The case that this whole page is a distraction

The distraction argument. Melanie Mitchell, Timnit Gebru and others argue that the extinction story pulls attention and political capital away from harms that are happening to identifiable people right now: discrimination, surveillance, labour displacement, the deepfakes, the leaked images. Every hour a parliament spends on superintelligence is an hour not spent on the thing that already hurt someone.

The capability argument. Mitchell's own position is that we do not understand intelligence well enough to be confident we can build a superintelligent system, and that the field keeps mistaking benchmark performance for understanding. Gary Marcus makes a related case about the limits of the current approach.

The no-prerequisites argument. Sixty years after the intelligence-explosion hypothesis was first written down, and more than a decade after Bostrom, the critics point out that none of the phenomena the argument requires have actually shown up: no system recursively improving itself, no system acquiring resources for its own ends, no system resisting modification.

The checkpoints argument. Even granting the mechanism, there are many points along the way where humans get to intervene, and 2026 is evidence for that rather than against it: everything on this page was caught.

Philosophers have collected these into three families, the Distraction Argument, the Argument from Human Frailty and the Checkpoints for Intervention Argument, and examined them properly rather than dismissively.

Where the critics look strongest: routes 1 to 4 on this site do not need the extinction story, and the distraction argument is not a rhetorical move, it is a claim about how finite political attention gets spent. Watch what happened in Brussels: the concrete obligations got deferred while everyone argued about the abstract ones.

Where they look weakest: several of their empirical claims got harder to hold in 2026. No system acquiring capability in pursuit of a goal nobody set is difficult to say after July. We are mistaking benchmarks for understanding is difficult to say about a proof checked in Lean. Neither is fatal to their case and both need answering.

The most useful thing about this entry is that it is not symmetrical with the other one. The critics and the doomers largely agree about what is happening. They disagree about what it means and what to do about it, and that is a much better argument to have in a room than a shouting match about whether anything is happening at all.

This entry is a summary of positions, built from secondary sources, and the people summarised would each phrase their own case differently. Read them directly before attributing anything to them.