So-called chat bots are all over the internet – usually trying to sell us something – and it’s becoming harder and harder to distinguish actual humans from the sophisticated software designed to emulate humans. Technology writer Clive Thompson, who blogs at collisiondetection.net, encountered exactly that dilemma recently when he got a message from someone with the handle babygurl01475. Thompson explains what happened when he engaged.
BOB GARFIELD: Last month we talked on this show about Twitter bots, accounts on Twitter whose tweets are generated not by people, per se, but by a program.
Of course, so-called chat bots are all over the Internet, usually trying to sell us something, and it’s becoming harder and harder to distinguish actual humans from the sophisticated software designed to emulate humans.
Longtime technology writer Clive Thompson, who blogs at collisiondetection.net, encountered exactly that dilemma recently when he got a message from someone with the handle, babygurl01475.
CLIVE THOMPSON: The opening salvo was basically just, you know, hey, what’s up? And I had actually seen babygurl etc., etc., trying to engage me in chat several days before that, and I'd always just declined, declined, declined. So finally I said, okay, I'm game, and I said, oh, God, what is it? And they were like, you know, oh, you know, I'm just bored and lookin' for someone to chat with, which I recognize as the universal opening salvo of someone who’s trying to get you to go to some, you know, webcam chat sex site and then give up your credit card.
I got interested as to whether or not this was actually a human or a chat bot because, you know, it wasn't entirely clear.
BOB GARFIELD: Now, this sounds already like something called the Turing Test, right, which is for people to, in fact, try to detect whether they're speaking to software or speaking to a live human being, and it’s an ongoing measure of how sophisticated semantic Web software is, right?
CLIVE THOMPSON: Yeah, that's right. Back around the Second World War, Alan Turing, who was a famous mathematician who helped break the German codes, allowing the Allies to understand the German messages – pioneering computer scientist – he wrote an essay that basically said, you know, is it possible that a computer could be thinking?
And he said, let's take off the shelf everything except for like just conversation. Let's say that that’s, you know, the marker of whether or not you’re really intelligent and whether you’re thinking is whether you can appear to be a convincing human in conversation. And he was talking about like talking on a teletype machine, which is kind of like the chat of the 1940s and '50s, right?
And he said, let's imagine that you’re sitting in a room and you’re talking on a teletype to two entities and you have to try and figure out which one’s the human and which one’s the machine. And if you can't figure it out, then the machine is effectively human. You can say that it’s sort of thinking and it’s passing the Turing Test.
BOB GARFIELD: So you decided [LAUGHS] to go all Turing Test on whoever was sending you the messages. How did you go about that?
CLIVE THOMPSON: Some of these bots are very sophisticated now. People have been programming these bots for like 10 or 15 years, and they'll have tens of thousands of responses that can seem quite convincing.
One of the things that the bots are really bad at doing, though, is they won't be able to talk about something that recently happened, right? So I said, what did you think of Bridesmaids, the movie that just came out? And the bot really just dodged the question, gave me a sort of a canned response. I was pretty sure after that question I was talking to a chat bot.
BOB GARFIELD: Now, if the chat bot had responded, oh, super gross but super funny, then you realize you have been spending time talking to an actual person called babygurl [LAUGHS], then what?
CLIVE THOMPSON: The conversation would have gone in a very different direction [LAUGHS], believe me.
BOB GARFIELD: Yeah, okay, but one of the things about engaging online is just because an answer is canned and stiff doesn't mean that it’s necessarily coming from a computer. It could be just someone working from a script, right?
CLIVE THOMPSON: Yeah, that's right. So basically the conversation was, what did you think of Bridesmaids? And the response was, oh, well, I'm still in college right now and I've been working here online lately trying out something new – wanna check it out? – which, a) was not an [LAUGHS] answer to my question, what do you think of Bridesmaids, and b) you know, a transparent gambit to try to get me to probably a sex website, at which point in time I openly began discussing with the bot something I was wondering, which was this.
When you talk to chat bots you’re getting these canned responses, but sometimes when you talk to humans they're forced by their employer to only say something in the script, right?
We've all had this situation. You call a help desk, you’re asking a question, and you can hear them clicking the keyboards and they're reading something. Effectively, the human is behaving like a chat bot, right?
And so I said, you know, I can't figure out whether you’re a chat bot or a human who’s working with a preset script that’s sufficiently rigid that you appear to be a chat bot.
And the fascinating thing was at that point in time, babygurl said to me, no, I'm not a bot, are you? Whoever or whatever was on the other end of this conversation had thrown the Turing Test back in my face.
I got very philosophical, so I'm like, well, how would I know whether or not it was a chat bot? You know, this is almost like [BOB LAUGHS] a Bladerunner moment, you know. Is, is Deckard a, a replicant or is he not? And so I said, you know, that’s an interesting question. How would you be able to tell whether I was one or not? I got a response that didn't follow up on that at all and literally was just, they gave me a link and said, go click on the “invite” tab. And, you know, there, see, that’s clearly a chat bot that has been programmed to respond to the question “Are you a chat bot?” with, “No, I'm not a bot, are you?”
BOB GARFIELD: Right. So how was the website?
CLIVE THOMPSON: No way I was clicking on that link, my friend. [BOB LAUGHS] That was probably just filled with spyware and spamware and people who'd just show up and steal my toaster.
BOB GARFIELD: Okay, at this stage you’re certain, or almost certain, that you’re speaking to software. How does it end?
CLIVE THOMPSON: Well, I kept on saying things, making these philosophical points with the Turing Test, and finally [LAUGHING] I realized it’s gone away, you know. So my final statement was, I, I appear to have bored you into silence. And the commenters on my blog all said, you know, that was the most poignant moment, when I chased a bot away by boring it.
BOB GARFIELD: [LAUGHS] And actually, if you kept at it, on the verge of being a stalker, Clive.
CLIVE THOMPSON: Yeah, yeah, sure. Maybe the bot’s talking to the other bots, saying, don't go near that guy, you know. He’s weird. He'll just talk for hours -
BOB GARFIELD: Yeah.
CLIVE THOMPSON: - about philosophy. You'll never get a credit card out of him.
BOB GARFIELD: Next time we talk, we can talk about the bot restraining order that’s been issued. Clive, as always, a delight to talk to you. I'm reasonably sure you’re human.
CLIVE THOMPSON: Thank you, and I'm reasonably sure I am, too.
[MUSIC UP AND UNDER]
BOB GARFIELD: [LAUGHS] Clive Thompson writes about technology for The New York Times Magazine, Wired Magazine, and on his blog collisiondetection.net.
WNYC 93.9 FM and AM 820 are New York's flagship public radio
stations, broadcasting the finest programs from NPR, PRI and American Public Media, as well as a wide range of award-winning local
programming. WNYC is a division of
New York Public Radio.