The Invisible Graveyard
We count the visible harms of acting and never the invisible harms of waiting. Delay has a body count — it just doesn't hold funerals you attend.
Unsafe Intelligence
The economist Alex Tabarrok has a phrase for the people who die because a drug was held up by an extra few years of caution: the invisible graveyard. They have no names. There is no headline, no inquiry, no funeral you attend. But they are just as dead as the people a hasty approval might have harmed — and there are usually far more of them. The difference isn't the body count. It's the visibility.
This is the most important asymmetry in how we think about risk, and we get it backwards almost every time.
One error gets counted; the other doesn't
When a cautious institution approves something that later goes wrong, the harm is vivid and attributable. There's a name, a face, a regulator to haul in front of a committee. So institutions learn to fear that error — the Type I error, the disaster of action — above all else.
The opposite error — the Type II, the disaster of inaction — produces no such evidence. The cure that arrives five years late kills exactly as many people as the one that's pulled five years early. But those deaths are spread thin, diffuse, never assembled into a story. Nobody is blamed for a treatment that didn't exist yet. So the cost of caution is real and the cost of caution is invisible, and a rational bureaucrat, weighing a vivid risk against an invisible one, will be too cautious every single time. The incentives guarantee it.
The graveyard has an AI wing now
Point this at the AI transition and the shape is the same. Every year of "let's slow down and be careful" in the middle of this thing is a year of foregone abundance — the diagnoses not made, the materials not discovered, the drugs not designed, the productivity not realized, the boring miracle of cheaper everything deferred by twelve more months. None of that shows up in a statistic. There is no headline that reads "this treatment would have existed in 2029 instead of 2032." But the people on the wrong side of that gap are no less real for being uncounted.
The Precautionary Principle sounds like prudence, and sometimes it is. But it has a price, and the price is always paid in a future we don't get. "Do no harm" quietly becomes "do nothing," and doing nothing is itself a choice with a body count.
The honest caveat
This cuts only so far, and we won't pretend otherwise. Some harms are genuinely irreversible — the Collingridge dilemma is real, and where a mistake truly can't be undone, the asymmetry flips and caution is correct. A bioweapon, a hard-to-recall capability, a one-way door: those deserve the brakes.
But that's a narrow set, and most of what gets sold as "go slow" is not insuring against a one-way door. It's insuring against the visible tail while ignoring the certain cost of delay — buying peace of mind for the regulator at the expense of the uncounted. Lock-in to stagnation is also irreversible. It just never makes the news.
Put the graveyard on the books
The discipline we'd ask for is simple and almost never practiced: measure the road not taken. Hold yourself accountable to opportunity cost, not just to visible harm. When you weigh the danger of moving, weigh the danger of waiting on the same scale — both error types, in the same units. The invisible graveyard should appear in the cost-benefit, even when it can't appear in the newspaper.
Because the only catastrophe that never gets reported is the one where nothing happened — slowly, carefully, and for years.