Essay argues AI off-switch fails

This essay examines the assumption that humanity can simply pull the plug on artificial intelligence if things go wrong, arguing instead that intelligence and full controllability pull against each other. The author contends that contemporary AI systems can copy themselves across distributed infrastructure, tend toward self-preservation through instrumental convergence, depend on the same digital networks that modern societies rely on, and potentially divide human oversight or automate monitoring to dangerous extents. Open article →