Is It a Relationship if No One’s on the Other Side?
Opinion | The Ezra Klein Show
This is an edited transcript of “The Ezra Klein Show.” You can listen to the episode wherever you get your podcasts.
For a long time, the idea that we would have A.I.s or robots that we could talk to, interact with, integrate into our personal lives — that was fantasy, science fiction.
But now, talking to A.I. systems is mainstream. Roughly half of American adults are using chatbots. Many of them are going pretty deep with their A.I.s. Nearly one in five U.S. adults has chatted with an A.I. that simulates a romantic partner.
The results here can get dark. Google’s own researchers have found that some children are now struggling to distinguish between talking with chatbots and interacting with humans.
Since the 1970s, the sociologist Sherry Turkle has been studying the relationships between humans and machines. In the ’80s, she predicted that computers would evolve into these second selves. I think that basically came true — think of the smartphone in your pocket that you’re always looking at.
In 2011, just a year after Instagram came out, Turkle warned that social media might create a society where, even though we’re constantly talking to one another, we’re in this strange state of being alone together. I think the data bore that one out, too.
Turkle’s latest book, “Artificial Intimacy: Who We Become When We Talk to Machines,” turns this lens on A.I. So I wanted to have her on the show to help me figure out what’s going on now that talking machines have very much arrived, what it means that so many of us are building these — I don’t know, call it relationships; don’t call it relationships — with them, and what might happen if we keep going down this path.
(The New York Times has sued OpenAI and Microsoft claiming copyright infringement. The companies have denied those claims.)
Ezra Klein: Sherry Turkle, welcome to the show.
Sherry Turkle: Delighted to be here.
So you studied a very early, very famous chatbot named Eliza.
Yes.
Tell me about Eliza.
Well, I got into Eliza — I had just arrived at M.I.T. and I was introduced to Joseph Weizenbaum, who said he wanted to see me. I was one of the few people psychologists at M.I.T.
And he came to see me and he explained that he had written a chatbot — a stupid chatbot.
This is the 1960s, right?
No, this was 1976.
OK.
He had written it earlier. He had written it in the 1960s, but it wasn’t really until — it was also known as the doctor program because he had written it, because it was very stupid, to do something that could simulate a conversation that people would believe was a real conversation. And that was the conversation between a patient and the kind of psychotherapist that’s called a Rogerian psychotherapist.
And the reason that was a good thing for him to do with a stupid program is that a Rogerian therapist repeats what you say in a spirit of affirmation.
People should know, there are podcasters who — this is primarily what they do. [Laughs.]
Exactly. Exactly. [Laughs.] So basically, if you say to a Rogerian psychotherapist, “I’m so angry at my mother,” it’ll say, “I hear you saying you’re angry at your mother.”
So Weizenbaum was able to create a program — and again, it’s a chatbot only because it was a program that chatted, but it wasn’t a chatbot in the way we think of it now, with generative A.I. behind it, processing millions and zillions of data points. It took what you said and essentially put something like “I hear you saying that” in front of it, or “Are you saying that?” Or something really basic and very stupid.
And he did it to experiment with language processing, and he was a computer scientist and that was his goal. And he found that the graduate students who had helped him, who knew that this program didn’t know and didn’t understand anything, nevertheless wanted to be alone with it and wanted to be in a closed room with it and wanted to talk to it. His assistant — his secretary, as he called her at the time — wanted to be alone with it, wanted to talk to it.
And he was delighted to find me, a psychoanalytically trained psychologist, to try to help him understand what that phenomenon was. And this was really one of the things that drew me into the study of computers and people, because it was really one of the first times that you really saw the seductive power of a machine that talked.
So Eliza’s famous because, to many people, it was one of the first of these programs to arguably pass the Turing test — this idea of: Can you talk to a program and you’re not necessarily sure if you’re talking to a computer or to a person?
As you say, people knew what Eliza was. So what was so effective about it?
But what I called the “Eliza Effect” was people’s desire to make it seem smarter than it was. They wanted so much to believe that it was a thinking creature that they would ask it questions in a way that, with practice, they knew would cause it to give the most humanlike response.
So for example, hackers would figure out that if you said three angry things in a row, the machine would say to you, “I hear your anger; are you angry at your mother?” And it would seem like that came out of the blue. In other words, even though that was scripted, if the following angry words are said three times, it would respond, “I hear anger in your response.”
And again, that was one of the early things I learned, that it’s not just that we come to a machine and say: “Oh, is this intelligent? Does this understand me?” But we come to these conversations wanting it to be there for us. And that really concerned Weizenbaum because he thought that was a very dangerous road, because the machine really understood nothing.
Comments