Synopsis: The Academy Award-nominated film, “The Imitation Game” brought to light the contributions of Alan Turing in breaking the Enigma code during WWII. Throughout his career, he thought that maybe we could build a computer that could think like a human. But could even the best tech whizzes of today build a machine that could not only think but also converse like a human? We talk to a man who researched that issue and even competed in a contest to prove he was human, and find out what it takes for chatbots to beat humans at their own games.
Host: Marty Peterson. Guests: Brian Christian, author of “The Most Human Human: What talking with computers teaches us about what it means to be alive.”
Links for more info:
Most Human Human
Marty Peterson: The Academy Award-nominated movie, “The Imitation Game” starring Benedict Cumberbatch and Keira Knightley brought the life and work of Alan Turing into the public consciousness, along with the idea that maybe you could build a computer that thinks like a human. We broadcast a story in 2011 about that topic, and thought it would be fun to revisit it now. You probably remember the I-B-M computer “Watson” winning the “Jeopardy Challenge” in February 2011. When the two all-time human champs went down in defeat, it was a triumph for the men and women who created and programmed the one-of-a-kind computer. Although I-B-M engineers insist that Watson is far from human, the computer spoke, used strategy to run the board and even made some unusual wagers. He seemed to be at least a little bit like a real flesh-and-blood contestant. But can a computer really mimic a human? Brian Christian wanted to find out, so he competed in the annual “Loebner Prize.” Christian chronicles his adventures into what makes us human – or at least different from computers – in his book, “The Most Human Human: What talking with computers teaches us about what it means to be alive.”
Brian Christian: I first got involved around 2008. I’d been reading these articles about a competition called the Loebner Prize, which is an annual Turing Test competition within the field of artificial intelligence. Basically the way it works is a number of computer programs are talking over chat message with a panel of judges. And they’re essentially trying to convince this panel of judges that they are actually human beings typing back and forth. And at the same time the judges are talking to several real people who also trying to say that they are, in fact, the real people. So it’s this very strange and quite intriguing competition that not only gives us a technological benchmark, but it also for me asks these deeper, more philosophical questions of “what does it mean to be human? What does it mean to act human in this particular contest?”
Peterson: Christian says that the test was proposed back in the 50’s by the British mathematician and computer scientist Alan Turing. It was a way to answer the philosophical questions, “can we create a computer that really does think? And if we did, how would we know?” The Loebner Prize consists of 100-thousand dollars and a gold medal to the creator of a computer program—or “chatbot” — that can fool human judges into thinking it’s actually human. Christian was a confederate…one of the real people who texted online with the judges — so he had to devise a strategy to convince them he was actually “human.” Wait a minute. He is human. Why the strategy?
Christian: That for me is one of the fascinating paradoxes of this test, is that it’s this strange space of high-pressure chitchat. And of course I am human, so a big part of it is just be myself, to act natural. But to some extent, I didn’t want to go in unprepared, you know, I wanted to know what I was getting myself into. I think to some extent it was just a question of wanting there to be an even playing field. Because a lot of these programmers spend years looking over transcripts of these Turing Tests, trying to really fine-tune their programs, really kind of poring over what makes a conversation work structurally. We humans, in part because it comes so naturally to us, often don’t think about those things.
Peterson: So Christian prepared by studying up on linguistics and looking back at previous Turing Test transcripts. He also delved into very “human” topics like chess, jokes, poetry, graphic arts, and even how pick-up artists operate to try and figure out what makes our conversations human….
Christian: In human society there’s this tremendous demand for public speaking coaches, or conversation coaching, or dating advice, that kind of thing. So it seems actually to me that the art of making good conversation isn’t something that we’re just automatically experts at, but it’s something that we have to try to strive for. So part of my quest as a human competition is to try to have the best conversation I could. And so in this sort of counterintuitive way, looking at how these programmers were trying to imitate human conversation really shed a lot of light for me on how human conversation actually works.
Peterson: Of course, he wasn’t conversing in the truest sense of the word. Everyone who texts, emails or actually writes real letters knows that writing and speaking are completely different. When you type, there’s no inflection or body language to key off of. So how do you get your “human-ness” across in an online chat?
