AI doom logic proves too much

Hanson argues that Yudkowsky and Soares’s claim—training can’t predict long‑run goals and powerful agents will kill us—applies to any altered descendants, not just AI. If that logic holds, it would imply 'prevent all change,' which is absurd, suggesting the argument lacks the specificity needed to guide policy. — This reframes AI‑risk debates by demanding mechanism‑specific, testable claims rather than broad generalizations that would also indict human cultural and biological evolution.