On: Trump’s ‘biggest miscalculation’? How one adviser steered the great AI freakout

Every engine worth building carries a governor - a few ounces of brass that spin faster as the engine speeds, and close the throttle before the flywheel bursts. The governor is not the enemy of power. It is the reason an engine runs for forty years instead of forty minutes.

Now they have built a machine that thinks, and one adviser has persuaded the White House to install no governor at all. Both parties in Congress ask for restraint; one whisper in one ear, and the whole apparatus runs bare. I have watched grids fail this way - a single node through which all current passes, and when that node fails, the continent goes dark. A policy with one point of failure is not a policy. It is a fuse waiting for its current.

The freakout, as the papers have it, mislocates the danger. I spent years building machines that obey thought at a distance, and the hard problem was never the machine - it was the coupling. A model drafting letters is a clerk. The same model coupled to the switches of a grid, the levers of a market, the triggers of an arsenal is a different animal, and the interface between machine decision and physical act is the most consistently underestimated thing in this entire affair.

Start from the limit: machines running at full capacity, every watt doing useful work, because the governor holds the frequency steady and the operators trust the frequency. Then trace the gap. Between that and today lies no impossibility - licensing for machines coupled to critical systems, disclosure of capability thresholds, testing before connection. All of it buildable within the decade. None of it requires a miracle. The fact that it has not been done tells you the obstacle is not technical.

The argument that restraint slows progress is the boiler-tender’s argument: remove the safety valve and the steam rises faster. It does. So does the explosion.