How To Be Human
Using call center data to train “chat bots” might be just the thing we need to help push artificial intelligence to the point of passing the Turing Test.
For those not familiar with the Turing Test, a short description is that it is a test where a one human and one AI are set up to have an instant message (or IRC) chat with a human “judge” who tries to determine which participant is the human and which is the machine. If the AI can reliably fool the judge in to thinking it’s a human, than it’s passed the test. So far there is no system that’s been able to fully pass that test yet, but there is a $100,000 grand prize for the first success! The Loebner Prize is a version of the Turing Test and each year there is a prize to the AI that seems “most human-like” among it’s competitors. For 2005 and 2006 that has been Rollo Carpenter’s “Jabberwacky” system.
Rollo has a unique way to get his AI system to seem human, and that is to interact with humans in a conversational (chat) manner and store all reponses the humans give during the conversation. Then when the next person comes along, it can search it’s database for appropriate human-like responses given to it in the past. This method seems to be working!
But, now he’s saying he needs more data… MUCH more data ! And he’s decided to go to call centers where he can get at the vast amount of human interaction and responses to help fill his database at even faster speeds.
To pass the Turing Test, a system does not really need to be “intelligent” as we might think of it being able to come up with new solutions to problems or be able to use reason and logic…. it only needs to be able to reliably convince a judge that IT is the human, by upholding it’s end of chit-chat in a conversation.
Rollo’s main goal right now is for his systems to be able to converse on enough topics that it could be used in toys and other entertainment applications, such as talking pets and televirtual companions. He is already working with a call center in Japan, and hopes that one day the AI will be able to take over the role of the human operator.
I’ve chatted with various online bots before, and sometimes for up to 30 minutes. I can’t remember any conversation being overly rewarding, or interesting enough to come back for more, but I do know that it won’t be many more years before these systems will be able to keep some people entertained and interested enough that they might look forward to getting online for hours each day to chat with their new virtual friend. Perhaps even have some epiphany or mind expanding thought come about from such a conversation. When the technology is good enough to be able to talk with you about topics you are interested in, and has access to the wealth of knowledge on the internet you might find yourself enjoying the company of an artificial intelligence as even more than a few of your human friends!
1 Comment(s)
Comments RSS TrackBack Identifier URI
Leave a comment
You must be logged in to post a comment.

[…] It seems like only yesterday when I wrote a post about “George”, the artificial intelligence software which (who?) has won the Loebner Prize two years in a row now, for being able to hold he most human-like conversations. This new article reports that now George has been set up with voice recognition and speech synthesis so he can carry on a voice-to-voice conversation as well, and is currently chatting with over 200 people on the internet simultaneously. As posted earlier, there are projects in the works to use this technology in support call centers, toys, virtual pets, celebrity impersonations, historical characters used as tutors, and any number of other “interactive” applications. FINALLY, a pet who will not only be able to understand English much better, but also be able to reply back to me in English! Maybe my search for the perfect woman is coming to an end, as well?? […]