Stanford WP: attacking tasks worth 1% of GDP could slash output 68% in weak-link economy
Afinetheorem · x · 2026-11-15
A new early-stage working paper by Stanford's Charles Jones and Christopher Tonetti, "The Weak-Link Asymmetry: AI's Peril and Promise," applies Kremer's O-ring model to AI:
- With task substitution elasticity of 0.2, boosting productivity 10x in tasks worth 1% of GDP raises output by <1%, but cutting it 10x drops output by 68% on impact.
- The same complementarity that makes automation gains slow makes attacks more harmful and easier to execute.
- Applied to AI: attacks on financial systems, power grids and critical infrastructure; when a larger AI sector makes models more capable and attacks more likely, positive productivity spillovers are negligible versus potential damage, so the market overinvests in AI.
- The FAA ground stop is cited as a vivid illustration of micro-level systemic fragility.
More from AGI Musings
- 426 games slated for Steam in 3 days as AI creation floods the market — jhnnassky · 2026-11-15
- Ethan Mollick ranks AI slop replies by annoyingness, from copy-paste to fake curiosity — emollick · 2026-11-15
- Only 3% of Americans accept AI-only diagnosis, Pew survey finds — PAstynome · 2026-11-15
- Geoffrey Hinton renews warning: superintelligent AI may be impossible to keep under control — VraserX · 2026-11-15
- 'The lab is the bottleneck, not the molecule': a take on AI drug discovery — shyamalanadkat · 2026-11-15
- Thousands protest AI data centres in Edinburgh amid European backlash — emax · 2026-11-15