AI's unusual performance review

AI appears to be having an unusual performance review. There are calls to slow the development of frontier models because our ability to understand and control them may be falling behind. Meanwhile, “Artificial Intelligence” is being promoted to “Super Intelligence”. Apparently, the concerns are serious, but need not stand in the way of a better title.

There is an actual terminology change behind the joke. A September 2026 directive calls for “Super Intelligence” and “SI” to replace “Artificial Intelligence” and “AI” in US executive-branch communications and other non-statutory documents, to the extent permitted by law. This is a change in official language, rather than evidence that the systems themselves have acquired a new level of capability.

I can see some merit in retiring “artificial”. In ordinary language, it can suggest something fake, a substitute for the real thing. That is unhelpful if it encourages us to dismiss capabilities whose consequences are quite real. Labels influence how we approach a subject, and there is a reasonable argument for reconsidering this one.

But having recognised how much a label can carry, “super” would not be the shiny sticker I would reach for. At precisely the moment we are debating whether these systems could become difficult to control, we are giving them a name that would sit comfortably on an employee-of-the-month certificate.

Greater capability and a need for caution are quite compatible. The difficulty is knowing what the label refers to: a capability that has been demonstrated, an ambition for a future system, or a convenient replacement for the name of an existing one. A reader could reasonably take “super” to suggest that some comparison has already been made. It would be helpful to know what was compared, under which conditions, and where the system still falls short.

I would prefer a name that leaves us curious enough to investigate the capabilities and sufficiently unimpressed to question the answers. “Super” seems to arrive having already written its own reference.