ai

Take It To the Limit

By in ai, writing No Comments

Matthew Gasda:

What must be preserved from the encroachments of the calculator are forms of writing in which excellence was always an uphill battle: the long-form essay, literary fiction, poetry, plays, memoir, as well as personal letters. But even this preservation must make room for reasonable carve-outs. There’s a difference between a journalist who prompts Claude to generate a sentence, a paragraph, or a whole article, on the one hand; and one who asks Claude a factual question, on the other. The latter is doing something no different from consulting Google (Google results now are AI-driven anyway).

Norms can and should be developed within these professions for certifying organic work, which will no doubt command higher prices, just as handmade furniture from a carpenter comes at a higher cost than a chair from IKEA. If this means fewer people will be thought of as true writers, is that really such a bad thing?

I think we’re in the “bargaining” stage of the AI discourse. “OK, fine, you can have all of this writing. But surely you’ll leave me poetry, literary fiction, and personal letters?”

“Organic work commanding higher prices” strikes me as “capitalism wins again.” If some people get a thrill out of cultivating a New Yorker-type of relationship to writing, good for them. No different than if they spend thousands of dollars sniffing wine corks, even if I think it’s absurd. The human animal will always find a way to turn even a new status quo into a status game. Human writing becomes artisanal cheese. There will almost certainly be people for whom “human-written” becomes part of a cultivated relationship to language in the same way other people cultivate architecture, jazz, watches, coffee, typography, or restaurants. They may possess genuine discrimination. They may also be unbearable about it. Usually both are possible at once.

This is another instance where I wonder if the form of the piece doesn’t cause the tepid conclusion. It’s like: a writer has an insight about AI. He sets out his thoughts. But his thoughts don’t end on a summarizing note, and certainly not an uplifting one. So a “call to action” sort of ending gets perfunctorily tacked on, perhaps at an editor’s urging. Obviously, there’s no way to know. But it’s such a common occurrence that I wonder if it happens often. The conclusions are often so rote and uninspired.

The interesting part of the essay is exploratory. Gasda keeps discovering complications: the AI/non-AI binary is unstable; cognitive outsourcing long predates LLMs; writers have no principled reason to exempt their own favorite cognitive activity without explaining why; detector culture is corrosive; and the boundary between legitimate and illegitimate assistance is difficult to draw. But then, with very little argumentative preparation, we get: preserve literary genres, create professional norms for certifying “organic work,” accept a premium market for it, and perhaps recognize fewer “true writers.” It has the sound of an essay pilot being told, You need to land the plane. But the actual intellectual landing point might simply have been: “I don’t know where the line is. The old line doesn’t survive contact with the technology, and any new one will be contingent, socially negotiated, and probably unstable.” But that wouldn’t feel like an “ending” in the conventional magazine sense. It lacks resolution. It gives the reader no prescription, no hopeful gesture, no institutional program, no “what we must do.” I suspect this produces a great deal of bad nonfiction.

The first-person writer can stop at the exact point where his judgment stops. He can say, “I noticed this. I followed it as far as I could. Here is the shape it seems to have. I don’t know what follows from it.” No editor needs a call to action. No audience-development person needs the reader to leave feeling empowered rather than vaguely unsettled. Blogs at their best have a quality that magazine essays often lack: they preserve the scale of the thought. The writer doesn’t have to inflate an observation into a thesis or a thesis into a program. He can just say, “Here’s how I see it. Take it or leave it.”

I Would Be Lying If I Said That It Wasn’t True, I Only Want Bad Things to Happen to You

By in ai, media/propaganda No Comments

Charlie Warzel:

Of course, I’m gonna be defensive about this. I don’t want something to come out there and take the one marketable skill that I have and just make it irrelevant. But I also think that there’s something else going on here.

No, Charlie, that’s about it. You don’t want AI to put your worthless journalist ass out of work. There’s no deeper subtext to it.

This is from an Atlantic podcast summary titled “AI Writing Has Entered Its Post-Shame Era.” The subtitle is “Media Already Have a Trust Problem.” One of Warzel’s guests is Elizabeth Spiers, the founding editor of Gawker. All this reminds me for some reason of when OJ promised to find the real killer.

On June 3, 2020, J.D. Vance tweeted, “I’m still amazed at how quickly the moral scolding ceased as soon as elite-favored protests began taking place. (I guess I shouldn’t be amazed anymore. I should just expect it.) But it has further eroded trust in our country’s experts.” The tweet has since been deleted, but you can find it quoted here. He was talking about the fact that following George Floyd’s death, the media and public health experts suddenly became supportive of mass gatherings and street protests for favored causes, after three months of insisting that lockdowns were absolutely necessary. This wasn’t some conspiracy theory from the fever swamps of right-wing Twitter. Some publication called the Atlantic granted space to two epidemiologists, one from Harvard, one from Yale, to argue exactly that case in response to growing outrage: “Public-Health Experts Are Not Hypocrites.” The argument was so appallingly weak that even some idiot blogger could dismantle it in a paragraph. But enough reminiscing. The reason I thought of it was because Vance’s tweet was what inspired this response from some guy named Charlie Warzel:

Sigh. It’s all just so much. Pity the poor establishment hack, forced to explain that the Emperor’s fly is not down, and that is not his shriveled hypocrisy dangling in the breeze there, all while enduring calumnies from fishwives and costermongers. Can you believe the impertinence? Frightfully bad form! People are going to use our brazen double standards against us! Is that cricket? Is that Marquess of Queensberry rules? I ask you!

Six years ago, Warzel was not concerned with truth, trust, or institutional credibility, and God knows he was far beyond shame. He only cared about his job, his status, and the party line. Today is no different. His “one marketable skill” is being a propagandist and empty-calorie content-dispenser for Acela corridor midwits, and now comes a machine that can do it better and faster without even having a soul to sell first. OpenAI should incorporate this into their marketing. If they promise me that upgrading to Professional tier would help put Charlie Warzel and canutes like Spiers out of work even faster, I would smash that subscribe button hard enough for it to be measured on the Richter scale.

All We Are Is Piss In the Wind

By in ai, education, foolosophy, the geist of the zeit No Comments

David Shapiro:

What really needs to happen, and what has happened many times throughout history, is that we need to integrate the new technology into society. Like coal and steel, like electricity and radio, and like the internet before it, AI will have its pros and cons. That’s just a fact of life.

As some like to say in the South, fighting that is like pissing into the wind. I will pick my battles, namely about Post-Labor Economics, but that’s integration work, not unserious “shut it all down or everyone dies” nonsense.

I don’t think of myself as a Stoic, but I recognize that Stoic psychology is awfully useful. I just accept that change is always uncomfortable to those who have stakes in the status quo. How I feel about that is irrelevant, and childishly indulgent if I make it the centerpiece of my reaction. I also don’t need to be a Hayekian to notice that social orders routinely produce outcomes that no one designed. If a tyrant confiscates ten million homes for some vanity project, there is a decision-maker, an intention, a victim class, and an obvious moral question: was this act justified? Anger has a target because agency is concentrated. But if ten million people gradually decide they prefer streaming to DVDs, online shopping to department stores, AI-assisted work to unassisted work, or automated service to paying another human to perform the same task, what exactly is the corresponding moral event? There usually isn’t one. There are ten million little decisions, each of which may be perfectly intelligible from the standpoint of the person making it. The video-store clerk loses his job because nobody in particular decided that video-store clerks ought to suffer. People just stopped renting DVDs.  But ideologues, demagogues, and everyday morons turn “millions of people are voluntarily adopting a technology because they find it useful” into a story about a handful of sinister executives doing this to us. The moral grammar is much more satisfying that way. Someone can be blamed, resisted, or forced to stop.

I am mostly tired of what I think of as “theorizing from 30,000 feet.” The discourse around AI is full of humanities blowhards pontificating about What It All Means, despite the fact that none have them have even tried to use it, or understand how and why others have used it. I see Jonathan Pageau eructating about “shining souls” who want to “foster wisdom and virtue” and “practice truth and love,” in opposition to the tech lords building island fortresses who want to force AI on us, a moral framework better suited to Saturday morning cartoons. “Who does it serve?” he asks ominously, with the flashlight under his chin for spooky effect. Well, as Hayek might interject, it serves countless people in countless different ways, like all complex phenomena, and you would have to demonstrate some attempt at understanding those people’s perspectives if you want to be convincing. “The 1% vs. the 99%” was retarded demagoguery when Occupy Wall Street types said it, and it’s no less so when Christian humanists say it.

Most disappointing to me is the fact that so many of the humanities types, who I would expect to have greater-than-average perspective gleaned from engagement with thousands of years’ worth of culture and literature, are disproportionately among those panicking over their own high-altitude, worst-case theorizing. I don’t expect ideologues, demagogues and everyday morons to be capable of sober reflection. I did expect better from anyone who has actually absorbed thousands of years of literature in which we encounter human beings finding ways to be vain, lazy, brave, stupid, ingenious, loving, self-deceiving and disciplined under an astonishing variety of material conditions. But exposure to history does not automatically produce historical perspective. A person can spend decades reading old texts and still experience the present in a radically presentist way. In fact, expertise can sometimes make that worse. You acquire a vast stock of analogies, concepts and quotations, but those can be used either to relativize the present or to dramatize it. If you use them to enlarge your sense of the range of ordinary human adaptation, AI looks like another extraordinary circumstance imposed upon a very old animal. If you use them chiefly to furnish grand analogies, every new circumstance can become Armageddon.

Nothing to Declare But His Genius

By in ai, writing No Comments

Venkatesh Rao:

Starting with this essay, I will no longer be explicitly declaring AI use, or segregating my AI-assisted writing in a separate experimental section. You should assume by default that I’m using AI. See my revised About page for a detailed update on my AI use philosophy.

Rao to the Pangram police:

I love the judo move here, simply sidestepping the attempted thrust of moral authority and using its momentum against it. And I’m glad somebody prominent in this particular intellectual neighborhood is willing to say it plainly. The more respectable writers behave as though being “Pangram’d” is a reputational catastrophe, the more they hand detector companies and accusers the authority to define legitimate writing. But authorship is an agency problem, not a particle-analysis problem. There are no disputes over royalties to settle. Anyone for whom AI assistance is disqualifying has the information needed to stop reading him. Everyone else can go back to asking the old-fashioned question: Is the writing any good? Anything beyond that would only become interesting to a reader if, upon follow-up discussion, it became evident that Rao wasn’t capable of expanding upon “his” ideas. The reader might think, “Wow, Claude must have been doing all the heavy lifting for this guy.” But that’s a risk for Rao to be concerned with, and if he isn’t, I’m not either. Too bad for all the newly self-appointed commissars who will lose their jobs if this example spreads. On second thought, swive ’em.

How Much Do I Love You? How Hard Must We Try?

By in ai, bring me the head of nicholas carr, foolosophy, technology, writing No Comments

Derek Thompson:

In my corner of the world, which is journalism and academia and writing in general, Pangram has become nothing less than a cultural force. A meme, even. I’ve seen prominent politicians, chief executives, and public intellectuals exposed as slop-slingers. Their public humiliation sometimes fills me with righteous indignation. But it also makes my toes curl a bit, as I think about people spending all day policing the internet for slop and attempting to Pangram-shame people with a technology that can make mistakes.

Today’s interview is with Pangram founder and chief executive Max Spero. Among other things, we talk about:

  • what if AI writing takes over the internet and nobody cares?
  • how Pangram works—and how and when it makes mistakes
  • why AI writes “like that”
  • my fear that AI-writing witch hunts would be an unwelcome addition to online life
  • Pangram’s relationship with Substack
  • what if a big tech company, like OpenAI or Google, creates their own AI detector that puts Pangram out of business?

Evidently, I’m destined to be a gadfly, though with any luck not the kind that earns a glass of hemlock. I just think that most of what our commentating classes produce is empty verbiage that avoids the ostensible subject. What I find interesting here, as always, is what our choices reveal about our deeper sentiments. Fifteen years ago, when Nicholas Carr was beating the drum over how hyperlinks, search engines and the internet itself were making us dumber, I was zeroing in on an assumption he took for granted. He assumed that we truly wanted to do more difficult tasks like read challenging books, and that we were being prevented, more or less against our will, by technology. I simply noted that this was a big ol’ fat-ass begged question. Maybe all technology did was reveal our base inclinations once given multiple choices.

There was a whole hierarchy of “wanting” worth reflecting upon in Carr’s case. I want to read difficult books. I want to have read difficult books. I want to be capable of reading difficult books. I want to think of myself as somebody who reads difficult books. I want other people to regard me as somebody who reads difficult books. And, at 8:45 tonight, I want to aimlessly scroll around on the internet instead of reading one of the fifteen difficult books I have piled up on the nightstand. Those desires can all coexist, and they tend to be knotted together like last year’s strings of Christmas lights. Untangling them is hard. It’s much easier to blame an outside agent for “rewiring my brain” and preventing me from being my true self.

Now, here we are, where people are driving themselves crazy wondering who is using AI and to what extent, and they’re in agreement that this is unquestionably a bad thing. And once again, I say: isn’t it interesting that the thing we all profess to want is the thing to which none of us apparently want to devote time and energy? To me, all LLMs are doing is finally providing an equal alternative: you mean I don’t have to waste any time trying to write some LinkedIn bullshit? And it turns out that no one actually wants to devote time to writing LinkedIn bullshit!

But — crucially — we can’t bring ourselves to be honest about this, especially when it turns out that we don’t particularly want to devote time to writing posts and essays, either. We fear what it might say about us if we spend most of our lives wishing we could be elsewhere doing other things. We fear an honest accounting of our sunk costs and an honest reckoning with our regretted compromises. We fear an honest look at how much of our lives we would happily automate away if we had the option. Rather than face that squarely and decide what to do about it, we retreat to the earlier stage: proclaim that, say, writing LinkedIn posts by your own unaided effort is a good and worthwhile thing to do, and anyone who doesn’t want to do it is a bad, shameful person. It’s like a religious fanatic who catches himself thinking sinful thoughts and responds with horror and redoubled efforts at prayer and self-flagellation. Pray the AI away is a strategy, I guess, but I don’t think much of its chances.

We have spent ages assigning moral dignity to activities partly because we had to do them. AI is forcing us to discover how many things we called “meaningful work” because, until about five minutes ago, we had no practical way not to do them. As usual, sudden exposure to undiluted self-awareness sends us into a pants-pissing panic. Who are we in the absence of compulsion? How much choice is too much choice? What are my true desires, or are “desires” too ephemeral to bear the weight of a personality anyway? This sort of reflection seems much more penetrating to me than counting em dashes, which is probably why we’re talking about the latter.

The Signal Has Gone Dead, All That’s Left Is Noise

By in ai, calvin and hobbes, foolosophy, writing 1 Comment

Juliette Ryan:

I can fairly quickly spot a piece of AI-generated writing but I know my judgement is not infallible. This is why I support Substack’s decision to integrate AI detection into their platform, a decision that was met with great uproar amongst many writers. To me, the backlash reveals how far the virus has spread. The loudest objections weren’t about the detection itself but what it implies: that the machine did the thinking.

The gist of Ryan’s argument is that LLMs have dramatically lowered the cost of producing misinformation by giving it the camouflage of fluent prose. Before LLMs, producing a polished 1,500-word argument required somebody to possess at least a certain amount of verbal skill, time, or money. That did not make the argument true, but it imposed a production cost. Now someone who barely understands a subject can produce a plausible-looking essay about it in minutes. The old signal has become noisier. True enough. But earlier in the essay, she writes:

After all, to truly understand something, you need to be able to explain it. And to explain it, you need to have assimilated extensive knowledge and the ability to synthesise it — to recognise where and how to build bridges over that dense knowledge so that they can be walked by the reader. Before AI, fluent writing signalled that someone had laboured for hours trying to collapse three-dimensional thought into a two-dimensional stream of words that could convey that understanding to another person. Good writing meant good thinking. Or it used to.

Good writing =/= good thinking. They are different skills. I know intelligent people whose prose is stilted and clunky. I know English majors who are astonishingly dumb. Rhetorical fluency is the ability to smoothly join a premise to a conclusion. It tells you nothing about the validity of either. Writing can be clear without being good in the sense of true or accurate. We know this from something as small as the aphorism. Her own post illustrates this.

Leaving aside her blithe assumption that detectors like Pangram work as well as their marketing copy would have you believe, her entire “mind virus” metaphor assumes a reading environment that I simply don’t recognize:

AI has fooled a large population of online content creators into thinking they know what they truly think. Their superficially fluent output is then read by unsuspecting others, unknowingly simulating (possibly false) understanding in their own minds, and they pass it along with a cursory hit of the “share” button. And so the virus spreads this pseudo-understanding, infecting the minds of all who say they don’t care whether they’re reading AI-generated content.

Here she seems to suggest that no one ever questions anything they read. It’s like she assumes that the interaction begins with a piece of writing and ends with its first impression on the reader. But most writing that makes an impression on me stays with me and gets chewed on for a while. Eventually, some vague dissatisfaction from early on makes itself known as a coherent objection. It sounded good at first, but something just didn’t feel right. I mulled it over, and finally, the problem becomes obvious.

She also acts like the author’s pseudo-profundity, once established, remains. But let’s assume that someone has pulled the wool over everyone’s eyes by using AI to cover up a weak thesis. Or perhaps the thesis is strong, but it’s not truly the author’s, because Claude supplied it. In either case, the solution is to follow up with pointed questions or criticisms, perhaps in live conversation. A person who truly doesn’t have command of the ideas will not be able to keep faking it. Eventually, you’ll notice the lack of true familiarity with the subject. We all know this from our days of giving book reports in front of the class. You always knew the kid who didn’t actually read the book, because he wasn’t capable of summarizing it or expanding upon it. (I may have even been that kid on one occasion in third grade; I know my subject from the inside, as it were.)

In both instances, she acts like first impressions are forever. Her model of the reader is strangely passive for someone arguing that people need to preserve active cognition. She wants authors to think strenuously, but readers apparently just absorb whatever washes over them. Yet reading itself can be a form of thinking. The reader can do the intellectual work the writer failed to do, which is exactly what is happening here. Her prose is fluent. The argument initially has shape. Then you keep turning it over, and the hidden assumptions start coming loose. The irony is that her essay is explicitly worried about fluent language creating the illusion of understanding, and “mind virus” is itself a fluent phrase that can create the illusion of understanding. It’s such a stupid, reductive way to describe what is going on in this context. It’s fine for polemic, but useless as analysis.

Online polemics

Sculptural Writing

By in ai, aphorisms, books, technology, writing No Comments

Machines are still only in their trial period. One day, perhaps, an electrophonic kaleidoscope will compose musical figures in their thousands — will invent whole series of rhythms and melodies. Man will have machines that create just as he will have machines for precision thinking — His role will be reduced to choosing. Combination factories.

The most intense feeling is only one combination among others.

— Paul Valéry, Cahiers / Notebooks 3

The standard complaint about large language models is that they will make writers sound alike, and perhaps eventually think alike. But Valéry’s prescient suggestion makes me wonder about another possibility. What if these machines do not merely automate an existing kind of writing? What if they also call forth a somewhat different skill in writers?

We ordinarily imagine writing in painterly terms. The writer confronts a blank surface and begins putting things onto it. A word, then a sentence, then a paragraph. The work accumulates. Even when we know perfectly well that revision will follow, our basic picture of authorship remains additive: there was nothing here, and now the writer has made something out of his inner resources and aesthetic vision.

An LLM makes possible a more sculptural way of working. Instead of confronting a blank page, the writer can begin with an excess of material. The machine can produce the boulder of text. The writer’s task is to decide what, if anything, is inside it. That sounds close to Valéry’s prediction that our role will be “reduced to choosing.” But choosing may be doing a lot of work there. A sculptor also “chooses.” This part stays; that part goes. This line needs to emerge; that mass is obscuring it. The finished form depends not on the ability to manufacture marble but on seeing possibilities in it that someone else would not see. Two sculptors can begin with approximately the same block and end with objects bearing no resemblance to one another.

Give two people the same serviceable paragraph from an LLM. One may read it and think it’s fine as is. The other notices immediately that the opening sentence is unnecessary, the metaphor in the middle is the only interesting thing there, the conclusion merely repeats the premise, and one odd phrase suggests a better direction for the whole piece. This would shift some of the emphasis in writing from generation toward discrimination. An LLM can give you ten openings before you have finished thinking about the first. It can propose twenty metaphors, five structures, three counterarguments, a dozen ways of phrasing the sentence you are stuck on. Under those conditions, producing another possibility is nearly worthless, but knowing which possibility is promising becomes more valuable. That sounds less like the traditional ideal of the writer and more like the work of an editor.

I haven’t experimented with this yet. I don’t know how it might go in practice. It probably would not go one single way for everybody. But this seems more interesting than treating AI-assisted writing solely as a question of whether the machine has “done the writing.” That formulation assumes we already know where the important part of writing resides. Well, maybe. Or maybe abundance changes where the difficulty lies. The blank page tests one set of capacities; the overflowing page may test another.

In a World of Machines, What Can I Do But to Serve

By in ai, the geist of the zeit, writing

Way back in the day, my mom had a part-time business typing up term papers and dissertations for university students. Well, I did say “way back in the day.” We’re talking about a time when grad students would come in with handwritten theses and manuscripts. They didn’t have their own typewriters, let alone word processors. And yes, I was alive to witness this. Phones were attached to walls and desks, restaurants had smoking sections, cars had paper maps in the glove box, and everything was sepia-toned.

Point is, I’m considering resurrecting the old family business with a modern twist. In ye olden times, students supplied messy handwritten intellectual labor and paid somebody with technological/clerical expertise to turn it into professional-looking typescript. Today they could supply immaculate machine typescript and pay somebody with humanistic expertise to put the mess back in. (I also bet “humanism” will start to acquire new connotations in the age of AI.)

Why not? We have apparently reached the stage where software companies sell machines that generate prose, other software companies sell machines that detect the generated prose, and still other software companies sell machines that make the generated prose look less generated. At some point the cheapest solution will be to hire a human who grew up in a sepia-toned world.

It Seems Strange How We Used to Wait For Letters to Arrive

By in ai, the statusphere, writing

Giles Crouch:

When a human has written something, as I am writing this now, there is a cost (I’ve got a ton of other things on my to-do list), and we inherently know this cost. Just as we are, when reading something, paying a cost with our attention. We like good writers, something I am striving for, because they have thought something through, done the research, crafted and organized the story. If they are good at it, we like that. It’s why we have the Pulitzer Prize or in Canada, the Giller Prize. Most nations have them, or a lucky writer becomes a best-selling author. We are rewarding the cost of their attention to the writing because we are happy to pay the cost of reading their works. Part of Mauss’ gift economy idea.

The labor theory of value was dubious enough when economists used it. I see no reason to resurrect it as a theory of writing.

Years ago, the Lady of the House and I did some freelance corporate copywriting. This was the golden age of SEO, when every business needed a steady supply of “fresh content.” The work was excruciating for me because there was absolutely nothing that needed to be said. It was the ne plus ultra of bullshitting. Good spelling and grammar were important, but every other rule of writing as I understood it was irrelevant. I was simply required to manufacture several hundred new words about an old subject. There was no fact-checking. Hell, there was no audience but Google itself. We were basically just offering generic words up in mass sacrifice to the inscrutable algorithm so that anyone Googling “how do u use dental floss” would find a post for Dr. Yankem’s dentistry services somewhere in the first few results. I labored mightily, but the resulting prose did not thereby become meaningful.

Crouch’s argument about AI writing rests heavily on this connection between effort and value. Good writing costs the writer attention; the reader reciprocates by spending attention on it. AI supposedly disrupts this “gift economy,” a concept he borrows from anthropology, because the machine allows the writer to avoid paying his share. But effort is a fact about production, not a measure of value. We don’t know how many hours a Pulitzer winner spent on a particular chapter. We don’t reduce our estimate of a novelist after learning that he writes quickly; we might even see him as a natural genius. Conversely, my old SEO copy could absorb hours of my attention without becoming worth anyone’s time.

Crouch also describes an “uncanny cognitive valley”: AI prose feels almost human but somehow off, and once readers recognize its tells they resent the writer for not offering them genuine attention. I have heard this complaint before. Writers used to complain about email, text messages, and social media in the same way. In 2010 I wrote about people who called online communication cold and impersonal. My explanation was less exotic: most people are simply unskilled writers. Plenty of intelligent, interesting people are poor at turning their thoughts into nuanced sentences that capture the flavor of their personalities. Their emails were brief, awkward, misspelled, utilitarian. The problem wasn’t email; they just weren’t good at prose.

Sixteen years later, the same people have better equipment. An LLM can remove misspellings, fix grammar, supply transitions and produce sophisticated-looking sentences. What it cannot automatically supply is judgment. A weak writer may not recognize that a metaphor is awkward, a paragraph redundant, an argument too neat, a phrase unlike anything he would actually say. He now produces something different from the clumsy email of 2010: polished generic prose. The underlying problem is the same. What Crouch identifies as machine uncanniness may only be fluency without discrimination. Like the sorcerer’s apprentice, the animated tool has gotten out of the pedestrian writer’s control. He isn’t skilled enough to shape and direct the output, and the reader can tell: “This doesn’t sound like him.” People who aren’t good at shaping their own thoughts are also not good at shaping prompted thoughts. It’s not because an LLM inherently makes it impossible to produce good prose, it’s that most people aren’t good at it even if an LLM gives them a leg up.

So, if we don’t need elaborate theories from anthropology to explain the extraordinary agitation about AI writing, what would suffice? A cold look at basic human psychology seems good enough to me.

People are conservative in the elementary psychological sense: radical novelty makes us uneasy. Most of us also lack the time or inclination to investigate complicated new subjects deeply, so we glean a basic sense of what our peer groups think and conform to that. Add a nonstop barrage of posts and articles telling us that AI will steal our jobs, drink all our water, eat all our books and perhaps terminate civilization, and a certain amount of reflexive hostility is unsurprising.

But writers, academics, journalists, editors, and other people whose self-conception is bound up with verbal facility have an additional problem. Fluent prose has long functioned as a status signal. It suggested education, wide reading, practice, professional competence, cultivation. You could often locate somebody roughly within a cultural hierarchy after reading a page of his writing.

AI has has abruptly cheapened a form of cultural capital, and the people who accumulated that capital are understandably unhappy. Someone who once advertised his limitations through awkward sentences can now produce something resembling educated professional prose in seconds. The old distinction has become harder to see at a glance. Hence the sudden emotionally-charged obsession with restoring provenance: AI detectors, mandatory disclosures, watermarks, “human-written” certifications, threats to stop reading anyone caught using the machinery. If the clothes no longer reliably identify rank, somebody has to start checking invitations at the door. It’s a linguistic version of sumptuary-law anxiety: “Wait, how am I supposed to know who is entitled to wear the velvet now?”

Harmony’s Gone, the Rhythm Remains

By in ai, aphorisms, religion, writing

An obvious first question: which religion? (Orthodox Christianity, in her case; condolences to all you heretics.) An obvious follow-up question: what about the excluded middle option, that we decided “religion” was useful despite considering and rejecting the possibility of its truth?

And the possibilities aren’t limited to “true” and “useful-but-false” anyway. A religious tradition might contain some true psychological observations, some effective social technology, some beautiful mythology, some destructive doctrines, some historical falsehoods, and some metaphysical claims whose truth remains undecidable. “Is religion true?” is absurdly underspecified.

As much as I love aphorisms, epigrams, etc. as a form, I can never fully relax around them. Their rhetorical tidiness, which makes them so attractively quotable, also makes it easy to overlook all the flaws. We can be swayed by the seemingly-self-contained presentation to think that the matter has been settled. In reality, the compact saying might be a little Procrustean bed, with all the important details littered on the floor around it.

Even there, though, it’s too broad to complain about “aphorisms, epigrams, etc.” There are multitudes of good and bad examples. I’m mainly just inherently suspicious of rhetorical confectionery. Certain formulations arrive pre-glazed. The antithesis is perfect, the cadence closes neatly, the conceptual roles are cleanly assigned, and the reader gets the little reward of understanding before he has actually had to understand very much. The sentence gives the sensation of intellectual completion. When you’re younger, the sweetness itself can register as intensity: Christ, what a line! After enough reading, you start tasting the corn syrup, and your aging taste buds revolt.

A useful distinction might be between compression and reduction. A good aphorism is compressed: there’s more in it than the sentence can explicitly contain, and unpacking it makes it stronger. A bad one is reduced: inconvenient possibilities have been removed so that the sentence can achieve its pleasing shape. Unpacking it makes it weaker because you discover that the apparent depth depended on nobody asking obvious questions.

We have acquired this strange retrospective suspicion that polished symmetry is somehow an artifact of generative AI, when humans have been mass-producing parallelism, antithesis, metaphor, repetition, and strategic omission for centuries. No transformer required. All you need is an ear for cadence and a willingness to let the melody carry you along.

The old anti-sophistic suspicion is that persuasive form can become detachable from truth, belief, or character. Give the skilled rhetorician a proposition and he can make it sound weighty; give him its opposite and, with enough ingenuity, he can make that sound weighty too. The listener experiences the force of conviction without knowing whether any conviction exists behind the performance. This may explain some of the anxiety around LLMs. They demonstrate clearly that the external signs we associate with thought — fluency, balance, confidence, elegant qualification, apt metaphor — can be generated without anything resembling human conviction behind them. But that revelation cuts backward too. We were already too willing to treat those signs as evidence of depth when humans produced them. The clankers merely industrialized a snack food humans had been making at home for millennia.

You Scurvy Scalawag!

By in ai, calvin and hobbes, jests japes jokes jollies, writing

I don’t want to make bruising fun of this poor girl. It’s mainly just sad that she seems to feel she’s committed a mortal sin against the sacred craft by having AI do basic proofreading. But it does make me genuinely chuckle to see her itemize by nationality the fifty-three people she “deceived” like casualties in an international incident. You can practically hear the cello as the map slowly darkens country by country. If she had simply written, “I realized my ‘100% no AI’ statement was technically inaccurate because I used ChatGPT to proofread one post, so I’ve changed it,” there would be nothing to laugh at. Instead she has imagined a blast radius. Fifty-three people walked innocently into her website, believing themselves to be in a pristine human-text environment, and emerged with invisible synthetic shrapnel lodged in their souls. Even funnier, the odds are good that they were mostly search crawlers, link-preview bots, or other automated scanners rather than a solitary Dane sitting thoughtfully by a window, coffee cooling beside him, devastated to discover that the author had used ChatGPT for proofreading. It may instead have been a server in Copenhagen requesting the page for 0.14 seconds.

But as I said, it’s more sad than anything. The “contamination” model has overpowered an individual conscience. Grammarly? Word’s grammar checker? The red spelling squiggle? An editor who changes a comma? Those are presumably compatible with “100% me.” But ChatGPT correcting precisely the same comma introduces some tiny metaphysical admixture of Not-Me. This is how you know a purity regime has really taken hold: you no longer need inquisitors. The prosecution is internalized. Congratulations to all the self-appointed Torquemadas of the online world; your fanatical efforts have made a difference.

Antonette, dear, you’ve done nothing wrong. If I can offer one bit of advice, it would be not to make grand oaths about how you will or won’t write, especially if you’re new at it. Just practice, experiment, and enjoy yourself. Use AI or don’t; that choice is yours, and you’re free to change your mind tomorrow. But don’t let an imaginary audience of suspicious readers turn writing into a purity test, with every new tool or bit of assistance treated as a possible moral failure. You’re learning how you want to work. Give yourself room to find out.

Come Back, Signal, Fill This Void

By in ai, art, foolosophy, music, writing

Me, I’ve been existentially depressed ever since people stopped making music purely by vocalizations and started using artificial “instruments” that don’t exist in nature. It’s all been downhill ever since clicks, hoots and whistles. An animal gut stretched across a resonating box?  Suspicious. Somewhere in the Upper Paleolithic a cultural critic looked at the first bone flute and thought, So this is it. We’re outsourcing the human larynx. That first flute was Big Tech’s war on hooting.

Seriously, though, he seems to feel that either humans “produce” culture or machines do. The actual situation is already spreading itself across a huge continuum. Someone can type “write me a song about heartbreak” and upload the first result. Someone else can write the melody and lyrics, use AI to generate an orchestration, reject half of it, rewrite the rest, sing the vocal himself, and mix the recording conventionally. Somewhere else a musician may improvise against an AI-generated part and discover a musical idea neither input contained straightforwardly beforehand. Which of these is “no longer produced by human beings”?

There is also something revealing about the nostalgia implicit in that phrase. Produced how? Apparently there exists some blessed technological moment at which human cultural production was sufficiently pure. Was electric amplification all right? Photography? Word processors? Photoshop? Sampling? MIDI? Spellcheck? Search engines? Dictionaries? Editors? Session musicians? Ghostwriters? At some point the natural human apparently fell from grace and began cheating with artifacts. AI may be an unusually powerful and strange extension, perhaps strange enough eventually to require new ideas about authorship. But “artificial means entered the creative process, therefore the result ceased to be human” would condemn an astonishing amount of civilization.

A musician may say, “I invented this chord progression,” but in one sense the progression was always latent in the combinatorial space of tonal music. What he really did was discover it as meaningful for this song. Likewise, an aphorist does not invent the English language or even all the conceptual relations inside a sentence; he notices that certain words, placed together under the pressure of a certain thought, suddenly become exact.

AI can widen the field of discovery enormously. But widening the field does not tell you what to notice. If ten people ask for “an unusual chord for the bridge,” they may receive overlapping suggestions. But they will not hear those suggestions identically. One person hears pretension. Another hears melancholy. Another hears the exact pivot the song needed. Another thinks, “Interesting, but if I lower the third and move the bass note…” The machine output becomes raw material the instant it encounters judgment.

The creative act often consists in selection under pressure from a particular project. The musician hears or encounters something and thinks: that. There. Which means the genealogy of the input is often secondary. He might find the chord in a theory book, steal it from Coltrane, stumble into it by mis-fingering, hear an HVAC unit hum a strange interval, or have an AI propose it. Those routes differ historically and perhaps ethically, but they don’t necessarily differ in the part that matters artistically: the musician perceives a fit between an external possibility and an internal intention. The refrigerator compressor making a weird noise is not “collaborating” in any ordinary sense. It has no conception of the song. Yet it can still trigger something decisive because the artist already has a receptivity shaped by the work in progress. A thousand other people hear the same noise and hear only an appliance. One person hears the missing texture in the bridge. That seems very close to what inspiration has always been. The world throws off material constantly; the artist’s mind is a selective field.

AI may be to intellectual and artistic work what the Industrial Revolution was to material production: not the abolition of human agency, but the mechanization of tasks that once consumed large amounts of it. A writer once had to search books, wait on memory, or labor through dozens of possibilities to find the right metaphor or rhetorical device; a musician might spend hours experimenting before stumbling onto the right chord or texture. AI industrializes that search. It can generate possibilities at machine speed and scale, just as machinery multiplied the output of physical labor. That does cheapen some forms of accomplishment and may displace particular skills, but it also shifts the scarce resource elsewhere — from producing possibilities to judging, selecting, combining, and directing them.

The deeper continuity is human temperament. Industrial machinery did not make craftsmanship, ambition, laziness, ingenuity, or mediocrity disappear; it gave them new conditions in which to operate. AI will likely do the same. Some people will use it to avoid effort and become little more than approval buttons. Others will use its abundance as cognitive leverage: more examples, more counterarguments, more metaphors, more experiments, more practice — and thereby enlarge their own capacities. The machine can make fifty suggestions; it still cannot decide which one belongs in this particular human project, or why the project should exist at all.

