AI is the worst
The fashion among proponents of AI is to squawk, rather parrotlike, about how the technology is, at any given moment, the worst version of itself that we'll ever use. According to them, the texts, images and code that AI spits out will improve with each iteration, to a point where its quality first becomes indistinguishable from humanity's best, then surpasses it. Every last dendrite of the human mind will be trampled upon, and from the wreckage of our grey matter a synthesised omniscience will emerge, to which it will be inappropriate to apply the term "artificial".
For these AI-AI-Oh!-ers, then, the endgame of AI is dropping the "A" and arriving at, simply: Intelligence, of a super-human kind.
I see a different endgame. Or, more accurately, I see two possibilities which, like a false fork in the road, eventually converge back into a single rut. Not to foreshadow overmuch, but this rut terminates at the place where all the world's books go up in flames. The Library of Alexandria, writ global.
A caveat: my remarks here are confined to the social, para-social and cultural spaces that AI is willy-nilly deconstructing, displacing everyday human ingenuity with machine-generated infill. In the areas where AI might actually improve people's lives — early cancer detection, say, or finding that stubbornly elusive next prime number, the better to encrypt our data with — I for one welcome our robot overlords.
A further aside: my estimation of that colossus of SA Lit, André Brink, rose considerably when I learned that he used Yogi Berra's aphorism, "when you come to a fork in the road, take it", for the title of his 2009 memoir. Do ensure there's a copy of A Fork in the Road in your own personal library, if you can, before it all burns down.
Fork in the road
At the fork in the road AI has led us to, let's pause and consider semantics. I believe the word "worst", as in "AI is the worst it will ever be", is a poorly-chosen pejorative. What the AI boosters continuously bruit is not that the technology will somehow become its "best" version one day — that is self-evidently ludicrous — but that it will become ever more optimised, infinitely refining and improving as it learns.
If there is no pinnacle, similarly, there can be no nadir: AI is not currently the worst version of itself, but merely the least optimised. It's still graced by the experimental; it still falls into lapses and traps; it still entertains its users' follies.
But all the while, at the behest of increasing the revenue and reach of the companies that own it, AI is learning to pre-empt experiments, fall less, and ignore more — a continuous striving for the most highly adoptable forms of output. To put it another way, AI's own commercial logic will dictate that it become more predictable, more homogenised, and more flattened out, for these qualities will guarantee its appeal to the widest set of users. This is a far cry from the technology’s initial dark promise, that it will surpass humanity’s skills in toto and become, i.e., the best writer of emails, the best painter of lilies, the best wave-particle theorist, the best EDM DJ, and so on.
Instead, to maximise for what might be termed the “broadest common denominator”, AI will eventually have to turn inward, processing fractals of optimisation that only it perceives. All the other competing AIs will follow suit, learning from one another as they churn data in service of their corporate KPIs, honing in on the spot on the horizon where the parallel lines touch. They will optimise themselves to a standstill, lost in the minutiae. We won't have superhuman intelligence; but we will have chatbots that are the best pickers of, i.e., shades of beige, for when we start that long overdue kitchen remodel and need advice on the cupboards.
Pabulum machine
What these AIs are inching toward, even now, is a networked, hyper-leveraged, and possibly self-aware pabulum machine.
Do you know how much energy is required to run an infintely-escalating, ever-optimising pabulum machine for millions of pabulum-hungry customers? Customers famished for it, stuck in their voracious meatsuits controlled by inferior wet organs — see how they strap in for eight to ten hours a day, boot up, and mainline the stuff?
According to Goldman Sachs, a ChatGPT query uses almost 10 times as much power as a Google search. The bank’s recent report on AI predicts that, in the next five years, AI demand will grow to about 19% of overall data centre electricity consumption, helping produce CO2 emissions "that will represent a 'social cost' of $125-140 billion".
That’s a lot of green. Meanwhile, remember three years ago, when UCT's library burnt to the ground?
In the close future, the fork in the road that the AI corporate pabulum machine is striding down will also, by dint of ever-billowing emissions, likely see one or more further libraries, at Wits, the University of the Western Cape, Stellies, Fort Hare, Tuks, the University of KwaZulu Natal, or Kovsies — or all of them — similarly apotheosise into burnt offerings made to technology-driven omniscience.
Of course, I can't say for sure that AI will be the direct agent completing the metamorphosis of libraries into insurance claim slips — correlation versus causation and what-what — but I know without question that more libraries will go up in smoke, and AI will be spotted in the front ranks of the arsonists.
Voracious meatsuits
Let's leave that thought for now, though, and visit the other fork in the road. Call it the fork on the right.
Speaking of voracious meatsuits controlled by inferior wet organs, does the name Ted Kaczynski ring a bell? Perhaps his other name, then: the Unabomber. The Unabomber was a domestic American terrorist who sent out manifestos and parcel bombs via the US Mail. He murdered three people, maimed 23, and died last year, having served almost three decades in prison. His entire ideological conceit centred on denouncing technology and the "surrogate activities" that humans are conned into pursuing because of it, so he is perhaps not the most felicitous example to use for this particular column. Still, he gained notoriety for lobbing hand grenades from society's margins, and in that sense, he fits the bill.
Imagine the Unabomber with access to custom large language models and social media, able to run an influence machine that produces and publishes manifestos faster than the speed of thought.
The good news is, you don't have to: we're well down that fork in the road already. All deranged misfits today have Unabomber-style extreme personas and the equivalent of anonymous 20th-century fax machines, but their respective manifestos are now hypercharged by prompt engineering, A/B testing, engagement farming, click baiting, the support of the tax-revolt libertarians who control the various digital "town squares", and the endlessly creative psyops tips and tricks that are freely available, courtesy an army of helpful Kremlinbots, whose activities now actually make up 78% of the entire Internet*.
*This is not true. But it feels true. Have you read anything online lately?
The new benightedness
Make no mistake, these deranged misfits are using AI to their full advantage, training the machines to create optimized versions of their conspiracy theories, even handing said theories over in their entirety. Just ask the AI to cherry-pick facts and coincidences from all sources available, then weave cusp-of-plausibility elaborations and outright fictions into the main narratives - et voilà! Your conspiracy theory is now an online siege engine whose torsion springs never wear out. All that's left is for the AI to "select all the tiles with motorcycles", venture into the online shallows and heave its buckets of slop, ad infinitum.
Just as with the pabulum machines, the conspiracy machines will inevitably optimise toward a single point. As facts, reasoning and depth of knowledge are anathema to conspiracy, but quite common in books, that point will be marked by a certain sign — a bright light drawing us onward toward a new benightedness, marked by conspiracy theorists' favourite activity, the literal setting aflame of all the pesky counter-narratives.
Our two forks in the road, then — neither less traveled by — have reconverged; and the blaze lies just ahead. Instead of infinite monkeys sat at infinite typewriters typing infinite texts, through AI we’re about to witness infinite monkeys sat at infinite typewriters typing the same text, fully-optimised, conducting a holocaust of other forms of thought. It might already be down to just six words. In the Age of AI, AIs chant together with one voice:
All books are good for burning. DM