Comparing the value of perceived human versus AI-generated empathy

Area of Science:

• Human-computer interaction
• Psychology
• Artificial intelligence

Background:

• Large language models (LLMs) exhibit advanced social-emotional skills, potentially enhancing human-AI interactions and emotional support.
• The perception of empathy, including understanding, emotional resonance, and care, may differ when attributed to AI versus humans.

Purpose of the Study:

• To investigate whether empathy is perceived differently when attributed to AI compared to humans.
• To assess the impact of AI-generated empathic responses on human emotional perception and preference for interaction.

Main Methods:

• Nine studies involving 6,282 participants evaluated AI-generated empathic responses labeled as either human or AI-provided.
• Varied response characteristics (length, delay, iteration) and LLMs were employed to test the robustness of findings.
• Participant emotional states and preferences for human versus AI interaction were measured.

Main Results:

• Human-attributed empathic responses were rated significantly higher in empathy and support than AI-attributed ones.
• AI-attributed responses elicited more positive and fewer negative emotions compared to human-attributed responses.
• Participant belief that AI assisted human responses diminished perceived empathy and support, particularly for responses emphasizing emotional sharing and care.

Conclusions:

• Human empathy is perceived as more authentic and supportive than AI-generated empathy, even when the AI's role is undisclosed.
• Emotional sharing and care elements in AI responses do not fully bridge the perceived empathy gap with human interactions.
• Individuals consistently prefer human interaction over AI for emotional engagement, indicating current limitations in AI's capacity for genuine empathic connection.