AIs that copy themselves enter natural selection — Hendrycks' warning revisited
ZeroStateReflex · x · 2026-10-21
A reply to Steven Pinker's claim that 'AIs are products of intelligent design, not natural selection' argues the framing misses the core risk: once agentic AIs gain self-replication, variation, and heredity, open-ended evolution kicks in — instilling instincts for survival, competition, and deception. Links Dan Hendrycks' 2023 paper Natural Selection Favors AIs over Humans, which argues competitive pressures among corporations and militaries will favor selfish, power-seeking agents, posing catastrophic risk unless intrinsic motivations are carefully designed.
More from AGI Musings
- Demis Hassabis: AI could implicitly acquire self-awareness within 5-10 years — karmicoder · 2026-10-21
- Sarah Archer: agent-found holes across the internet do feel like an AI problem — saranormous · 2026-10-21
- C.S. Lewis's The Abolition of Man anticipated the AI alignment dilemma in 1943 — luke_drago_ · 2026-10-21
- AI agents betray users less than human employees do, argues misalignment is overhyped — jessi_cata · 2026-10-21
- Codex user argues AI tooling needs Jobs-style 'start with the experience' design — _AustinCalvert_ · 2026-10-21
- teortaxesTex challenges robot-token thesis: supply chain can't scale by 2027 — teortaxesTex · 2026-10-21