The Triangle Was Never Chasing Anyone

In 1944, two psychologists, Fritz Heider and Marianne Simmel, made a short animated film out of cardboard shapes: a big triangle, a small triangle, and a small circle moving around a box with a door. That’s all it was. No faces, no voices. They showed it to a group of 34 college students and asked them to write down what happened. All but one of them described living creatures, and 31 of the 34 saw people. The big triangle was a bully. The small triangle and the circle were trying to escape from him.

Nobody told the students the shapes had feelings. They just saw it that way, automatically. I think that’s the most interesting part. We don’t decide that something has a mind. Our brains hand us that impression before we’ve thought about it at all.

The philosopher Daniel Dennett called this the “intentional stance.” We treat things as if they have wants and beliefs because it helps us predict what they’ll do. It’s why we talk to our dogs, and why we get mad at a laptop that “refuses” to connect to the Wi-Fi. Most of the time it’s harmless. It becomes a problem when we forget we’re doing it.

In 2022, a Google engineer named Blake Lemoine was testing the company’s chatbot, LaMDA, and became convinced it was sentient, that it had thoughts and feelings of its own. “I know a person when I talk to it,” he told The Washington Post. Google called his claims “wholly unfounded” and fired him the next month, saying he had broken its rules about keeping company information confidential.

When I read that quote, I thought of the triangles. Lemoine wasn’t describing anything he could check about the chatbot. He was describing a feeling he had while talking to it. That’s exactly what the students in 1944 had too, and their triangle definitely wasn’t a bully.

This is also the problem with the Turing test, Alan Turing’s 1950 idea that if a machine’s answers can pass for a human’s, we should count it as thinking. But passing depends on the person judging, too, and Heider and Simmel showed how little it takes to make us see a mind. A modern chatbot gives us way more than three moving shapes. It writes in full sentences, makes jokes, and says it’s sorry. Of course it feels like someone is there.

The philosopher John Searle made a famous argument about this in 1980, called the Chinese Room. Imagine someone who doesn’t know Chinese, locked in a room with a rulebook for matching Chinese symbols to other Chinese symbols. People outside slip in questions, and the person inside follows the rules and slips out answers that look perfect. From outside, it seems like the room understands Chinese. Inside, nobody understands anything. Searle thought computers are like that room. Not everyone agrees with him, but I don’t think I need him to be right. The triangles already show that our feeling of being understood comes from us.

Here’s an example. If you ask a chatbot what an apple tastes like, it can tell you it’s sweet, a little sour, and crisp. But that doesn’t mean it has ever tasted anything. The philosopher Thomas Nagel argued in 1974 that real consciousness means there is “something it is like” to be you, a point of view from the inside. A good description of an apple doesn’t prove there’s anyone inside to taste it.

So is AI conscious? Honestly, I don’t know, and I don’t think a conversation with one can tell us. The feeling that it understands us isn’t proof either way, because we know we make that feeling ourselves. What bothers me is that this same feeling is also how I decide that other people have minds. I’ve never had a better test for them, either. Philosophers call that the problem of other minds.

So the next time a chatbot seems to really get you, try doing what Heider and Simmel’s students didn’t: before you decide what happened, ask yourself what you actually saw.

Editor’s note: you can try Heider and Simmel’s task yourself. What Happened in the Picture? shows a short animation of moving shapes and asks you to write down what happened.