Superintelligence is having a moment. A recent executive order proclaims the dawn of an era of "Super Intelligence," and public remarks have applied the term to today's AI models. Sandra Rogers, PhD, published a careful piece this week arguing that this is a category error worth correcting before it hardens into habit: AI is not superintelligence, not yet anyway. She is right — and for the people we work with, the reason matters well beyond semantics.
We are an engineering firm, not a linguistics seminar. But our whole method starts with questioning foundational assumptions until what is actually true falls out, and vocabulary is a foundational assumption. When the words are wrong, everything built on them — budgets, timelines, training plans, safety rules — inherits the error.
Three words, three different things
The field has a standard ladder, and it is worth being precise about the rungs:
- Narrow AI (ANI) is everything that actually exists today. Chatbots, image generators, recommendation engines, agentic assistants that book meetings and write code — extraordinary at bounded tasks, and without understanding beyond them. Generative and agentic AI live here.
- Artificial general intelligence (AGI) would be a machine with human-level intellect across domains: able to reason, plan, and learn in unrelated contexts the way a person can. It is a theoretical milestone. It does not exist.
- Superintelligence (ASI, or SI) would be an intellect beyond the best human minds in practically every field — thinking faster, and in directions we could not follow. It does not exist either, and serious researchers debate whether it should ever be built. Rogers walks through those arguments, including the alignment problem; her piece is the deep end, and it is worth your time.
Every tool we deploy, teach, or evaluate sits on the first rung. All of it. No vendor demo, however polished, changes that.
The confusion is not harmless
It would be easy to shrug this off as Washington vocabulary drift. It is not harmless, for three practical reasons:
Buying. When a pitch deck says "superintelligence," a buyer hears a capability that does not exist — and budgets, schedules, and staffing get built against it. We teach a line in AI 101 that applies verbatim here: confident wording does not mean the claim is right. A term is not evidence.
Planning. Last month we wrote about Michigan's workforce numbers, and the discipline there was that different measures cannot be added together just because they sound similar. The same discipline applies to technology categories. A training program designed for narrow AI — the tools that exist — looks nothing like one designed for a hypothetical general intelligence. Plan for science fiction and you misallocate real money, real curriculum, and real people's time.
Oversight. Governance fits the category of the thing governed. A narrow tool needs what we already know how to do: access controls, verification, a human accountable for the output. Blurring the category cuts both ways — it makes people fear the tool they actually have, and under-prepares them for systems that don't exist yet. Precision is what keeps both errors out.
What we are changing in our own materials
Starting now, our website and our AI 101 course materials teach the ladder explicitly, and our terms are fixed:
- AI stays — for today's real technology. It is the word everyone knows, and for the first rung it is accurate. We are not going to confuse working people by renaming the tools on their desks.
- SI (superintelligence) is used only for the third rung: the theoretical category, discussed as such. Where public usage is drifting — and it is — we will say so plainly rather than quietly swapping words.
- The course keeps its name. AI 101 teaches what AI is; it now also teaches, on the record, what AI is not.
This post is the reference for that choice. If the vocabulary shifts again, we will update the materials and say why — the same way we would correct a number.
The takeaway
Words are the first system anyone engineers with. Get them wrong and the error propagates through every decision downstream — which purchase, which policy, which training plan, which fear.
So here is the whole discipline in one question, free to steal: the next time a pitch, a headline, or a press release says superintelligence, ask which rung? If the speaker can't answer, you have learned something important about the pitch.