AK or just okay? AI and economic growth (jzmazlish.substack.com)

Economist Nick (writing after attending AI-focused events) argues that the big question isn’t whether AI will matter but whether it can reach “true” AGI capability—because conditional on that, we should expect explosive growth. Using the AK model (where AI capital substitutes for labor), he explains how full automation turns diminishing-returns dynamics on their head: capital accumulation yields sustained exponential growth and, combined with accelerating technological progress, could produce double‑exponential or even hyperbolic GDP outcomes. Crucially, skepticism about transformative growth often conflates doubts about economic bottlenecks (regulation, adoption) with doubts about underlying AI capabilities; he stresses these are distinct and falsifiable hypotheses. On timelines and technical signals, he points to METR/task-time horizon trends as the best available monitor: models’ coherent, longer-horizon performance is improving and, by simple extrapolation, suggests one-week task horizons around 2030–31 and one-year horizons by ~2034, yielding AGI-ish timelines near 2035–40 (with room for faster or slower paths). Reasonable skeptic arguments remain—chiefly financing limits and a “data wall”—but persistent scaling laws, RL-driven domain gains (esp. in software), and the plausibility of recursive self‑improvement make outright dismissal of near-term AGI appear faith‑based. For the AI/ML community this implies urgent priorities: build long‑horizon benchmarks, stress-test safety and deployment pipelines, and prepare policy responses for potentially rapid economic and R&D acceleration.