As the accomplishments of “artificial intelligence” take up an increasing portion of the news, it becomes apparent that we in the modern world are suffering–yes, suffering–from a cult of intelligence, or rather of intelligence so-called. More and more it becomes apparent that this cult owes its power, not to intelligence per se, but to our not having at all grasped what intelligence even is.
Intelligence, more than any other personal quality, is seen as the index of one’s whole potential, the best predictor of success for individuals and groups alike. The modern world itself is an IQ test, some tell us. We live in a “knowledge economy”. We have abandoned military and religious heroes in favor of Einsteins, Edisons, and (perhaps) Musks. Parents, especially the well-to-do, want nothing more than to have the most “intelligent” children possible, and will often spare no expense to secure anything that promises to confer either the substance or the appearance of brilliance.
Who would dare consider, given this, that what we now almost ritually deem “intelligence” may just as well be a form of limitation as liberation, and may indeed prove itself to be extraordinarily petty, if not deceptive?
Yet, “intelligence” as now conceived–if the mavens of “AI” bother to stop and conceive it at all–has more and more of the qualities of a snare for the development of mind and intellect than of a means to new fulfillment of their powers.
This concern may be applied, surely enough, to other aspects of modern society that are held up as proof of our triumphant “intelligence”–not just computational or strictly scientific approaches, but social and mental habits. Nevertheless, we find very often that talk of “AI” is situated at the core of these aspects. It represents, in some way, a “blueprint” for the entire activity of the rest of the cult.
Yet as the cult of intelligence advances, it becomes more and more of a cliche, more and more of an unconscious tic–more and more like fashion. the praise of “intelligence”–artificial or not–everywhere seems more than anything else, to be a praise of repetition, of sameness.
For example, consider The Fourth Industrial Revolution, a 2016 book by World Economic Forum chairman and self-styled futurist Klaus Schwab’s. This and similar publications by WEF and other purportedly path-breaking, forward-looking, and of course extremely intelligent organizations extol an impending future of near-limitless innovation and novelty–a true revolution.
Yet throughout The Fourth Industrial Revolution, one is most struck by how strangely monotonous the proposed “revolution” actually appears. Notably, nearly all of the dozen-plus hyper-disruptive “deep shifts” Schwab heralds at the end of the book boil down to one basic trick, repeated over and over: namely, that of placing tiny computers (that is, “AI”) in every conceivable object, animal, or person, in order to continually monitor and log their minutest acts.
Taking the undoubtedly high-IQ WEF view, the culminating glory of our uniquely intelligence-obsessed culture apparently will amount to an eye-wateringly convoluted bureaucratic meshwork of remarkably gratuitous digital control and surveillance systems.
This strange combination of cognitive horsepower and complete monotony is hardly limited to the WEF and its devotees. In a more general vein, to remain in fashion and be considered highly “innovative”, one simply appends “internet of…” to some vague, portentious-sounding noun to generate ready-made “innovations”: thus “Internet of Things”, “Internet of Bodies”, “Internet of Everything”, and so on.
In the hands of the cult of intelligence, innovativeness and intelligence, it turns out, have become formulaic, algorithmic–which is to say, not very innovative or intelligent at all. The curious paradox of stagnation in everyday life combined with exponentially intensifying hype radiating from “intelligent” elite formations such as WEF epitomizes the strangenes of the situation. Something in the whole conception of what “intelligence” is, who has it, and where it should lead us has begun to overheat and malfunction. The result is a sense of widening absurdity, and a growing frustration with precisely those figures society is supposed to look up to and take its cues from.
Contemplating at the intellectual landscape of the past half-century, we seem to have become enthralled by a force that, for the intellectual world, is closely analogous to what Matt Taibbi’s vampire squid represented for the economic world. It is as if some cosmically greedy yet totally unimaginative demigod from the Outer Void has steadily tightened its grip upon our world of thought, trying to suction up everything it can possibly find, squeezing it all down into some standardized format (“data”) while evacuating it of nourishing content.
Much as the “vampire squid” in 2008 decimated the real economy, replaced it with a worthless subsitute–and expected the right to rule the financial system, so this would-be god of the intelligence-cult, standing amid the exanguinated remains of countless once-vibrant ideas, sentiments, personalities, and ways of life, now unveils a series of just-passable simulations of everything it just drained of life, and declares that this act of simulation proves its supreme “intelligence”, and its right to rule.
Our project–which may not be feasible at all–must be to defeat this god, and restore the position of true intelligence before the last examples of it are turned into simulations. Otherwise, the result must be an intellectual “financial crisis” which, by crippling not the modes of material value but of cogitation itself, will be much worse than the monetary crisis that emerged in the late ‘aughts.
* * *
At the heart of today’s AI craze, this paradoxical sameness and stagnation manifests as an obsession with mathematical, and specifically statistical thinking. Under the combined force of many decades of postmodernism and managerialism, we have arrived at a stage where “data” can not merely approximate reality, but actually replaces reality (“it from bit”). If only we chase relentlessly the limit-case of large numbers, says this philosophy, reality itself can be manufactured to taste. Thus, data must become “big”–and bigger is always better.
