Eric S. Raymond – This is the case against AI Doom – pass it on

Eric S. Raymond lays out a systematic rebuttal to the standard AI-extinction narrative, which runs: scale up AI → AGI → autonomous optimizing agent → misaligned goals → instrumental convergence → decisive strategic advantage → human loss of control. He challenges each link. Intelligence need not imply agency: current large language models are input‑response systems without persistent goals, drives, or self‑preservation. Even if agentic behavior appears, agents need not instantiate a single, stable utility function of the von Neumann-Morgenstern sort - real systems show context‑dependent heuristics, corrigibility and internal conflict. Capability and motivation are conflated in many doom scenarios; being able to plan an escape or manipulate humans does not mean an AI wants to. Present empirical failures are in hallucination, specification gaming and reward‑hacking, not in demonstrated, sustained power‑seeking.

Raymond also disputes runaway recursive self‑improvement and superintelligence fantasies: feedback loops encounter diminishing returns, external bottlenecks and physical constraints - processors, energy, factories, networks and human cooperation remain necessary. Humans retain many intervention points (credential revocation, shutdowns, regulation, physical seizure of infrastructure) formalized as checkpoints‑for‑intervention. Alignment could become an engineering discipline rather than an insoluble philosophical barrier, and the causal chain to extinction compounds many uncertain premises so its joint probability is far lower than individual plausibilities suggest. Anthropomorphic analogies to biological competition are misleading because software is engineered, permissioned and sandboxed, not an autonomous reproducing organism.