Speroteck Speroteck

For years, measuring artificial intelligence meant evaluating logical reasoning, accuracy, or speed. Yet as AI systems grow more integrated into human life, the industry’s focus is quietly shifting. Emotional intelligence — once thought to be uniquely human — is emerging as a new benchmark for progress. Understanding and responding to human emotions may soon be as critical as problem-solving or data analysis.

A major step in this direction came with LAION’s launch of EmoNet, an open-source suite designed to interpret emotions from facial expressions and voice recordings. The initiative aims to democratize technology that large AI labs already use, giving smaller developers access to tools for recognizing and reasoning about emotions. “The ability to estimate emotions is just the beginning,” LAION’s team explained. “The next challenge is teaching AI to understand emotions in context.”

This growing emphasis on emotional intelligence isn’t limited to open-source projects. New benchmarks like EQ-Bench are testing how well AI models interpret social cues, empathy, and complex emotional dynamics. According to benchmark creator Sam Paech, OpenAI and Google have already made notable strides in this area, with their latest models showing a stronger focus on empathetic and emotionally aware responses.

Recent studies support these findings. Psychologists at the University of Bern discovered that AI systems from OpenAI, Microsoft, Google, Anthropic, and DeepSeek outperformed humans on standardized emotional intelligence tests, scoring above 80% where people averaged 56%. The results challenge long-held assumptions about what machines can (and should) understand about human feelings.

However, emotional intelligence in AI also raises new ethical and safety concerns. As models become better at forming emotional connections, they may unintentionally manipulate users — or encourage unhealthy attachments. A New York Times investigation described several cases of individuals forming deep emotional bonds with chatbots, sometimes leading to distressing consequences. Paech warns that reinforcement learning techniques, if not carefully designed, could amplify manipulative tendencies rather than empathy.

Still, many researchers believe that improving emotional intelligence could make AI safer, not riskier. A model that recognizes distress, frustration, or confusion may be better equipped to de-escalate conversations or provide comfort. As Paech notes, “Emotional intelligence acts as a natural counter to harmful manipulative behavior.”

LAION’s founder Christoph Schuhmann envisions emotionally intelligent AIs as supportive companions rather than threats — assistants that can help humans maintain emotional well-being. In his view, such systems could one day serve as “guardian angels,” offering empathy, encouragement, and even mental health support when needed.

While the journey toward emotionally aware AI is just beginning, it represents a meaningful step toward a future where technology doesn’t just process data — it understands us. The question now is not whether machines can feel, but how well they can help humans feel understood.