The future of AI is already written (www.mechanize.work)

Matthew Barnett, Tamay Besiroglu, and Ege Erdil argue that the “choice” between building AI that assists humans and AI that fully automates work is largely illusory: once the technical prerequisites and economic incentives align, transformative technologies emerge reliably and rapidly. They marshal historical evidence of simultaneous discovery (Hall–Héroult aluminum smelting, independent jet‑engine inventors, Bell and Gray’s telephone patents), cultural convergence (independent development of agriculture, writing, metallurgy), and biological convergence (similar eye structures evolving twice) to show that a tech “tree” is discovered, not freely forged. They note a concrete recent example: compute capacity inferred from NVIDIA revenues suggests hardware able to train GPT‑4‑level models became available around 2020, and GPT‑4 followed in 2022—illustrating how capability thresholds trigger quick adoption. Technically and economically, the paper says full automation is inevitable because autonomous systems remove the human bottleneck inherent in augmentation (e.g., one human per task versus many robots scaling concurrently). Attempts to ban or delay uniquely powerful technologies have historically only postponed adoption when substitutes don’t match their advantages (nuclear weapons, encryption, printing). The implication for AI/ML practitioners and policymakers is clear: prepare for wide replacement of human labor, prioritize safe, scalable automation, and focus governance on managing transition risks and distributing the enormous potential benefits—medical breakthroughs, extended lifespans, and accelerated invention—rather than assuming we can indefinitely preserve the status quo.