84. Measurable Cognitive Factors Contributing to General Intelligence

(Epistemic status: taking a framework other people have started on and then abandoned; actually kind of predictive. Probably not complete, but the frame seems unique to me and I want other people to think about it. For NC, JM, and PR.)

 

Intelligence! A famously hard thing to pin down and operationalize. The part where people often assign it some kind of moral or valuative dimension - in both directions - and then allocate universally vital resources based on who seems to have it and who doesn't makes that even harder. The part where people either deny that it has a genetic basis and thus focus on misguided environmental schemes or consider the predominant genetic basis to be so all-important as to try to push breeding programs based on it makes it harder yet. Worst of all, some people even go as far as to deny that "intelligence" in the abstract even exists as a meaningful property of a person let alone a measurable one, or go heinously egregiously wrong and try to tie it to ethnicity or culture. We believe in the mythical g in this blog, in a directionally-correct sense if not more so. That said, aspiring skull-measurers need not apply, because we also believe in a larger intra-ethnic variation than inter-ethnic and also the humble author is mixed-race. (That's hybrid vigor for you.) But there's still something there: not everyone can do a doctorate in pure math, but at the same time it would be wrong to say that it requires more intellect than writing a coherent and moving novel, planning a military operation in depth, or understanding people well enough to charm and convince them.

Some attempts to define and measure intelligence try to break it up along different capabilities-ish lines: visual intelligence, spatial intelligence, working memory, reaction time, capacity for verbal abstraction, depth of lore, capacity for computation, social-emotional, and so on. This is largely not about such fine-grained abstractions. Rather, the framework I propose is based on qualitative judgements of cognition.

As I see it, there are five major loosely measurable aspects or factors to cognition:

• Thrust: The closest parallel to classic measure of IQ or something between processing speed, fluid reasoning, and working memory. Having high thrust means having lots of raw cognitive power, and being able to make single straightforward isolated deductive steps perfectly at speed. It means being good at grinding through masses of data and finding patterns in it, and being good at things like memorization, symbol manipulation, visualization, and keeping hold of ideas, frames, and abstractions.
• Sparkle: Something like creativity or lateral thinking. Having high sparkle means being able to make large intuitive leaps in reasoning and landing safely. It means being good at finding unexpected or unconventional answers, or at applying existing knowledge in a novel way. It means generalizing well and being comfortable with induction.
• Quash: If sparkle is about having few false negatives, quash is about having few false positives. Having high quash means being good at suppressing instinctive wrong answers, at blocking out an overeager System 1. It means being able to shut down lines of inquiry unlikely to lead anywhere helpful or productive, and at keeping yourself firmly on track when you reason. It means being firmly grounded in the unambiguously true without getting distracted.
• Reach: If thrust is about getting short obvious things done quickly and well, reach is about how well you can weave those steps together into a larger and more significant whole. Having high reach means that you can avoid losing much fidelity when thinking in terms of complex and heavily interdependent chains of reasoning. It means being good at thinking in systems, and in being able to hold one realization in mind while digging for a second, and then a third. It means flexibly repeatedly swapping what you're grinding through, juggling assorted different frames fluently without letting any plates fall.
• Bounce: Surprise! Everything you know is wrong, or your attempted reasoning is full of errors! Can you get yourself back on track quickly and without losing too much? If you have high bounce, you can. It means being able to bughunt and carefully isolate and correct errors in your reasoning or understanding

Notably, none of these have any intrinsic moral dimension to them. It's not even guaranteed to be a good thing to have lots of any given one of these; what matters more for quality of thinking than the absolute amount of any of these - if they were even easily measured quantitatively - is their rank-ordering and the ratios between them. Do you have more sparkle, or more quash? That will say something about whether you seize overly fast on wrong answers or instead shut down inadequately conventional lines of reasoning with gaps that feel too large on the way to the right answer. Do you have thrust vastly in excess of your reach? Then you might end up grinding away fruitlessly without ever thinking much about the broader picture.

Continuing to take this frame seriously, we might operationalize Intelligence as something like the minimum "score" among all five of these factors, plus some terms for the other scores which rapidly decrease as the difference from the worst one grows. Again - if you have high bounce and low quash, you're going to be good at catching and fixing your errors, but you'll make so many of them - assuming you've got reasonable sparkle - that you'll get overwhelmed. On the other hand, being reasonably well-rounded with fairly good scores on all five means being overall intelligent - especially prone neither to false positives nor false negatives, able to quickly and precisely reason in a way that chains into larger insights with enough mental workspace and capacity for abstraction to make that reasoning useful, and to detect and correct for those inevnitable errors that you end up making anyway.

A reseach mathematician might have excellent thrust and sparkle, strong reach, and respectable quash and bounce. By contrast, a world-class diplomat might have excellent bounce and reach, strong sparkle, and merely adequate quash and thrust. Where it gets especially interesting in the the modern day is what the frame has to say about frontier LLMs: in my experience - mostly with Claude 4.x Sonnet - is that they have world-class thrust, strong reach, adequate sparkle, barely adequate quash, and bounce that's all over the place depending on context and topic, but often not great either. Notably, this kind of incredibly lopsided spread is not at all what most human minds look like, and this seems suggestive of why LLMs can seem so smart in many notable ways and yet so terribly lacking in others. I think that this makes my ontological frame a promising one, if only to build on, for understanding the nature of intelligence and cognition.

Give it a try yourself. How do your friends stack up here? Or your family members? Or your partner(s)? Do some tests, or think through how much of each of these five factors you yourself have or others seem to. (If you're reading this blog and have understood all this, you're so clearly high in thrust and reach!)