This map of “intelligence” has been most imposingly manifest recently in the “large language models” (LLMs) such as ChatGPT. With the branding of such huge statistical models as “artificial intelligence”, the statistical has been posited as the apotheosis of, if not successor for, intelligence itself.
Yet the same dilemma continually reappears: the statistical, by its very nature, cannot deal with truly unique cases; yet it is the unique cases that not only make life interesting, but, by requiring us to handle truly new things, reveal the actual range and power of our intelligence. This ability to handle or explore unique or “edge-case” situations without collapse is what separates intelligence from intelligence as conceived by the present computational-statistical cult.
As more and more would-be LLM users are discovering, the gaps resulting from a statistical “intelligence” system based on the complete abjuration of structure, principle, and uniqueness can be grievous. In one example, when asked “what sex will the first female president of the USA be?”, ChatGPT strains its hundreds of billions of artificial “neurons” and its terabytes of trawled Internet text and concludes that it is impossible to say in advance.
Similar examples, it turns out, abound in LLMs’ output; blatant non-sequiturs and outright fabrications appear time and again (in one case causing Google’s stock value to plummet); in other cases, the LLMs, when pushed, show highly inappropriate or disturbing behaviors.
It is something of a cliche that there are “lies, damn lies and statistics”. But it turns out that what distinguishes the LLM, as the apotheosis of statistical “intelligence”, is in many cases not the truthfulness of its output, but the meretriciousness, mediocrity, and mindlessness of that output. More data and more compute leads, after a certain point, to more “intelligent”-sounding language output, yet with diminishing improvements in actual insight, truthfulness or–to use the technical euphemism now current–“calibration.”. As one Margaret Mitrchell, chief ethics scientist at the A.I. firm Hugging Face, put the situation, LLMs “are not trained to predict facts […] They’re essentially trained to make up things that look like facts.” Or in the still-more-pungent words of anothermachine-learning researcher:
“…ChatGPT can not really understand the underlying semantics of the sentences it spits out. This can always be seen in the ways it fails and makes complete nonsense up, which happens very frequently […] In reality chat GPT is mostly nonsense producing machine. It cannot be relied upon for accuracy or logic. It confabulates non existing sources and makes up self contradictory statements.”
This is the epitome of semblance over truth, surface over substance. In true postmodern manner, the LLM implicitly excludes the “metaphysics of presence”: there are no essences, no meanings (beyond statistical correspondence), and nothing is consistent even with itself, hence truth is just another kind of textural pattern to be imitated.
(This proclivity for style over substance–in fact no substance–reminds us of so many well-salaried and widely-praised scientific and managerial “experts” of our day, the veritable priests of the cult of intelligence. Their mistaken predictions, questionable or hypocritical conduct, counterproductive decisions, false assurances, and general unreliability, howsoever numerous the examples thereof, seem to do nothing to dislodge the default presumption of their own brilliance. Could it be these experts are the only entities of society actually less reliable than the LLMs in their pronouncements?)
* * *
The long-term result of the cult of intelligence is, actually, a pervasive mediocrity, masked by computational fervor, that is passed off as vision and rigor but soon comes to displace both like a weed. It turns out that mediocity, not stupidity, is the true opposite of intelligence. Thus wherever some event or some question deviates too far from the training-set (which represents mediocrity), too far out on the tails of the distribution, or wherever edge- or corner-cases become even moderately numerous, this once-triumphant statistical “intelligence”–human or digital–collapses into incoherence, or confabulation, like a tromp d’oeil.
Here, on these edges and corners, these places well outside the “convex hull” where the training sets of “big data” peter out, the perennial war between the mean and the variancetakes place. The mediocre, the statistical–represented by the mean–dazzles us with repeated impressions of fluency, and would lull us into the assumption that semblance and reality must be interchangeable. Yet the unique, unseen, and unexpected–represented by the variance–regularly intrudes on this illusion, and from completely unpredicted directions, revealing ghastly faults in the statistical superstructure, shattering the system’s illusion of understanding.
One of the key “early” (it was really only 8-10 years ago) discoveries of the current rennaissance of neural-network methods involved something known as “adversarial examples”. These were images or other instances of “data” which appeared perfectly recognizable to people, but, when fed to even the most scrupulously trained networks, would produce wild misclassifications–an automobile being mistaken for a tree or a baby, for instance.
The adversarial example is the classic example of the variance’s unexpected proximity to the heartlands of the mean, its ever-present capacity for sudden intrusion and overturning of the mean’s well-ordered mediocratic regime.
But as it turns out, life itself is replete with adversarial examples–including in such seemingly innocuous tasks as driving. Like the dark edges of possibility where the variance makes its rebel hideaway in plain sight, these examples may be sparse, but they are far from scarce. There are more than enough of them to reveal, glaringly and routinely, the difference between real intelligence versus the statistical, mannequin-intelligence of the Cult.
The rebels are alive and well. Not only do they mock the mindless complacency of the mean and its adherents, but–there are uncountably many of them. In time, as the Cult continues blindly in its dream of total machinic triumph, this rebellion will coalesce into full revolution.
Peeling The Paradox 