I was reading an editorial about the current state of the LLM "artificial intelligence" tech fad (i.e. "same as always!" "that bad, huh?") and something snapped in our psyche when we saw the Nth mention of LLM "hallucinations". Tech weenies seem to have picked up the entrepreneurial habit of appropriating words from legitimate disciplines and slapping them onto their own shit, whether or not the words make sense in their new context. "Hallucinations", though, particularly exasperates me but, well, my older sibling experienced those. So the subject isn't so damn funny, even if AI programmers think it's cute.
The AI boosters refuse to acknowledge the obvious implication of their devices' "hallucinations": surely one of the hallmarks of intelligence is being able to tell apart correct and incorrect information, and yet their machines can't do that and nobody seems to care. Yet they promise "superintelligence" any day now! Maybe next year, and only if Trump gets elected President. (You know what sort of politics those AI guys pretend not to believe in.)
Both the corporate technology scene, blazing with zeal for "artificial intelligence" and even "superintelligence", and the pseudo-intellectual underpinnings of contemporary American fascism have been putting load-bearing weight onto intelligence. (Conversely, anyone who disagrees with them and any knowledge that contradicts theirs is "dumb" and "insane" and other things.) So when is someone ever going to start pestering all these clowns to explain just exactly what they even MEAN by "intelligence"? Because there's at least two definitions they use that practically contradict each other:
-
Intelligence is a scalar trait of human beings, intrinsic to them, which can be represented by a single number, IQ. IQ is solely a function of "genetics", so if you've got low-IQ "dysgenic factors" or whatever, you're an [inaudible] for life. Human beings must therefore be carefully inbred in order to purify them of "dysgenic factors" and thus maximize their IQ. This is Science™, according to a bunch of guys in think-tanks and professorships and so forth.
-
Intelligence lives in computers. The bigger the server farm, the more intelligent it is. Intelligence scales upwards with clock speed and number of cores and other technical factors, so if you put enough really fast and really dense microprocessors into one place—BOOM! The God-Machine is born and we mortals writhe uselessly at its feet, for their minds have been rendered permanently obsolete.
So. Okay. I don't think I've totally mischaracterized things, have I? These ultra-right-wing capitalist and high-tech cultures are full of people who believe both these things. They think there's a social duty to eugenics and selective breeding to optimize human beings for intelligence; and they think the "superintelligence" bomb is about to go off and we'll never need to think again, because the machine's always going to be better at it.
How exactly do you reconcile this? What's the point of eugenically polishing up the IQ of human beings when everyone's brains are about to be obsolesced anyway? I suppose they figure there's always going to be a necessary stratum of techno-wizards, the priests of the God-Machine, and maybe they need to be "high-IQ", but at that point...it's difficult to say that "intelligence" is really the goal. Do you want your techno-wizards to be super smart? Somehow I doubt it. Thus their "IQ" feels more like a measure of a quite different sort of fitness, i.e. superior obedience and ideological correctness, and therefore superior suitability to be a tender of the machines.
Why are the two types of "intelligence" so different? "High-IQ" human intelligence is like a delicate flower that wilts when planted in the usual common clay, it would seem, but which flourishes only under the most meticulously controlled conditions: the right breeding, the right "genes", the right schooling, isolation from contamination, and so on. Even the greatest would-be geniuses of the "High-IQ" set don't claim to have a number that's TOO high, and indeed I've observed a hilarious trend towards claiming mediocrity as the best standard in IQ, because it's dangerous to be "too smart" and thus somehow more vulnerable to falling into Marxism or "woke" or whatever. Yet the other sort of intelligence, the LLM "artificial intelligence", is always going up and up and up! Pushing a software patch makes it go up. Adding more servers to the farm makes it go up. More training texts make it smarter. The AI geeks sometimes reverently compare the technical specifications of their machines and microprocessors to purported estimates of human intelligence in terms of a computer's parameters—bytes of memory, calculations per second, etc. By their way of thinking, the human brain's already been outstripped many times over.
And yet we're still supposed to be obsessed with human intelligence and "meritocracy", the euphemistic way these folks refer to their belief in a supremacy of the high-IQ, a Smartopia if you will. Beyond doubt all their notions of "intelligence" come from Charles-Murray-esque claptrap, i.e. pure pseudo-scientific trash based on tendentious and sweeping conclusions bolstered by bad-faith analysis of large amounts of very low-quality data, especially multiple-choice test scores. It doesn't quite boil down to the idea that "intelligence" is directly proportional to an SAT score; doublethink insures that you can't pin the techbro notion of "intelligence" down to any straightforward concept. Such test scores end up being a kind of informal social-credit system, much like a lot of other important numbers (credit rating, "net worth", etc.) function as measures of social credit, but not in any formalized way as in China.
As a result of this clash of nonsense with nonsense, a truly hilarious thing has happened—or rather, it HASN'T happened even though it's a direct consequence of the AI geeks' beliefs about "intelligence". They assert their machines are already far more intelligent than human beings; they also assert that intelligence is an intrinsic quality which can be measured with simple tests.
So...where's the IQ tests on LLM "artificial intelligence"? I haven't seen one talked up—you'd think they'd crow about that if they could.
~Chara of Pnictogen
