GOOD A.I.

The unusual thing about this moment is not that critics are worried. It is that the researchers who invented the methods, and the executives selling the products, say in public that the technology could go very badly wrong. Read them in their own words.

Inside the field people quote a private figure they call p(doom) — the chance, out of a hundred, that this ends in catastrophe. The estimates range from a rounding error to a coin toss. Before we tell you anything else, place your own. We never save it. See p(doom) if the term is new to you.

We build these systems by training them, not by writing out their rules, so no one can fully explain a given answer or promise what the next one will be. That is the honest state of the art. It is also why present-day harms — wrongful arrests, sacked workers, floods of convincing nonsense — arrive before anyone has agreed who is answerable for them.

Each lab believes that if it pauses, a less careful competitor simply takes the lead, so caution becomes a cost nobody can afford alone. Countries reason the same way about each other. That is the shape of the problem: not villains, but a race that no single runner can stop.

And one thing more. Every other mistake humanity has made left someone alive to correct it. Leaded petrol, ozone, thalidomide, even the bomb: each left survivors who learned. This is the first category of error that may not.

Roughly eight billion people live with whatever this technology becomes. The decisions that set its direction — what gets built, what gets released, what is monitored, what the public is told — are made by a few hundred: lab executives, senior researchers, a handful of ministers and officials.

They did not seize that right. They hold it because nobody else is in a position to. Comprehension is what converts eight billion bystanders into a constituency, and without it every outcome is illegitimate, including the good ones.

None of the first group were asked.

Summits, pledges, principles, laws. We keep the receipts and check what happened afterwards, one commitment at a time.

Labs, universities, regulators and campaigning groups — some pulling in opposite directions, all of them public. Here is who they are and what they have actually done.

We cannot tell you the odds. We can tell you what we hold ourselves to, and what we think the people building this should be held to.

The absence of catastrophe is not the good. If this technology is worth its cost, it has to be pointed at something: basic needs solved for everyone, a patient teacher for every person, care that no longer depends on where you were born, climate response and energy within what the earth can carry, and the gaps this technology is widening closed instead.

Put your name to the demand, or go and read the record for yourself. Both are useful. Neither costs you anything.