We’re back with Joe Dillon, the social psychologist, who is introducing you to one of his graduate students. The young woman – who doesn’t, in fact, look much younger than her academic supervisor – is called Leah Hart, and the fall of her neat black hair frames a round, earnest face located at a level with Dillon’s bicep.
“Leah is working on the cues by which people identify Gubots and distinguish them from Gupes,” says Dillon. “She reckons people are developing greater ability and better strategies to do so as the pseudo-AI gets more sophisticated.”
“That’s right,” says Hart. “I’ve compiled evidence from a number of sources – real-time brain tracking, experiments I’ve done, surveys – and I believe I can show that human discernment of non-human, but humanlike, actors is increasing in power.”
“That sounds like a thesis title,” you say.
She smiles shyly, dipping her head. “Almost. Of course, it’s a kind of arms race, because when researchers in my field find ways in which people distinguish between fellow humans and Gubots, the Gubot researchers try to find ways to overcome that, to make the illusion more convincing. Eventually they’ll win.”
“You see it as a struggle?”
“For some people it’s a struggle. They want to be able to choose their responses differently based on whether they’re interacting with a real human or an imitation one. Some people don’t care. Some people pretend they don’t care, but when you observe them they actually do behave differently when they believe they’re interacting with a Gubot. And some people think they can’t tell, but again, when you observe them, they have a subtly different response – and they’re right more often than they’re wrong, in many cases.”
“Can you give me an example of the differences?”
“Well, most of them are matters of degree rather than kind, but if someone thinks they’re talking to a Gubot, their syntax tends to simplify. They don’t make as much eye contact, and sometimes they’ll catch themselves making eye contact and then pull away. They instruct more and collaborate less.”
“How can you tell that apart from arrogant people with poor social skills?”
Hart laughs. “Sometimes, we can’t. We’ve developed some informal categories, actually, for different kinds of interactors. There are the ‘uncaring interactors’, who fall into two categories: those who treat anything they interact with as if it was a human being with feelings, and those who treat nothing that way. They don’t distinguish, in other words. Then there are the ‘subtle discriminators’, who will act one way when they believe they have a Gubot and another way when they believe they have a human, but the difference isn’t highly marked, you need to be observing closely to see it. Perhaps the largest group, though, is the ‘differentiators’, who behave distinctly differently if they think they’re talking to a human. If a differentiator was talking to you and thought you were a bot, you’d probably get quite offended with them.”
“And how quickly do people make up their minds?”
“That also varies. Some people make up their minds very quickly and ignore any cues that they’re wrong, which they quite often are. Others go back and forth for an entire conversation. But usually what happens is that people will begin by observing until they spot something which they consider a ‘tell’, a unique marker of language or behavior that tells them, ‘Houston, we have a Gubot.'” She blushes and laughs nervously at her own joke. “Then their behavior changes. Or, of course, if they spot a ‘tell’ that they’re confident marks a human.”
“Can we see any examples?”
“Sure, here’s a nice straightforward one.”
A young man, his features digitally anonymized, is talking to – something made out of Gu and shaped like a young woman, which may or may not be controlled by one. We enter partway through the conversation. He is speaking.
“So, what did you think of Bowline’s speech last night? Did you catch that?” He looks at her intently for her reaction.
“Yes, I did. I don’t usually follow politics but something that important… I thought he had some good points, actually.”
“Like what?”
“Well, people are getting complacent. I know my parents… they used to be activists, but they’ve become comfortable, you know? It’s all too easy. I hope I never get that way.”
The man’s body language changes. He leans forward, touches her lightly on the arm and makes eye contact. “Yeah, I know. Mine too. That’s… so what are you doing? About Bowman’s thing?”
“Well, that’s what I don’t know. Is there, like, a group we can join on campus or anything? To like raise awareness?”
“Well, let’s check the campus groups directory.” He pulls up a projection and moves round so he’s sitting next to her and they can both see it.
“Oh yeah, duh, I should have thought of that.” She laughs.
“Takes some getting used to,” he says, and smiles at her, making eye contact. “Now, what have we got…?”
“Now that,” says Hart, “is a beautiful example of the moment when someone decides to believe that he’s talking to a human.”
“And is he?”
“Funnily enough, he is, despite the fact that the conversation itself could quite easily have been AI fabricated. What’s a lot harder to fabricate is the body language, and hers is perfect, it completely matches the tone and content of the conversation and responds to his very precisely. What he hasn’t picked up, and never does, is that the person on the other end is in fact male, and that’s the other dimension to this kind of interaction – people who are representing themselves truthfully as human, but untruthfully when it comes to other matters.”
