Why a temporary slowdown is the wrong ask, and what we actually want governments to do.
Assumes safety will catch up on a schedule. Treats superintelligence as inevitable, just early. Hands the restart button back to the same race once the clock runs out.
Prohibits systems past a general-capability threshold until control is actually solved. No scheduled restart. The ban holds for as long as the science cannot make the gamble survivable.
Not at small scale. Not at large. Leading labs say so in public. A pause that assumes alignment will finish on time is a bet with no method behind it. Prevention is the off-ramp from that bet.
Medicines, weather models, industrial optimizers, coding assistants: tools with no goals of their own stay legal and funded. The line falls around the autonomous general mind, not around AI as a whole.
The Nuclear Non-Proliferation Treaty does not "pause" bombs for five years. It limits who may build them, with inspections. Chemical weapons and ozone-depleting chemicals got the same treatment. Superintelligence needs that shape of law, not a kitchen timer.