Several times this year, a year that marks the hundredth anniversary of the birth of Alan Turing, I have come across cases of computers being able to masquerade as humans. A few months ago I talked to a telemarketer for almost two minutes before I worked out it was a bot.
Around the same time an old friend sent me text from a chat session he had with AT&T. In this chat session, my friend asked about internet service for his home. AT&T didn’t serve that area but had wireless products. AT&T gave him several links to detailed info on those products without providing any reasoning as to why these products might meet the needs of someone who originally was looking for residential service. At that point he became convinced the AT&T agent was bot, although the agent pointed to his own spelling errors as evidence he was human. I read the transcript, and I cannot be certain whether it was software feigning human mistakes or human being with limited fluency in English.
Another case of software feigning human foibles cam up last month in the Botprize competition. Botprize is a competition in which software engineers try to create software that plays a video game in a way that emulates a human being. The game is a first-person shooter. Players indicate while playing the game whether they suspect an opponent is a human or bot. For the first time, two teams developed software that players perceived as human-like as human players.
One of the winning teams said their bot is convincing because it mimics irrational human behavior like holding a grudge and it evolves by repeating behaviors that were convincing in the past. On the surface, revenge seems like a completely irrational behavior, especially when it causes someone to take actions that are not in his interest and won’t undo the wrong being avenged. (See the short story Cask of Amontillado for chilling insight into the through process behind holding a grudge.) Revenge actually has a selective advantage because it acts as a crude system of deterrence if one person knows that another person or his family will be inclined to avenge a wrong even at the expense of their own interests. This is an ineffective system when compared even to a flawed system of criminal justice, but it was the best system available prior invention of laws. (See The Better Angels of Our Nature for further reading on how revenge worked and how civilization replaced it.)
It seems like benefits of giving computers anthropomorphic behaviors will continue to crop up. Will computers interacting as if they were humans become commonplace within the next 50 years?