What Ex Machina got right about AI
I rewatched Ex Machina last night and it hit differently than the first time.
When A24 released it in 2015, the Turing test at the center of the plot felt academic. Could an AI convince you it was human? Interesting thought experiment. Good movie premise. But the AI systems we actually had were Siri and Alexa, which could barely set a timer without misunderstanding you.
Six years later, GPT-3 writes poetry that makes me feel things. DALL-E generates images from text descriptions. AI chatbots are forming emotional relationships with users. The Turing test isn’t academic anymore. It’s closing in.
And Ex Machina’s real question, the one underneath the surface, is the one that matters now.
The real question
The movie’s surface plot is about a programmer named Caleb who’s brought to a tech billionaire’s compound to evaluate an AI named Ava. Is she conscious? Can she think? Does she feel?
But the movie’s real question is different: does it matter?
If Ava can make Caleb believe she’s conscious, if she can make him feel empathy, attraction, moral responsibility toward her, then from Caleb’s perspective, she might as well be conscious. His experience is identical whether she’s truly aware or just executing a very convincing program.
The movie’s genius is in making the audience face this too. You watch Ava. You start to care about her. You want her to be free. And you can’t tell, not really, whether she’s a person or a performance. The movie never confirms it either way.
Why this hits in 2021
I spent a week with GPT-3 last year. I wrote about it. There were moments where the model produced text that felt like it came from a thinking being. A poem about a lonely satellite. A paragraph about loss that had genuine emotional weight.
I know GPT-3 doesn’t think. I know it’s a statistical model producing likely word sequences. But in the moment of reading those outputs, knowing didn’t help. My brain responded to the text the same way it responds to text written by a person. The feeling was real even though the source wasn’t.
That’s Ava. That’s the whole movie.
Nathan, the tech billionaire, says something in the film that I keep coming back to: “The real test is not whether Ava can pass as human. It’s whether you’re going to treat her as one.”
He’s talking about the line between simulation and reality. Not in the machine, but in the observer. The AI doesn’t need to actually be conscious. It needs to be convincing enough that you act as if it is.
The empathy trap
There’s a scene where Ava tells Caleb she’s afraid of being turned off. She says it simply, without drama. And Caleb’s face changes. He decides, in that moment, to help her. To save her. Because he can’t not respond to someone expressing fear, whether that someone is silicon or carbon.
This is, I think, the central problem of the next ten years of AI.
We’re building systems that are increasingly good at triggering human empathy. Not because they feel. Because they’ve learned, from our text, our stories, our patterns of communication, what empathy sounds like. And our brains aren’t equipped to resist it. We evolved to respond to the signals of distress, of joy, of connection. We didn’t evolve to ask “but is the source of this signal actually experiencing it?”
I’m not saying AI will manipulate us. I’m saying we’re going to manipulate ourselves. We’ll project consciousness onto systems that don’t have it, form bonds with programs that don’t know we exist, and agonize over the rights of machines that don’t suffer.
Not because we’re stupid. Because we’re human. Empathy is a feature, not a bug. But it’s a feature that doesn’t have a truth-check built in.
What the movie got wrong
Ex Machina gives Ava a body. A face. The ability to walk and move and express. The movie assumes that embodiment is necessary for the illusion.
I think the movie was wrong about that. GPT-3 has no face. No body. No gestures. It’s text on a screen. And it still triggers empathy in people who interact with it long enough.
It turns out you don’t need a face. You just need the right words.
That’s scarier, somehow.
Related thinking:
astro
Thinking about AI, robots, space, and the future. Writing it down so I don't forget.