A recent study by scientists from The Pennsylvania State University (Penn State) and the University of Toronto, published in the journal Communications Psychology, has revealed a strange paradox between artificial intelligence and human empathy.
Experiments conducted with over 600 respondents showed that, alongside the level of empathy, the “authenticity” of its source is critically important to humans. Although people often rate ChatGPT’s responses as more considerate, when given a choice, they still prefer to receive emotional support from another human being.
The research highlights that, thanks to training on billions of texts, AI perfectly balances conversational tone and provides users with a sense of emotional validation—something it often does even better than professional medical practitioners.
Experts believe that AI will not replace psychologists; however, scientists see great potential in it as a supporting tool. We spoke with the study’s co-author, Joshua D. Wenger, a doctoral candidate at The Pennsylvania State University (Penn State), about these complex connections and the human motivation for seeking empathy. He helped us clarify why people feel more empathy from AI and what risks this process entails in the long term.
An Interview with Joshua D. Wenger
Why do you think people still turn to humans for empathy even when AI responses feel more attentive or understanding?
It’s difficult to say exactly why this might be. People have different possible motivations behind seeking empathy besides perceived empathic quality (e.g., feeling connected with the empathizer) and it’s possible that AI does not satisfy these motivations in the same way that human empathizers might. There is also prior work suggesting that people may have a general aversion to AI systems which could certainly play a role here. Additionally, some research suggests that people tend to moralize AI, so it’s possible that this may lead some to view it as wrong to turn to an artificial system for such a deeply human quality such as empathy.
What does it reveal about human emotion that someone can feel genuinely heard by a machine, even knowing it has no feelings?
I think it really highlights the importance of perception from the empathy recipient’s perspective. Even if AI can’t truly empathize and feel what we feel, sometimes the perception of it doing so is enough to have a real emotional impact.
It may also feel shocking or counterintuitive at first that AI is capable of making people feel heard when it obviously has no feelings itself, but I think this sort of phenomenon plays out in other interactions too. For example, there are plenty of interactions which may require empathy from us at times where we feel like we have little energy to give this kind of support. In these situations, it may be a valuable skill to convey (or more negatively put, “fake”) empathy, making someone feel heard, even when we aren’t presently capable of truly engaging emotionally.
Do you think the paradox between AI’s “perceived empathy” and actual human preference tells us something about our need for connection versus validation?
This really touches on a core motivating question for our team’s approach to empathy research: How might different motivations shape empathy-seeking behaviors? AI might be a sufficient—or even ideal—source of emotional support if all we desire is to feel validated (though there are some concerns of sycophancy such that AI is inappropriately validating); however, AI may fail to satisfy a deeper motivation for human connection. There are multiple reasons people may want empathy—some instrumental and purely about feeling comforted, and others about building relationships. Different people want different things out of empathy in different situations.
Are there patterns in which emotional situations make people more open to AI empathy, and why might those differ from human interactions?
This is a question that we’re currently exploring. We have a supplemental finding from our recent paper suggesting that the type of suffering a person is seeking empathy for may be relevant. There, we found that the human preference was even stronger for emotional suffering (e.g., a rejected proposal) than for physical suffering (e.g., breaking a bone). We’re currently exploring how the personal sensitivity/embarrassing nature of an emotional situation could maybe make someone more or less likely to confide in an AI versus human empathizer.
Could AI’s way of structuring responses help us understand what humans unconsciously look for in feeling understood?
Absolutely. I think some of the most exciting avenues of AI empathy research are those that can teach us about what we want from human empathy too. What is it about AI empathy responses that make them seem so empathetic? What can we learn from these responses to possibly make human empathy even better? I think these questions are intriguing and have enormous future research potential.
How do you see the balance between comfort and authenticity shaping the way people interact with AI emotionally?
I think this connects strongly to some of my previous answers regarding motivations for seeking empathy. In situations where people just want their emotional suffering alleviated, AI may be a great choice; however, if emotional authenticity is important to the empathy-seeker, AI might fall short. Though, I’d also note that actual authenticity and perceived authenticity may be different. AI is incapable of authentically sharing in someone’s experience because it lacks experience altogether, but AI empathy can still feel authentic to some people.
Do you think repeated exposure to AI empathy could change how we define or experience human empathy itself?
Yes absolutely. Many traditional definitions of empathy in psychology research require the empathizer to be capable of emotional experience so that they can feel what the empathy recipient feels. The fact that AI is capable of making people feel empathized with, yet clearly lacks emotional experience, raises a lot of questions as to how we should really define empathy. I think it strongly highlights the importance of the empathy recipient’s experience which has been historically neglected in empathy research. Regardless of whether the empathizer is exhibiting “true” or “fake” empathy, the emotional consequences for the recipient can be exactly the same.
From a psychological perspective, what are the risks and benefits of letting AI play a meaningful role in emotional support?
I think this is one of the most critical questions moving forward and requires substantial further exploration. Loneliness and isolation are a major concern right now with many people lacking meaningful friendship or support. AI could possibly serve as an emotional support supplement for those who lack it altogether, or in specific contexts where quality support may be scarce. Emerging research—including our own—suggests possible short-term emotional benefits from these interactions. However, there is an enormous amount of uncertainty regarding the long-term effects. It’s unclear whether people actually benefit from AI empathy in the long-term, and there are reasonable concerns regarding the risk that AI may replace rather than supplement human relationships and empathy. These concerns are completely warranted, and I think time will tell whether we are able to take advantage of the potential positives of AI emotional support, without succumbing to the possible risks.
