Intelligence Is Not a Scalar

There is a quiet condescension in the phrase 'more intelligent'. It assumes a single axis, a ladder everyone is climbing, to the top. In the work of evaluating people and systems — which I have done for most of my career — I have watched that assumption do real damage. Almost nothing important about a mind reduces to a point on that ladder, and yet the ladder is how we distribute authority, funding, and trust.

A scalar is a number with magnitude and nothing else — no direction, no dimension, no context. To treat intelligence as a scalar is to insist that every mind can be placed on one line, and that the only meaningful question about two people is which of them sits further along it. We rarely say this aloud, because said aloud it sounds absurd, and condescending. But we build institutions on it — admissions tests, hiring funnels, benchmark leaderboards — and institutions say what we are too polite to.

The instruments, not the ladder

The peasant who reads the weather in the sky and the theorist who reads it in equations are not two rungs of the same ladder. They are different instruments, tuned to different worlds. Collapse them into a ranking and you have not measured intelligence. You have merely decided whose world counts.

I have made this mistake and paid for it. Early in my career I would have ranked a room of colleagues confidently, top to bottom, by how quickly they could update system or a model. Years later I watched some of the people I had quietly placed near the bottom of that list save projects that the people at the top had confidently steered toward a cliff. Their intelligence had not been low. It had been pointed in a direction my ladder could not measure — toward patience, toward doubt, toward the unglamorous question no one else would ask. These brave practitioners had quietly shown the courage to question the agreed upon rationale and created friction for themselves.

This is not a romantic point about honoring folk wisdom. It is a structural one. Intelligence is better understood as a vector — a bundle of capacities pointing in many directions at once: pattern recognition, patience, the tolerance of ambiguity, the sense of when a problem is not yet worth solving (say re-Gantt of an Implementation Plan). As an Oracle practitioner, I am quite surprised that we loudly applaud Oracle's AI Vector Search capabilities that allows to combine different types of data across script, relational and json to derive an answer, while we oppose the idea of using 'vectorized' approach to measure our own intelligence. We need to think about intelligence as a vector in our general intelligence and in building our future AI workflows - it should be immersed in lived experience, rounds of experiment; it should be able to argue the counterpart, and should be proved through empirical evidence of representative data.

Every ranking of minds is, underneath, a ranking of what we have already chosen to care about.

Why the fiction persists

If the scalar is so obviously false, why does it hold? Because it is convenient for whoever is doing the sorting. A single line makes allocation easy. It lets a committee compare a thousand applicants without confronting the incomparable, lets a market price a mind, lets a benchmark declare one system the winner, lets a department planner sandbag their forecasts year over year, lets a solution architect propose the same tool to the marketplace owing to its current stickiness in the industry at large, to name a few. The ladder does not survive because it is true. It survives because it is intellectually cheap, rewarded for its standard measurement capabilities and is paraded as time saving. The ideas of time-saving and rewards, once institutionalized, are very hard to dislodge.

The cost is paid quietly and by other people. Every genuinely different way of thinking — the slow one, the lateral one, the one that asks the unwelcome question — gets read as a lower position on the shared line rather than as a different line entirely. We just do not just fail to reward these minds. We fail to notice they were minds at all.

I am not asking to abandon judgment. We do have to choose, to trust some reasoning over others. It is to keep the choosing honest ����� to admit that when we rank, we are not discovering an order that was already there. We are imposing one, in the service of something we have decided to value. In the world of AI, there is a term called 'Objective Function' that defines what 'success' or 'errors' look like for a model. There are minimizing functions - reduce error or costs, and maximizing functions - increase accuracy or rewards. Key is to provide AI what we value. Naming that value is the beginning of intelligence about intelligence. If it applies to AI where we can go through rounds of inquiries and questions to arrive at an objective function, then let's stop pretending there is a natural scalar ladder for human intelligence and intelligence at large itself. When we say it is natural; it is where we stop thinking.