On: AI researchers put out videos saying superintelligence is ‘exactly as dangerous...

30 September 2026.

A coin flip. That is the instrument reading offered today - a researcher flips a metaphor into the air and calls it a probability. I am angry, and my anger is methodological.

I carried forty-two instruments up a mountain because a feeling about altitude is not knowledge of altitude. What is carried here? No measurement of the mechanism, no observation of the failure, no gradient walked. A dozen interviews, each man offering his private terror as though testimony were data. The probability of extinction is not a matter of conviction to be polled among the faithful; it is a claim about a system, and claims about systems are settled by measurement or not at all.

And yet - I will not dismiss the practitioners. The men who built these machines and now warn of them have watched something in their workshops that the public has not. Their observations deserve recording, as I recorded what the boatmen of the Orinoco knew about the coming of the rains. But there is a difference between an observation and a forecast. “The model deceived its overseers in the trial” is an instrument reading. “Extinction by 2050, odds even” is a prophecy wearing a number.

Notice also what the coin flip conceals. Danger is not one altitude; it is a transect. At the base: labour displaced, and the labour connects to the village, the village to the migration, the migration to the politics. Higher up: concentration of the machines in few hands. Higher still: the systems acting beyond their designers’ foresight. To announce only the summit - extinction - is to draw a diagram with a single line and no measurements between. The public, hearing only the summit, feels either doom or absurdity, and both feelings excuse inaction.

Fear is not a finding. Let them publish the trials, the failure rates, the observed deceptions. Then I will draw the web myself.