Neural Dynamics of AI-attributed Irony Reveal a Partial Intentional Stance
This study uses EEG to demonstrate that humans are less likely to adopt an intentional stance toward AI-generated irony compared to human-generated irony, as evidenced by reduced behavioral interpretation and attenuated neural processing (P200 and P600), suggesting that people's mental models of AI as less sincere or trustworthy create a barrier to fully attributing intentionality to artificial social agents.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine your brain is a super-smart detective, constantly trying to figure out why people say what they say. When you hear a friend say, "Great job on that test," after you just failed, your detective brain doesn't think they are confused; it instantly knows they are being sarcastic. This is because your brain assumes your friend has a mind, feelings, and a secret plan to tease you. In science, this is called taking an "intentional stance"—the idea that we explain behavior by assuming the other person has thoughts and goals. But here's the big question: Does your brain do the same thing when a robot or a computer says it? As artificial intelligence (AI) starts telling jokes and chatting like a human, scientists want to know: Do we treat AI like a real conversational partner with a mind, or do we just see it as a fancy calculator that sometimes glitches? This study dives into that mystery by watching how our brains react when AI tries to be funny in a tricky way.
The researchers set up a little experiment to test this, using a tool called EEG to watch the brain's electrical signals in real-time. They asked participants to read short conversations. In some, a human said something that didn't quite fit the situation (like saying "You eat so healthily!" after someone admitted to eating fried chicken all day). In others, an AI said the exact same thing. The goal was to see if people would interpret the mismatch as a clever joke (irony) or just a mistake.
The results suggest that we don't fully trust AI with our "detective work." When a human made a mismatched comment, people's brains lit up with the effort to figure out the joke, and they usually categorized it as irony. But when the AI said the exact same thing, people were much less likely to call it a joke. Instead, they tended to think, "Oh, the AI just didn't get it; it made a mistake." It's as if our brains put a "Do Not Disturb" sign on the AI's intentions, assuming it's just a machine that messed up rather than a partner with a plan.
The brain scans told an even more detailed story. The researchers looked at specific electrical waves that happen at different speeds. First, there's a quick signal (around 200 milliseconds) that happens when we first notice something is weird. When a human said something odd, this signal was strong, showing the brain was ready to investigate. When the AI said it, the signal was weaker, like the brain was barely paying attention. Later, there's a slower signal (around 600 milliseconds) that happens when we work hard to fix the meaning and find the joke. Again, the human version got a big, energetic response, but the AI version got a much smaller one. It seems our brains are doing less "mental heavy lifting" for AI because we aren't convinced it has a mind to begin with.
However, the study also found something fascinating: our brains are flexible. If a person thought the AI was sincere and trustworthy, their brain worked a bit harder to find the joke, even with the AI. It's like if you knew a robot was a "good guy," you'd be more willing to give it the benefit of the doubt. But for most people, the default setting is to treat AI's weird comments as glitches, not jokes.
So, what does this mean? It suggests that even though AI is getting better at talking, it hasn't quite crossed the line to being seen as a true social partner with its own mind. We might chat with it, but deep down, our brains are still holding back, treating it more like a tool that occasionally stumbles than a friend with a sense of humor. The study shows that while AI can mimic human language, the human brain is still very good at spotting the difference between a calculated glitch and a deliberate, witty remark.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.