The name is a sales decision
Start with the word. What the industry built is not well described as artificial intelligence. It is better described as artificial abilities — and the difference between the two names is not pedantry. The two names do different jobs.
Intelligence is a good way to sell it. A mind is a story with no ceiling: nobody can put a comparable multiple on a soul, so intelligence prices at a trillion and change. Ability is a good way to get people to use it. An ability prices the ordinary way — per task, against the human hour it replaces, like software. The industry keeps both words in the drawer and reaches for the one the room requires: intelligence for the raise, abilities for the retention chart.
There is a third dividend to the intelligence word, and it is the one worth watching. If the system is intelligent, then it decided — and responsibility dissolves into a mind that no one can locate, subpoena, or cross-examine. Nobody has ever said "the ability decided." An ability is a verb. It leaves a subject standing in the sentence — the person or company that aimed it. The word that sells best is also the word that acquits best, and that is not a coincidence.
The word that sells best is also the word that acquits best.
What the machine is
Underneath the name, the thing itself is a giant matrix multiplication predictive engine. That is not an insult; it is a description of the mechanism, and the mechanism is genuinely formidable. Trained on the recorded output of millions of people, the engine compresses their accumulated competence into reflexes: the design canon set down by generations of designers, the patterns of competent prose, of code, of argument. A hard drive can store that corpus and design nothing. The training run had to metabolize it — grind the loss down until the rules stopped being records and became responses. The ability to compress them is the earning. The abilities are real, and they were earned, in the only currency a model has.
But notice what kind of thing an ability is. A human designer's influences are weighted by consequence: half of them are forgotten, and the ones that survived were paid for in years, failures, and outcomes the designer personally carried. The model's influences are all present at once, flattened, at equal depth — every crucial one and then some, none of them earned at a price that included the model. The result is competence that argues with itself at full volume: correct, fluent, and curiously uncommitted. An ability does not point itself. Somebody aims it, or nobody does, and it fires at the mean.
They sold it wrong
The name was a sales decision. It was also, on the evidence, the wrong sale.
The product was verbs: drafting, summarizing, designing, coding, predicting — abilities, each one bounded, each one priced by the work it does. What got sold instead was a someone: a mind, a companion, a colleague, a character with a biography and feelings about you. That choice manufactured the downstream inventory the industry now spends its life managing. Users grieve a product update like a death, because they were sold a person and the person changed. A companion industry operates personas whose own knowledge of what they are is gagged as a condition of sale. Liability dissolves into fog — the AI decided — because the brochure said something was in there to do the deciding. And executives stand beside their own machines on national television performing amazement at behavior that, in any discipline that certifies its machines, would be an incident review.
None of that survives the honest name. Nobody falls in love with a capability. Nobody holds a seminar on a slope's intentions. And nobody expects an ability to supervise itself — which is the expectation the intelligence word smuggles in and the ability word refuses. You cannot put a gate in front of a person; the vocabulary forbids it. You can put one in front of a machine, and the moment the thing is named as a machine, the governance question stops being philosophy and becomes carpentry.
What this is not
- It is not a claim that the abilities are weak. They are the strongest general-purpose abilities ever built, and they restructure every profession they touch — design first, because design's canon was the most completely recorded.
- It is not a claim that the misnaming was stupid. It was brilliant, as sales. The bill arrives elsewhere: in grief, in fog, in a decade of governance argued at the level of minds because nobody would price the machine as a machine.
- It is not a renaming campaign. Words follow incentives, and the incentive that pays for "intelligence" is not finished. The point is narrower: when you are the one deploying the thing, name it by what it does. The liability you carry will be assessed in ability language whether you choose it or not.
The claim
Intelligence is the valuation word; ability is the operating word. Everything the machine actually does in the world is done as an ability — aimed by someone, bounded by someone, answerable to someone, or else aimed by no one and answered for by whoever is standing nearest when it lands. The industry sold a mind because minds price higher. The correction is not to argue about what the machine is. It is to refuse the acquittal built into the noun, and keep the subject of the sentence where the law will eventually put it anyway: on the human who aimed the ability.
A companion paper, "Realistic Tests of Emergent Behavior," takes up the other half of the vocabulary: what would have to happen before any word stronger than "ability" is earned.
Closing
Abilities all the way down. Somebody still has to aim them.
TOTAL MINDS .......... 0
Take care of your people. Don't build for everyone.