April 15th, 2010
As the world Turings…
For two years I’ve flirtatiously circled the Pygmalion myth, toying with human-machine interactions in which it’s not necessarily clear to the human that s/he’s interacting with a machine or human-human interactions in which both participants are convinced that the other is a machine. I can’t seem to get away from this idea of tricking people into adopting mistaken mental models of interactions. I thought it would be fun to create two bots that would follow each other on Twitter. Caleb Larsen, whom I’ve written about before and with whom I’m beginning to believe I share an eerie and otherworldly mental connection (I found this today, compare to my Obama piece) created a script that updated his Facebook and tweeted randomly generated status messages as part of Whose Life is it Anyway, though in the end he abandoned algorithmically generated messages for appropriation of other people’s statuses—which I find conceptually stronger but no longer relevant to the topic at hand.
In any case, in 1950, Alan Turing wrote a paper about thinking machines. In it he proposed a thought experiment in which a person is asked to converse via teletype with a person and computer pretending to be a person. If s/he is unable to definitively distinguish between the two, goes his argument, the computer is effectively intelligent. People have taken issue with Turing’s conception of intelligence, but nonetheless, over the years, this “Turing test” has spawned doctoral dissertations, colloquia, academic prizes, late-night geek-outs, and many software implementations of computerized interlocutors or “chatterbots.” The first of note was Joseph Weizenbaum’s ELIZA, a Rogerian psychotherapist (you can still talk to her here). She was followed by PARRY, a paranoid schizophrenic. A match made in heaven, I know, but their conversations weren’t nearly as interesting as this exchange between more sophisticated later chatterbots (in this case ALICE and Jabberwacky). Awesome.
There have been a couple of really good ITP projects that riff on what I might call the nebulous interlocutor. Generative Social Networking is my absolute favorite ITP project—in conception, in execution, even in documentation. After using a Bluetooth exploit to download all the contacts on your cellphone, a program calls each number in succession, playing a recording of the last person it called as the other half of the conversation. The most amazing thing when you listen to the demo is the realization that most people have no idea they’re talking to a recording! And some of the “conversations” that develop would easily fool a casual observer too. The ritualized form of phone conversation combined with the latencies and poor connections to which frequent cell phone use have accustomed us make it really hard to tell the difference.
That was part of what made the Popularity Dialer so much fun (and ultimately led to its demise—though creators JennyLC and Cory referred to it last week as “dormant” rather than dead). The premise popular people get lots of phone calls, so what better way to enhance your popularity than by increasing the number of calls you receive? Enter your phone number on a website and schedule a call from one of five characters who think you’re awesome (girl dying to date you being my favorite). At the appointed time, your phone rings and the voice you’ve selected speaks its half of a recorded phone conversation, pausing several beats for you to respond. It seems totally real to onlookers. The problem? It seems totally real to many of the people receiving the calls after their friends entered their number as a prank. Worked great until a humorless FCC lawyer got a call late one night from dude wanting to get some beers.
I love both of these projects. They raise questions about the subjective nature of interaction that don’t get discussed all that much in the literature. So much of an interaction is in our heads. That’s the great lesson of Apple’s marketing—you can take a shitty phone that’s uncomfortable to hold and inconvenient to talk on, but if people are emotionally attached to it, they’ll find using it a pleasure anyway (I think Donald Norman might have said something similar a little more eloquently). In our case, if I think I’m talking to a real person, my experience of that conversation will be radically different from my experience of the exact same conversation if I know I’m talking to a recording—just think of that weird, disjointed feeling you get when a friend’s answering machine picks up and you think it’s him and start talking only to realize a second later that it’s a recording. Richard Powers’s Galatea 2.2 deals with this notion of artificial intelligence as deception, and I want to as well.
I propose a film of a woman with her back to the viewer. She is obviously concentrating hard, occasionally tapping a pencil or reaching for her coffee mug but otherwise moving very little. A phone number is displayed beneath the frame. The viewer calls the number and suddenly the phone on the desk next to the woman rings. She picks it up, and the viewer is amazed to hear her voice both on the screen and through his phone. He speaks to her. She responds that the connection is not clear, she can’t hear him well. He tries to gauge whether she is a real video or a clever program. She hangs up in anger and frustration. She looks at her phone and decides to call back. The viewer’s phone rings and when he picks up, she apologizes for the poor connection and asks him a question. When he answers, she asks another. Suddenly, she has to go. She apologizes, turns toward the screen, waves, and hangs up. The viewer scratches his head and calls back. Her phone rings, she looks at the number and sends the caller directly to voicemail with an over-the-shoulder wag of the finger. And scene!
I’ve seen a couple implementations of phone-enabled interactive movies, but they’re infantile choose-your-own-adventure narratives constructed like corporate phone trees (“if you’d like to see the hero die, please press pound now, otherwise, stay on the line for more options”). I want the interaction to be the purpose of the piece, not a means of advancing a canned story, though I do love the bizarro preview man voiceover in this German interactive “horrah” film:
My system works in a similar way, though without all the voice recognition. I’m interested in exploring how much of such an interaction is actually reactive. In Japan, for instance, it’s definitely over 50%, but I’m working on the assumption that it will be similar for the viewer speaking on the phone, that the character in my movie won’t need to respond directly to the viewer’s words because the social inertia that carries people through uncomfortable party conversations with socially maladroit companions will cause him to behave a certain way in this particular interaction—enough that I’ll be able to maintain some doubt as to whether they’re actually participating in a real conversation. Based on several recent interactions with customer service representatives over the phone, I can’t swear that health insurance companies haven’t already commercialized and adopted this system.