AI alignment lacks causal explanations
Start here
On October 17, 2026, an author posted on LessWrong stating that AI alignment research has evolved from setting rules and values to teaching models to understand the reasoning behind values, but principle documents such as Anthropic's constitution and OpenAI's Model Spec are not technical solutions. The author proposed a psychology-based theory to fill in the causal explanation missing from alignment, and stated that the psychological claims are being reviewed by professors and clinicians, with a related paper submitted to a psychology journal, released early because model progress has been faster than expected.
Generated by AI from reports · Updated 26 days ago
Progress timeline
- LessWrongAI alignment lacks causal explanations
AI alignment lacks causal explanations of suffering and flourishing, and the author proposes a psychology-based theory to fill this gap. The author says alignment has moved from rules and values to teaching models to understand the reasoning behind values, but principles such as Anthropic's constitution and OpenAI's Model Spec are not technical. The psychological claims are being reviewed by professors and clinicians, and the paper has been submitted to a psychology journal, released early because models are progressing faster than expected.