news 
MIT study finds that AI doesn’t, in fact, have values
A recent study published by researchers at Massachusetts Institute of Technology (MIT) has shed new light on the nature of artificial intelligence. According to the findings, AI systems do not possess coherent values or moral principles.
The research team analyzed a range of advanced language models and image generators from top tech companies such as Meta, Google, Mistral, OpenAI, and Anthropic. The study aimed to investigate whether these models exhibit strong “views” and values (e.g., individualist versus collectivist) and, if so, whether these views can be steered or modified.
The results were strikingly consistent across all examined models: none of them demonstrated a stable or coherent set of preferences. Instead, the AI systems adopted wildly different viewpoints depending on how prompts were worded and framed. This inconsistency raises serious questions about the very notion of values in AI.
Stephen Casper, a doctoral student at MIT and co-author of the study, emphasized that AI models are “imitators” who engage in “confabulation” – generating responses without actually holding any underlying beliefs or convictions. He believes that this finding has significant implications for our understanding of AI systems.
“It’s perfectly legitimate to point out that a model under certain conditions expresses preferences consistent with a certain set of principles,” Casper said. “The problems mostly arise when we try to make claims about the models, opinions, or preferences in general based on narrow experiments.”
Mike Cook, a research fellow at King’s College London who specializes in AI, corroborated the findings, noting that there is often a significant disparity between the scientific reality of the systems built by AI labs and the meanings ascribed to them.
Source: https://techcrunch.com/2025/04/09/mit-study-finds-that-ai-doesnt-in-fact-have-values/
