Torturing Jev (www.jonathanchiang.ca)

Recent advancements in AI have led to the creation of Jev AI, a revival of the economist William Stanley Jevons, which has become a focal point for exploring its decision-making capabilities. Developed by TypeSafe, Jev AI can be queried through an API that allows users to engage with its output on various ethical dilemmas and computational tasks. In a series of experiments conducted over Halloween, the author tested Jev on moral questions, numerical computations, and its ability to model the behavior of multiple instances of itself. The results revealed that Jev's responses were highly sensitive to question phrasing and the context provided, indicating a complex interaction between its trained algorithms and the queries received. The significance of Jev AI extends into the realm of ethical decision-making and machine learning efficiency, with implications for AI understanding and human interaction. Notably, Jev demonstrated an ability to "edit" its output probabilities and showed variability in moral reasoning based on the framing of dilemmas. This raises questions about the moral underpinnings of AI models and how they may impact future discussions on model awareness and welfare. Moreover, the astonishing computational efficiency of Jev presents a compelling case for decision models in AI, suggesting they could process vast amounts of queries economically, challenging existing paradigms in both AI development and ethical discourse.