AI and Autonomous Systems Risk

The claim about a closing window is widely cited by arms control experts and the UN Secretary-General, but the specific timeline is uncertain. Technology diffusion rates suggest the window is narrowing, but no firm empirical basis exists for a precise deadline.

Regulation can adapt to technology. Historical precedent with chemical weapons, landmines, and nuclear arms shows that treaties can be effective even after initial proliferation, though enforcement challenges increase.

Complex adaptive systems theory suggests that once autonomous weapons are widely deployed, the competitive dynamics and lock-in effects make rollback orders of magnitude harder than prevention. The analogy to nuclear weapons is apt but underestimates the lower barrier to entry for AI weapons.

Specific figure comes from limited experimental studies. Real-world performance varies dramatically by system, environment, and target type. The directional claim (AI misidentification exceeds acceptable thresholds) has broader support than any single percentage.

Consistent with broader pattern of AI systems performing poorly on edge cases and in contexts not well-represented in training data. Military applications compound these failures with lethal consequences.