I just finished The Most Human Human by Brian Christian, which is about computers simulating human conversation (one working definition of AI). You will get a detailed summary from this interview. Computer simulation of other supposedly exclusively human skills has been all over the news lately, too: moral and ethical improvisations in The New York Times, identification of emotions via facial expressions in the New Yorker.
AI is always all over the place. The IKEA chatbot, for example. It was programmed by a previous winner of the Loebner prize, but was completely unhelpful to me. I asked it if I could arrange for someone else to assemble my order, and it just referred me to generic web pages that did not contain the information I wanted. Then it asked whether its answer had been helpful. When I said No, it responded "Sorry. As an IKEA Online Assistant I don't know the meaning of 'no.' " Well, that's adorable, but shuts down the conversation, doesn't it? As I learned from the book, a good conversationalist offers plenty of holds with her responses, instead of just saying, "That's not what I wanted to hear."
My favorite robots are the ones that don't pretend not to be, like the CVS prescription reorder system. "Press or say 1," "The first three letters of the patient's last name are..." Okay, that's a conversation we can have. If I have any actual questions, I can press 2 for a real person.
So the topic of conversation simulation is interesting in itself, but the description of Christian's experiences as a Loebner Prize confederate went beyond interesting to inspiring. I even emailed the organizers about the possibility of participating next year! If nothing else, what a great excuse to take a trip to Bletchley Park!
No comments:
Post a Comment