When I was a kid I read everything I could find about science fiction, robots especially. Isaac Asimov was my favorite, with his Three Laws of Robotics. To me robots had personalities, different shapes, different voices, different behaviors. They had emotions. It looked like they had moods.
Years later, as an adult, I understand that's impossible. And I still catch myself reacting like it isn't.
Ask most people whether a robot can feel anything and you get a debate about the far future. Sentience, consciousness, some threshold nobody's built yet. Wrong question, and it's already been answered. Rosalind Picard, who founded the Affective Computing group at MIT Media Lab specifically to study this, has a simple answer for the robots we actually have. They don't feel anything. They can look like they do. Aibo, the robot dog Sony built in 2001, wags its tail when you walk in and droops when you scold it. "The dog does not have any conscious experience or feelings that go with being happy or sad," Picard says. It's matching an input to an output, nothing more.
Fine. Settled. And it changes almost nothing about how I, or anyone else, actually behaves around one.
Kate Darling, also at MIT, has spent her career on the part that's actually interesting, which is us. Her research keeps landing on the same finding. People anthropomorphize robots reliably and immediately, whether the robot earns it or not. Give something a name, a story, or a flinch, and people start protecting it, apologizing to it, feeling responsible for it. It's not a glitch in a few overly sentimental people. Darling's work has gone as far as asking whether that reaction is strong enough to warrant legal protection for robots, the same way animal cruelty laws exist not because the animal files a complaint, but because of what mistreating it does to the person doing it.
Expectation turns out to matter as much as the reaction itself. A study in Frontiers in Robotics and AI interviewed workers across hospitality, manufacturing, and care roles and found they weren't asking for warmth exactly. They wanted robots with basic, predictable social skills and clearly limited autonomy. What actually unsettled them was a robot behaving either colder than expected or more independent than they'd been told to expect. The reaction wasn't about what the robot could do. It was about whether the robot did what they'd been braced for.
A Frontiers editorial this year reviewed seven HRI studies and named the mechanism. Robots don't transmit emotion. People construct it, filling the gaps with their own expectations. A study out of Japan showed this directly. Researchers gave an android named Nikola 63 facial expressions and tested it on more than 600 people. Vaguer instructions about what to expect made people rate it as more emotionally expressive, not less. Less information, more feeling, invented on the spot.
Which brings me back to being a kid with an Asimov paperback, deciding a fictional robot was moody because of how it was drawn. Nothing about growing up and learning the science switched that reflex off. It just gave it better manners.
That reflex is the real terrain for anyone managing people near robots. What the robot can do matters less than what your people expect it to do. Get careless with that expectation and you'll get exactly what Darling and the Frontiers researchers keep finding: real feeling, real consequences, none of it started by the robot.
Source: Murphy, M. (2017, updated 2026). Can we make robots that experience emotions? MIT School of Engineering, Ask an Engineer (Rosalind Picard). Read the article →