the first person you ever kissed? Left or right handed? Favorite restaurant? Most memorable birthday? Best friend growing up? These are just a few of the basic background information we humans have. If you tried talking to an unidentified entity about any of these, you would know very quickly if you were talking to a human or an AI. Unless Dr. Esquivel and his team built in false answers so an advanced computer program could “pass” as a human.
No comments:
Post a Comment