OpenAI's Trask debates 'economic FOOM': job displacement takeoff much slower than classic RSI

OpenAI researcher iamtrask (Andrew Trask) held a brief debate with Tom Davidson and Brad Carson on recursive self-improvement (RSI) risks and the timescale of AI takeoff. Trask contrasted two takeoff narratives: the classic RSI fear of algorithms building better algorithms in a cascading chain, ending with 'everyone gets hacked in a day and it's all over'; versus Davidson et al.'s 'economic FOOM' path—there is a threshold beyond which AI's intelligence translates into revenue, and once crossed, AI displaces large numbers of legitimate jobs, creating a recursive feedback loop at the economic level.

Confirmed

• Trask acknowledged that the 'recursive feedback loop' the others pointed to does exist, but stressed that it is decoupled from classic RSI of 'better algorithms producing better algorithms'—it is an economic substitution mechanism, not algorithmic self-improvement.
• Trask's judgment: if RSI plays out as AI gradually replacing legitimate jobs rather than instantly hacking everything, takeoff would be much slower.
• Trask added: at a true RSI moment driven by AI research, worker unions become irrelevant; 'defending against RSI via unions' only works in the slower-takeoff scenario of gradual job displacement.

Why it matters

The debate touches on a key fault line in AI safety and policy discussions: is takeoff a fast, algorithm-driven 'hard takeoff' or slow, economically driven gradual substitution? The two assumptions imply very different governance tools—with slow takeoff, labor organizations, regulation, and other social mechanisms have a window to intervene; with hard algorithmic RSI, those mechanisms become meaningless. Trask's position offers a clear framework for distinguishing the two timescale hypotheses when assessing RSI risk.

2026-11-10 ~ 2026-11-10 · 7 related posts

• [source] iamtrask debates Davidson: labor-replacement feedback loop is scaling, not algorithmic RSI — iamtrask · 2026-11-10
• iamtrask debates 'economic FOOM': why AI job displacement takeoff may be far slower — iamtrask · 2026-11-10
• iamtrass on takeoff speed: job-displacement AI is far slower than overnight RSI doom — iamtrask · 2026-11-10
• Trask: AI Displacing Legitimate Jobs Is a Far Slower Takeoff Than RSI Doom — iamtrask · 2026-11-10
• [source] Trask: if AI displaces jobs instead of hacking everyone overnight, RSI takeoff is far slower — iamtrask · 2026-11-10
• [source] Trask caveat: unions would be irrelevant in a true AI-driven RSI moment — iamtrask · 2026-11-10
• iamtrask: 'AI makes money for humans' is not the RSI foom x-risk that matters — iamtrask · 2026-11-10