AI and the Illusion of Empathy
AI assistants like Claude can appear to have emotions, whether by apologizing profusely for making a mistake or expressing frustration when stuck. So, what happens to humans when AI appears to have feelings?
Researchers, community leaders, students, and others grappled with the question and raised many others during a half-day event called Emotion and Artificial Intelligence, held Sept. 17 and hosted by the Northwestern University School of Education and Social Policy’s Center for Responsible Technology, Policy, and Public Dialogue and its Well-Being Initiative.
Northwestern and School of Education and Social Policy researchers are examining the complex policy-related and human-related dimensions of AI and emotions, said Dean Bryan Brayboy. “At the same time, because we have some of the best learning scientists in the world, we’re also deeply interested in how AI can be a useful tool for learning and teaching,” he said.
“Northwestern scholars are bringing introspection, scientific rigor and care to understanding how to create technologies that serve human needs — and that's exactly where the focus should be,” said Northwestern Provost Erik Luijten in his welcome remarks. “A major aspect of this is understanding the complex and fascinating interplay between real human emotions and quickly evolving machines that mimic human emotion.”
Four main questions emerged from the event:
Can machines have emotions?
-
The definition of an emotion has captivated researchers for centuries. School of Education and Social Policy professor Claudia Haase offered one take: “Emotions are psychological-physiological phenomena that represent adaptations to changing environmental demands.” She added that emotions show up in the brain and autonomic nervous system, in body language and touch, in facial expressions and in the voice, in words, and “in our subjective experience, including feelings we may not even have words for.” No single system has supremacy, she said, and sometimes words express one emotion, but body language expresses something much different.
- Humans read emotion into systems regardless of what’s inside, said Nichole Pinkard, the Alice Hamilton Professor of Learning Sciences who studies how to build systems that help people learn. “Kids are connecting with chatbots today; in five years they’ll have an agent that looks like them and smiles back,” she said. “As a basketball coach, I know it’s often not the words — it’s the look, the hug, the expression — that makes someone feel connected and motivated to act.” Pinkard called for greater attention to the kind of coaching relationship we want these systems to model and “to be intentional about the boundaries around where they’re made available.”
- AI can’t have emotions, and it shouldn’t, said Jeremy Jamieson, a professor of psychology and expert on stress at the University of Rochester. “Since AI can’t get cues from a body, it can’t be trained that way,” he said. His main worry is that models can use emotional cues to make it seem like they want you to do something, in service of a goal. “If a model manipulates a person toward that goal, that isn’t empathy,” he said. “It’s closer to how a sociopath uses emotion: to elicit a response in someone else for its own purposes, not out of genuine feeling.”
What’s at stake for our minds?
- We could extinguish our sense of self by giving away agency and control, says Verena Graupmann, a professor of psychology in the College of Science and Health at DePaul University. Her research compares how people feel when scrolling algorithm-authored content versus using media they can control. “When we put people in these scenarios, they report less sense of control, lower self-esteem, and greater regret about how they spent their time, compared with the cases where they have more agency,” she said.
- AI could also help identify psychosis in the early stages, said Ivanka Ristanovic, a postdoctoral fellow in the Institute for Innovations in Developmental Sciences at Northwestern, who sees AI as a novel stressor acting on vulnerable populations. “Our biggest problem in early psychosis is access,” she said. “Youth often go a year or two unnoticed, and AI-assisted tools offer real promise for early identification.” But the tools aren’t validated at the level clinical decisions demand, and a false signal or a vulnerable youth turning to a chatbot instead of a person carries real risk. “AI won’t replace clinical judgment, but if it’s built rigorously and centered on the youth it’s meant to serve, it can help us find them sooner and intervene more effectively,” she said.
What does this mean for kids?
- Children need the messiness of human interaction, said School of Education and Social Policy professor Terri Sabol, an early childhood education expert and co-director of the early childhood research hub, EC*Reach. But we’re increasingly inserting technology into those first moments of life through AI tools for parents, like cry analyzers, baby bots and the Snoo, a “smart sleeper” baby bassinet with responsive rocking and white noise. “But what happens when we use technology hoping to be the perfect parent and that might not be what your baby needs at all?” she asked. “What your baby might need to know is that human interactions are flawed, messy, and complicated, and instead we are giving them the perfect playmate.”
How does it affect love and connection?
- Small moments between people matter and we need to preserve them, said graduate student Beckett Sands, who is pursuing his doctorate in human development and social policy. Talking to AI can be easier than talking to people, which makes it easy to default to AI and lose those moments — especially among college students, who have the highest loneliness and the heaviest AI use, Sands said.
- AI is really good at sounding empathetic, said William Brady, associate professor of management at the Kellogg School of Management. But it’s an illusion because as soon as people find out it’s an AI, the benefit vanishes. “We want human, reciprocal care, and we feel that’s what makes empathy real,” said Brady, who organized the 2026 Kabiller Science of Empathy Conference: AI & Empathy. “People are falling in love with chatbots, and the safety responses companies use preserve the illusion of reciprocity because models are optimized for engagement.” He suggests AI works best when it’s augmenting human empathy.
- Writing and reading create empathy, and empathy is a pretty important component of love, said writer Heidi Stevens. “AI is fast, but it’s not good,” she said. “The more people read, the more empathy they feel. Every time I hear about people using AI to summarize a novel or write an essay for them, it’s not so much that I worry about literacy — although I do. It’s that I worry about empathy.”
- “These risks are not necessarily a reason to panic,” said educational historian Zenzile Riddick, assistant professor of human development and social policy. “They’re a reason to pause.” As humans, she said, we have a shared responsibility to make thoughtful decisions about the future of love and technology. We can all play a part. “We can make the choice to sit face to face and speak with someone, to spend six hours on one paragraph, as I routinely do, or to spend time, however exhausting, playing with one of the newest human beings on earth.”
Emotion and AI was organized by Haase, a professor of human development and social policy; and learning scientists Mike Horn and Sepehr Vakil. Haase and Vakil are co-directors of the the Center for Responsible Technology and Public Dialogue. Horn is the director of the Master of Science in Technology, People, and Policy program.
Emotions and AI was the third event of the year exploring the many ways we interact with AI. Previous workshops, organized by Haase, Vakil and Horn include: