That dangerous supplement

Precisely 303 must-dos, must-tastes, and must-tries

Perhaps no one would have noticed if the now infamous “Heat Index” supplement distributed by slop syndicator King Features to the Philadelphia Inquirer, the Chicago Sun-Times, and other newspapers had reached only its intended audience, people who, for reasons of vanity or self-delusion or sheer physical habit, like to go through the dignified motions of “reading the paper,” briskly and determinedly turning pages and diligently scanning over columns of words (Hmm, what must I taste this summer?) while actually taking in nothing more than maybe the display copy and, of course, the ads, the reason such supplements exist.

Typically when a publisher sells ads, they add edit pages to accommodate them, though “edit pages” is an ironic term to use in this case, where no editorial oversight occurred: As this Slate piece notes, the Chicago Sun-Times admitted that “we falsely made the assumption there would be an editorial process for this,” apparently washing their hands of their own responsibility for what their publishers and sales reps presumably make them publish. These edit pages shouldn’t really be understood as an enticement that will get readers to look at the ads; they are scaffolding necessary merely to hold the ads in place. If anything, they serve more as the reader’s alibi for getting to consume the ads, and more broadly, as an alibi for the whole project of marketing, which must preserve the idea that people engage with things because they actually want to, if only to set a benchmark for what marketing then sets out to corrupt and override. Otherwise, people would be like marbles in those wooden labyrinth games, with marketers steering them left and right and up and down until they drop through a hole and the meaningless process starts all over again.

In supplements like the Heat Index, ads are the primary content, and the editorial “content” is a kind of simulation that is meant to distract everyone from recognizing that. I don’t think this is an especially cynical view. People like it when ads give them ideas about how to live, as long as they can disavow that process. (This is the function not only of edit pages but of the human advertorials known as influencers.) Marketing assumes that people can be convinced to do things they otherwise wouldn’t do particularly when they can remain convinced that they are inconvincible; the pretense that one is looking at a publication not for the ads but the content helps sustain that susceptible mood.

This calls for content that doesn’t ask to be read, produced by writers who don’t care if they are appreciated, who are deliberately absent from their own texts. There have always been people at the fringes of the publishing industry who are willing to produce this sort of copy, for money or for further opportunities down the road to maybe even write something less disposable, but it’s not surprising that they would be quick to adopt LLMs to generate the text that is not supposed to be anything beyond generic and predictable, whose whole point is to be soulless. And the “editors” for this material — the ad-sales people who have called it into being — probably prefer it that way. If you were an advertiser, why wouldn’t you demand that the places where you advertise use LLMs to make the personality-free, “brand-safe” copy that can be counted on to never interfere with the consumer’s appreciation of the surrounding ads?

The Slate piece contends that “the A.I. slop of the Heat Index tells us much about the declining standards of print journalism,” but the supplement could also be understood as service journalism’s apotheosis. The standards of “lifestyle” sections have never been those of journalism but of entertainment; their purpose is not to inform readers and assist them in their citizenship duties but to indulge their fantasies and treat them as easily manipulated marks (i.e. people who receive pleasure easily), and to encourage them to enjoy being addressed that way as a kind of flattery. LLMs optimized for sycophancy are especially useful for this, capable of relentlessly love-bombing some hapless interlocutors into a compliant stupor.

But for how long could we possibly find it flattering to be barked at from all sides by sycophantic machines, especially when it has become common knowledge that nothing at all goes into the production of all that flattery, not even the suspension of some hack writer’s ego? How soon before language stops generating the illusion that there is necessarily someone speaking it, and it turns into a stagnant pool of signifiers?

The current generation of “AI” has frequently been promoted as a kind of trickery that wise early adopters can use to get over on the laggards who still anachronistically believe in “communication” and “empathy” and other such nonsense. Why not fool the ones you “love” by getting an AI to “care” about them for you? What is “love,” anyway? Does anybody love anybody anyway?

The presence of AI is a sure indicator of a more profound absence of human consideration. Thought is not a private act in an windowless-monad brain but a token of implicit public commitment, a kind of attention that has been devoted that seeks recognition or reciprocation, that evinces faith in the existence of other consciousnesses that can share it. Thinking entails thinking about something, for someone, it posits a faith in a social relation. LLMs are set out to negate and stifle that faith. Where human thought inherently marks the recognition of other beings capable of their own subjective intentions and interpretations of things — human thought is a gesture of recognition as well as a “content” — mechanized simulation of thought is a deliberate refusal of that gesture, an elision of that recognition, an attempt to deny the other’s subjective consciousness and render them a kind of manipulatable content, a transformable output among other outputs. It is premised on the idea that other people don’t deserve attention but should be fooled into surrendering it along with their agency.

The LLM-generated Heat Index supplement is a somewhat low-stakes example of this, but it exemplifies how AI is fundamentally a technology of disrespect, of giving nothing where something is commonly expected among people trying to live as a society. But there is another way of thinking of the supplement. In Of Grammatology, Jacques Derrida develops his concept of the supplement to capture a paradox he identifies in the writings of Jean-Jacques Rousseau. In Derrida’s analysis, Rousseau sees writing as a degraded form of speech, which itself falls short of capturing the essential thing, the real presence of a human being. Writing ends up being a double-edged attempt to address that failing in speech: as it attempts to supplement speech’s ability to evoke presence, it also supplants speech, replacing and negating it.

As substitute, it is not simply added to the positivity of a presence, it produces no relief, its place is assigned in the structure by the mark of an emptiness. Somewhere, something can be filled up of itself, can accomplish itself, only by allowing itself to be filled through sign and proxy. The sign is always the supplement of the thing itself.

Using language to try to capture the “thing itself” is always a rowing against the tide: “an infinite chain, ineluctably multiplying the supplementary mediations that produce the sense of the very thing they defer: the mirage of the thing itself, of immediate presence, of originary perception.”

In Derrida’s account, Rousseau regards the doubleness of supplementarity — the fact of “mediacy,” that “real presence” is only ever known to us through an infinite chain of representations, through a kind of streaming media — as “evil.” He quotes from Rousseau’s educational tract, Émile:

As soon as [children] can think of people as tools that they are responsible for activating, they use them to carry out their wishes and to supplement their own weakness. This is how they become tiresome, masterful, imperious, naughty, and unmanageable.

For Derrida, this suggests that for Rousseau, “the supplement will always be the moving of the tongue or acting through the hands of others.” It marks a “regression toward an evil that is not natural and that adheres to the power of substitution that permits us to absent ourselves and act by proxy, through representation.” This perspective is readily applied to “AI”: a proxy actor that enables absence and obfuscates presence, that allows for exploitation beyond the “natural” tendency of humans to care for one another that asserts itself when the conditions of true presence are maintained. (Derrida also develops this analysis in light of Rousseau’s shame over his masturbation habits.)

The point of Derrida’s analysis is not to confirm Rousseau’s sense of “presence” as essence, and writing and textuality as evil indications of our having fallen away from it; instead, he makes his famous claim here that “there is nothing outside of the text” and tries to describe an approach to texts — eventually known as “deconstruction” — that would help prevent us from falling into the metaphysical trap Rousseau is in.

What does any of this have to do with the Heat Index supplement? It seems easy to take a Rousseauistic view of it as I did above, lamenting of how certain kinds of empty texts permit a kind of exploitation and dehumanization that is contingent on the suppression of thought and the human relations it implies. As automated text generators flood the world with ever more textuality, there will be ever more absence in the world, and it will become increasingly impossible for presence to announce itself, to make itself known and presentable.

But Derrida emphasizes that this has always been the case, and lamenting it does nothing but perpetuate that impossible chase for pure presence, for things that are finally and completely authentic. Derrida’s guarded and hedged suggestion of an alternative is presented, as he says, in an “entirely negative” way, but it doesn’t consist of salvaging “true presence” or the “real voice” or anything like that — nothing is ever sufficiently pure, things only seem more and more mediated, fake.

The fact that LLMs can generate endless amounts of explicitly “fake” copy with the traces of human intention and presence deeply diluted through countless layers of processing and concatenation could hopefully demystify not only that particular subject position that seeks safe harbor in “real texts” — i.e. an alibi in a “real supplement” for the dubious pleasures such supplements have always supplied — but also the fantasy of accessing perfect authenticity through media.

It seems likely that promoting and publicizing the kinds of tricks and shortcuts and opt-outs from the demands of being a human being that tech companies promise will engender a universal skepticism of the low-effort gestures that are now possible, raising the bar for what counts as care and increasing the value of human attention — when you can manage the increasingly arduous process of establishing its genuineness. AI could thereby inadvertently help further expose and discredit all the automaticity and instrumentalism that had already crept into social relations with capitalist modernity. But that doesn’t mean that the “real paper” will ever be rescued from its supplements. We still have to pursue a different way of reading everything.