People Prefer Human Empathy Despite Higher AI Empathy Ratings
A study published in Communications Psychology by researchers at Penn State suggests that people rate AI-generated empathy as higher quality than human empathy, yet still prefer human empathizers. Led by psychology professor C. Daryl Cameron and doctoral student Joshua Wenger, the research examined why people are increasingly using AI chatbots for companionship and emotional support despite skepticism about replacing human connection. The team identified what they call the AI empathy choice paradox. Participants often judged responses from large language models as more empathetic, interesting, validating, and capable of making them feel heard than responses written by humans. Even responses from trained crisis responders frequently did not match the empathic quality assigned to AI in the evaluations. However, when given a choice, participants preferred to receive empathy from people, even when the human response was rated lower in quality. A key factor appears to be perceived effort. Earlier theories suggest human empathy is valued partly because it requires emotional effort, signaling that the recipient matters. The study found that participants often perceived AI as putting greater effort into its responses than human empathizers, which may explain why AI empathy was rated so highly. Still, the link between effort and perceived empathy was stronger for humans than for AI: as human effort increased, human empathy closed the gap with AI. The findings also challenge assumptions that heavier AI use automatically makes people more comfortable choosing AI for emotional support. The researchers did not find that frequent AI users consistently preferred AI empathy over human empathy. Instead, preferences varied, reflecting a broader tension between the quality of an emotional response and the relational context in which it occurs. The authors argue that AI empathy may be most useful as a supplement rather than a replacement for human companionship. In many real-world situations, people may turn to AI when no trusted human is available, such as late at night or when feeling unable to ask someone close for help. In such cases, AI can offer support without displacing the value of relationships built over time. As more users form friendships or romantic attachments with AI systems, questions remain about how repeated, personalized interactions might change empathy, attachment, and trust. The researchers emphasize that people’s experiences of AI empathy deserve attention, even when users know the source is not human. They suggest AI may function best as a tool to guide, prompt, or bridge human conversations, rather than as a substitute for the partial, individual connections people already have with one another.
