In what seems like yet another setback for a domain where we believed humans would always surpass machines, researchers now propose that AI comprehends emotions better than we do.

Researchers have discovered that artificial intelligence demonstrates a superior grasp of emotions — achieving significantly higher scores than the average human when selecting the appropriate response to de-escalate emotionally intense situations.

In a recent study published on 21 May in the journal Communications Psychology, scientists from the University of Geneva (UNIGE) and the University of Bern (UniBE) administered widely-used emotional intelligence (EI) assessments (STEM, STEU, GEMOK-Blends, GECo Regulation and GECo Management) to well-known large language models (LLMs), such as ChatGPT-4, ChatGPT-o1, Gemini 1.5 Flash, Claude 3.5 Haiku, Copilot 365, and DeepSeek V3.

The research focused on two aspects: comparing AI performance with that of human participants, and evaluating AI’s ability to generate new test items consistent with the objectives of EI assessments.

By analyzing validated human responses from earlier studies, the LLMs selected the "correct" response in emotional intelligence tests 81% of the time—based on expert human judgment—compared to 56% by humans.

When tasked with generating new test questions, human evaluators found these efforts comparable to the original tests regarding difficulty level and confirmed they weren't merely rephrased versions of existing ones. The correlation between AI-generated and original tests was deemed “strong,” with a correlation coefficient of 0.46 (where 1.0 indicates perfect correlation and 0 indicates no correlation).

The general conclusion drawn was that AI is more adept at "understanding" emotions than humans.

Sign up for the Live Science daily newsletter nowThe deeper story

Upon consulting several experts, a recurring theme was the importance of considering the methodology used. Each of the EI tests employed in the study consisted of multiple-choice formats—scenarios not fully reflective of real-life situations where interpersonal tensions are high, as noted by specialists.

“It's important to remember that people don’t always concur on how someone else feels, and even professionals can interpret emotional cues differently,” said Taimur Ijlal, an expert in finance and information security. “Therefore, ‘outperforming’ a person on such a test doesn't necessarily imply the AI has greater insight—it simply chose the statistically anticipated answer more frequently.”

They added that the skill being assessed isn't truly emotional intelligence but rather something else. “Artificial Intelligence systems excel at recognizing patterns, especially when emotional indicators follow identifiable structures like facial expressions or verbal signals,” explained Nauman Jaffar, Founder and CEO of CliniScripts—an AI-based documentation platform designed for mental health practitioners. “However, equating this with a profound 'understanding' of human emotions overestimates what AI actually achieves.”

AI excels in structured, quantitative quiz settings rather than grasping the intricate subtleties required for genuine emotional comprehension, some specialists emphasized. One key observation was that AI performs better on tests involving emotional contexts removed from real-time experiences—unlike how humans encounter them.

Jason Hennessey, founder and CEO of Hennessy Digital—who has spent years studying how search and generative AI systems process language—likened the study to the Reading the Mind in the Eyes Test, a commonly used method for assessing emotional states and one where AI has shown potential. But as Hennessey pointed out, when variables like lighting or cultural context change in such tests, “AI accuracy plummets dramatically.”

Most experts generally considered the assertion that AI "understands" emotions better than humans to be somewhat exaggerated.

“Does it show LLMs are effective at categorizing typical emotional reactions?” asked Wyatt Mayham, founder of Northwest IT Consulting. “Certainly. But it's akin to calling someone an excellent therapist just because they aced an emotionally themed BuzzFeed quiz.”

Still, there remains an important exception: despite relying on pattern recognition instead of true emotional comprehension, AI has demonstrated superior performance in identifying and responding to emotional states in certain cases.

Aílton, a conversational AI utilized by over 6,000 long-haul truck drivers in Brazil, serves as a multimodal WhatsApp assistant using voice, text, and images. Its creator, Marcos Alves, CEO & Chief Scientist at HAL-AI, claims Aílton detects stress, anger, or sadness with about 80% accuracy—approximately 20 percentage points higher than its human counterparts—all within real-time emotional interactions.

In one instance, Aílton swiftly and appropriately responded to a driver who sent a distressed 15-second voice message following a coworker’s fatal accident, delivering thoughtful condolences, offering mental-health resources, and automatically notifying fleet managers.

The above is the detailed content of New study claims AI 'understands' emotion better than us — especially in emotionally charged situations. For more information, please follow other related articles on the PHP Chinese website!