AI Alignment Research Criticized for Being Overly Technical

joshua_saxe · x · 2026-07-16

The author agrees that while their critique of the AI alignment research paradigm may not be as rigorous as a formal investigation, it remains valid.

The core argument is that the current AI safety research paradigm has "embedded" a certain bias in a Kuhnian sense. This paradigm consists of a whole set of simplifying assumptions, research subjects, capital and infrastructure, training programs, and classic texts (e.g., MATS, Bostrom / Superintelligence), which collectively frame "misalignment" primarily as a technical problem.

The author believes this framework severely underestimates the importance of treating society and social processes as first-class research subjects.

Original post →

More from AGI Musings

AGI Musings channel →