This new framing of human communication around power (weak vs powerful) danger (safe vs dangerous), and order (ordered vs chaotic). Super interesting with profound implications.
This would overturn an emotion-based framing of valence, arousal and dominance (VAD) that has been the consensus for more than 70 years — a consensus that has has far reaching implications in fields from psychology and linguistics to LLMs/AI and neurobiology.
From the Phys.org article:
“In artificial intelligence, the implications are immediate. Many natural language processing systems rely on sentiment analysis grounded in VAD-like frameworks. If those frameworks misrepresent the essential nature of meaning, then AI systems may be systematically misinterpreting human language.”
My own thought, based on the article, is that the researchers study seems based on relatively small of sample of texts, limited to English and formal translations of other language literature into English. Significant work would need to be done to validate more comprehensively.
That said, it underscores how profoundly AI’s underlying assumptions, based on inaccurate models of human language and behaviour, may be misreading the texts they are analyzing.
https://phys.org/news/2026-05-human-language-deep-safety-bias.html