Understanding Color Metaphors: Insights from ChatGPT's Limitations

In a groundbreaking study published in the journal *Cognitive Science* on July 8, 2025, researchers explored the capacity of ChatGPT, a prominent artificial intelligence language model, to interpret color metaphors such as 'feeling blue' and 'seeing red'. Led by Professor Lisa Aziz-Zadeh, Director of the USC Center for the Neuroscience of Embodied Cognition, the study's findings highlight the distinctions between AI's understanding and human cognition regarding color perception and metaphorical language.

ChatGPT operates by processing vast amounts of textual data and generating human-like responses based on learned patterns. However, as Aziz-Zadeh points out, the AI lacks direct sensory experiences that influence human understanding of colors. "ChatGPT uses an enormous amount of linguistic data to calculate probabilities and generate very human-like responses," she stated. "But what we are interested in exploring is whether or not that’s still a form of secondhand knowledge, in comparison to human knowledge grounded in firsthand experiences."

The interdisciplinary research team, which included psychologists, neuroscientists, social scientists, and computer scientists from institutions such as UC San Diego and Stanford University, conducted extensive online surveys. Participants were categorized into four groups: color-seeing adults, color-blind adults, painters, and the AI model itself. They were tasked with associating colors with abstract concepts and interpreting familiar and unfamiliar color metaphors.

Interestingly, results indicated that color-seeing and color-blind individuals exhibited surprisingly similar color associations, suggesting that visual perception may not be a prerequisite for metaphorical understanding. Painters, however, significantly outperformed other groups in interpreting novel metaphors, underscoring the value of hands-on experience with color.

ChatGPT produced consistent color associations and drew from emotional and cultural contexts when explaining color metaphors. For instance, when discussing a 'very pink party', ChatGPT associated the color pink with positive emotions and good vibes. Nevertheless, the AI model struggled with more nuanced metaphorical interpretations, particularly with unfamiliar or inverted metaphors like 'the meeting made him burgundy'. This limitation highlights the constraints of language-based models in replicating the full spectrum of human cognition.

The research raises critical questions about the future of AI in understanding language. As AI technologies develop, integrating sensory inputs might enhance models’ capabilities in approximating human cognition. Aziz-Zadeh emphasizes this potential, stating, "This project shows that there's still a difference between mimicking semantic patterns and the spectrum of human capacity for drawing upon embodied, hands-on experiences in our reasoning."

With ongoing advancements in artificial intelligence, particularly in natural language processing, studies such as these are crucial for understanding the implications of AI in linguistics and broader cognitive science. Future inquiries may further elucidate how sensory experiences shape language interpretation and how these insights can inform AI development to better align with human understanding of complex metaphorical language.