Christian: I wanted to figure out basically how to use the medium as much as I could. So the Loebner Prize is done in this rather unusual style, where not only are you sending these typed messages back and forth, but you can literally watch each other’s typing. So letter by letter, and you can watch them make typos and backspace and all these things, which is funny because the computer programs are designed to fake typos. I think compared to something like text messaging or tweeting, there is this opportunity when you can see each other typing to create that kind of really dynamic conversation that we’re used to having out loud. Where people are finishing each other’s sentences, and kind of talking over each other and fighting for the control of the conversation in a way that is really natural to us especially when large groups of people get together. But we’re not used to it when we talk to computers, we’re really used to this very rigid question and answer format. So part of my goal was to try to break that format and create the kind of rich, dynamic sort of everybody leaping in style of talking that is much more common to us.
Peterson: Two of the things that Christian discovered that were inherently human were jokes and wit. Could chatbots learn to use and appreciate humor?
Christian: Really the only answer that we have is that you can tell them a joke and they can repeat it. Which is not the same thing as having a sense of humor. One of the critical things about wit is that it’s fast. The French have this great expression, “l’esprit de l’escalier,” I’m butchering the pronunciation, but it translates to “staircase wit” – the devastating verbal comeback that occurs to you just as you’re leaving the party, you know, and it’s 30 seconds too late. That’s really the beauty of wit – is that you have this really narrow window to kind of get it in and after that point, it’s kind of overripe, it’s past the point where it’s funny. And computers do not really have that ability to be operating in real time, you know, responding instantaneously to the situation.
Peterson: One of the challenges to programmers who design chatbots is how to make a conversation evolve, rather than have it be just a series of disembodied questions and answers. Christian says that a lot of the programs are cobbled together out of chat room transcripts and other conversations picked up online. This might give the chatbot a sense of what “real” communication is like, but it does so without context…
Christian: As you’re talking to these programs, they have a really smart answer for everything, but the answers don’t quite add up. So you’ll say, you know, “how far away are you from England?” and it’ll say “Oh, that’s all the way on the other side of the ocean.” And you’ll say, “tell me about your commute,” and it’ll say “oh, well I just ride the tube every day. It’s only half and hour.” It’s not so much that you’re not talking to a person, but you feel that you’re not talking to a person. It’s this sort of Frankenstein of all these different people. So I think part of what’s great about real-life human interaction is that you build up this really coherent sense of who someone is.
Peterson: So what was it like actually competing against chatbots to prove you’re the most “human human”? Christian says it was kind of weird…
Christian: A little bit like the SATs meets a speed date, or something like that. Where you know you have this strange situation, you’ve only got five minutes and you’ve got to try to come across as this distinct, living, breathing person with a unique personality. And so there was the weird sense of pressure, where you know the judge will begin the conversation saying something like, you know “hi how are you?” these really sort of classic, time-honored conversation starters. And you know I found myself in this very strange position of you know I’m typing “I’m good, thanks, how are you?” but realizing as I’m saying it that we’re participating in this real conversational boilerplate. You know, when someone asks you on the street “how are you?” they don’t expect you to really answer. And if you really answer, you’re sort of overstepping tact. But on the other hand, you know, we only had five minutes. So there was this weird sort of pressure to connect in an intimate way that most conversations don’t have.
Peterson: He says that winning isn’t just about sounding like a human. There’s the interaction with your judge that figures in as well…
Christian: For all my preparation and all that I knew about how the software worked, there’s still no accounting for the fact that you may really hit it off with that person and you may not. So, I noticed as I looked over at one of the other humans’ computers he was a Montreal hockey fan, and he was talking to a Toronto hockey fan. So they’re sort of letting rip with all these abbreviations and slang, they’re talking trash to each other and I just thought, oh gosh, you know I’m doomed. I don’t have that kind of connection to the guy that I’m talking to, so there’s only so much I can do to kind of make up for it, because these guys are just hitting it off.
Peterson: Christian did hit it off with the judge and he was chosen “The Most Human human” for 2009. For him it was an unusual victory…
Christian: It’s a fascinating and strange honor to be dubbed “The Most Human Human” of a particular year. You know I was certainly pleased that you know I thought I’d taken the challenge pretty seriously, I’d prepared pretty seriously and so you know I was happy to feel that that had paid off in some sense. But at the same time, of course, you know I don’t, I don’t take the award too seriously to mean anything in terms of me actually being any more human than anyone else.
Peterson: You can read all about Brian Christian’s road to winning, and read about his and other’s conversations in competition, in his book, “The Most Human Human” available at bookstores and online. You can also visit his website at Brian-Christian.com. For more information about all of our guests, log onto our site at Viewpointsonline.net. Our show is written and produced by Pat Reuter. Our production directors are Sean Waldron, Nick Hofstra and Reed Pence. I’m Marty Peterson.