Position: Evaluations of AI Moral Reasoning Still Miss Half of the Picture
Because most evaluation tools are built on descriptive ethics frameworks like Moral Foundations Theory or Kohlberg’s stages, they capture what people value but not how those values shift across contexts. The authors reviewed popular benchmarks (e.g., MoralStories, ETHICS, Social Chemistry) and found that over 80 % of items probe value alignment, while fewer than 20 % require […]
Because most evaluation tools are built on descriptive ethics frameworks like Moral Foundations Theory or Kohlberg’s stages, they capture what people value but not how those values shift across contexts. The authors reviewed popular benchmarks (e.g., MoralStories, ETHICS, Social Chemistry) and found that over 80 % of items probe value alignment, while fewer than 20 % require reasoning about competing norms or contextual exceptions.
Word count? Let’s count: Because1 most2 evaluation3 tools4 are5 built6 on7 descriptive8 ethics9 frameworks10 like11 Moral12 Foundations13 Theory14 or15 Kohlberg’s16 stages,17 they18 capture19 what20 people21 value22 but23 not24 how25 those26 values27 shift28 across29 contexts.30 The31 authors32
📌 Source: Arxiv Ai
Related Articles
DOW-KE: Anchor-Free Multi-Layer Knowledge Editing via Direct End-to-End Weight Optimization
Knowledge editing in large language models has become a practical way to correct factual errors or inject new information without
Proactive Road Safety Intervention in Australia: Predicting Risky Driving Hotspots from Connected Vehicle Data
Transport agencies in Australia have long depended on crash reports to spot dangerous roads, a method that only reveals problems
A decodability criterion predicts when hidden-state selection beats majority voting in large language models
When a language model generates several answers to the same prompt, the usual way to pick a final response is