“And are there any people who pretend to be Gubots?”
“There are, and it’s almost a performance art. I have a competition going: if you can fool me for 20 minutes that you’re a Gubot when you’re not, I buy you a beer. If your Gubot can fool me for 5 minutes that it’s human – I buy you a keg. It’s a lot easier one way than the other.”
“And have you had to buy much beer?”
“Not for a long time. I can spot the difference very reliably these days.”
“Can we try something? I’d like to see if I can spot the difference. Let’s have two figures, one controlled by you and one by a Gubot, and see if I can tell which is which.”
“All right, a classic Turing test. You’re on.”
“And if I get it wrong, I’ll buy the beer.”
“In that case you’re definitely on.”
Cut to the test. Two Brandy figures are sitting in chairs facing you, identical except that they have large red numbers on their fronts designating them as “1” and “2”.
“All right, first question,” you say. “Number 1. Is a skylark green?”
Number 1 blinks as if startled. “Um – I don’t know.”
“Number 2 then. How often each day do you clean your teeth?”
“Twice,” says 2. “Morning and night.”
“Not at lunchtime? Why not?”
“I don’t carry a toothbrush with me.”
“Number 1, if someone asked you if you had the time, what would you say to them?”
“12:37.” The answer comes without hesitation, but it’s accompanied by a glance up and left. Number 2 uncrosses and recrosses her legs.
“OK. Number 2, which is heavier, a kilo of lead or a kilo of feathers, and why?”
“A kilo of anything is always the same weight.”
“And number 1, which is more important, freedom or security?”
“I’d have to go with freedom.”
“Because?”
“Security will never be perfect. Also, if you sacrifice freedom for security you don’t deserve either.”
“Who said that?”
“Uh, Benjamin Franklin, I think.”
“And Benjamin Franklin invented what?”
“I don’t know. I know he was an inventor but I can’t remember what he invented.”
“Do you know, 2?”
“Bifocal spectacles. And a kind of stove.”
“Did you know that or did you look it up?”
“I knew it.”
“Do you remember where you learned it?”
“I think I must have learned it in grade school.”
“And where did you go to school?”
“Portland, Oregon.”
“The name of your first teacher?”
A pause. “Mrs… Cunningham.”
“OK, 1, how many roads must a man walk down? To the nearest approximation.”
“That’s a rhetorical question – there isn’t a numerical answer to that.”
“Can you give me an example of a similar question?”
“Umm… How much wood could a woodchuck chuck, if a woodchuck could chuck wood?”
“2, if you came into a room and saw your boyfriend kissing another girl, how would you feel?”
“I’d be very surprised.”
“Why is that?”
“Because I don’t have a boyfriend.”
“What about you, 1?”
“I’d feel angry and betrayed.”
“Do you have a boyfriend?”
“No, but if I did, that’s how I’d feel.”
“OK, one last question, and either of you can answer. What’s wrong with this sentence? ‘Is you is, or is you ain’t my baby?'”
The two figures think for a moment, then 1 speaks. “It’s not conventional English grammar. It’s not even the usual way that ‘ain’t’ is used in dialect.”
You look at 2, who says, “What she said.”
“All right,” you say. “The test is at an end. Would the real human being please stand up?”
1 starts to stand, but 2 says, “Wait a minute. You haven’t said which one you think is real.”
“You are,” you say.
“Yes, you’re right,” says 2. “How did you know?”
“I didn’t until that last moment,” you say. “The Gubot doesn’t care about the bet; it’s just programmed to pretend to be human as best it can. You weren’t doing as good a job, and that made me suspicious, but I was only certain once 1 stood up.”
“Actually,” says 1, “we rigged the test. I’m controlling both of them. Neither one was a Gubot.”
You stare for a moment, then crack up laughing.
Mike Reeves-McMillan lives in Auckland, New Zealand, the setting of his Auckland Allies contemporary urban fantasy series; and also in his head, where the weather is more reliable, and there are a lot more wizards. He also writes the Gryphon Clerks series (steampunk/magepunk), the Hand of the Trickster series (sword-and-sorcery heist capers), and short stories which have appeared in venues such as Compelling Science Fiction and Cosmic Roots and Eldritch Shores.
OK, I cheated. Thing is, I kept changing my mind myself throughout writing the dialogue. Is 2 the human who’s pretending to be an AI pretending to be human, or is it 1?So I finally decided it was both of them.Is that too much of a cop-out, do you think?