A chainsaw and a violin are both elaborate artificial extensions of the body, but their social and psychological effects are completely different. A typewriter, a search engine, and an LLM all intervene between thought and finished prose, but they intervene in different places and alter different capacities. So the useful questions become more local: What does this tool make easier? What does it make unnecessary? What skills does it cultivate or atrophy? What new possibilities does it create? What dependencies does it introduce? What kinds of people and institutions tend to form around it? That’s much more difficult than saying “technology went too far after X,” because you lose the clean moral geography. You can no longer live conceptually in the Good Village on one side of the river while the Satanic Mills smoke on the other.

The Decadence Preservation Society

By in ai, the geist of the zeit, the statusphere

Anastasia Berg and Jon Baskin:

Even more important is to create and protect spaces that allow people to read, write and discuss ideas without the encroachments of AI or any other digital technology.

…In response not only to the evangelists of posthumanism but also to a more pervasive despair about the ability of our words and actions to make things happen, these spaces must double as training grounds for the guardians of human language, thinking and imagination. For the “radical” task before us is not to hope that something resembling our concepts and practices will reemerge in the distant future; it is to do whatever remains in our power to preserve them in the present.

Like most reflections on AI from the writer’s guild, I don’t find this one very thought-provoking. What I end up thinking is that they have identified a real question but chosen the wrong scale for it. AI will force us to decide which activities are valuable only for their results and which derive some of their value from the human performance itself. A concept becoming difficult to apply is not the same thing as a concept becoming unintelligible. AI may make it harder to know whether Fred wrote his paper, whether Susan knows what she’s talking about, or whether some award-winning novel owes half its prose to Claude. That can obviously produce serious problems of verification and attribution. But the very fact that we care about those questions suggests that merit, authenticity and fraud remain perfectly intelligible. Indeed, fraud may become more salient when imitation becomes easier. Counterfeit money doesn’t make the concept of genuine currency unintelligible.

Their line about unassisted work becoming a mere “hobbyist’s predilection,” like using a typewriter, makes the confusion clear. It assumes that people undertake difficult activities primarily because there is no cheaper way of obtaining the output. But as I just said recently, a forklift has not made a 500-pound deadlift unintelligible. It has merely made obvious that moving 500 pounds from A to B and deadlifting 500 pounds are different activities.

But my main impression is that I’m once again a bit surprised at how uniform the reaction to the strange new tool has been among the supposedly-creative: pull up the drawbridge and ready the boiling oil. Where’s the curiosity, the experimentation? How psychologically conservative, even reactionary, of them! The stereotypical artist used to be the person who horrified respectable people by saying, What happens if we use this new thing in a way nobody intended? Photography, collage, tape manipulation, synthesizers, sampling, cut-ups, video, photocopiers, digital editing — the creative temperament was supposed to be promiscuous toward tools. Give the artist a machine and the artist immediately misuses it. Now the machine arrives and a surprising number of literary people become the church ladies: Absolutely not. That infernal contraption will alter the traditional relationship between Man and Paragraph!

In a funny coincidence, Arts & Letters Daily linked to this article, in the achingly-progressive Guardian, on the same day as the one above. The authors argue that contemporary people possess the costumes but not the spiritual substance of fin-de-siècle decadence: the luxury, excess, artifice, bodily display are there, but the authentic ethos has vanished. Which is, amusingly, structurally identical to every conservative lament about lost forms: we have Christmas without Christianity, universities without learning, manners without morality, architecture without craftsmanship. Except here, apparently, we have decadence without decadence. The old decadent said: Bourgeois morality is suffocating; propriety is fraudulent; good taste is merely convention; let us cultivate artifice, perversity, intoxication, useless beauty. The new nostalgic decadent says: Young people are practicing decadence incorrectly! There used to be standards! Aubrey Beardsley has somehow become Norman Rockwell.

I mean, I too think of scrolling, “looksmaxxing,” and self-quantification as dubious pleasures, but it’s simply hilarious how we’ve moved from “Away with Victorian moralism! Follow sensation!” to “No, no; not those sensations!” The interesting question is what an embodied human being can do with the artificial thing. If Wilde had somehow been handed Claude, I doubt his first reaction would have been, “Alas, this device possesses no phenomenological interiority; lock it in the cupboard.” I suspect you’d lose him for three days. That’s the missing decadent response: “What deliciously improper thing can I do with this?”

Instead, at almost precisely the place where their historical decadents would seem to recommend promiscuous experimentation with artifice, the authors become guardians of traditional authenticity. AI artifice is disqualified in advance because it isn’t backed by the correct metaphysics of embodiment. Again, how can you not laugh in delight? The self-appointed heirs of “the first duty in life is to be as artificial as possible” encounter perhaps the most powerful engine of artificiality ever devised and again reply, “No, no, not like that!” They make much of the decadent ideal that pleasure should be useless, an end in itself, opposed to Victorian utility, productivity and profit. Yet by the end they have turned decadence itself into a social-improvement program. They advocate for a “New Decadence” to resist optimization, restore embodied pleasure, revitalize culture and “force fresh blood through today’s complacent, anemic culture.” Poor des Esseintes. Even decadence has KPIs.

Fin-de-siècle fetishism aside, we’ve been living under the Baby Boomer ethos for a half-century now. David Brooks was writing about “bourgeois bohemians” around the turn of the century. That template of rebellious hedonism has long been a stale cliché in itself. But it really begins to feel its age when change arrives from the direction of the future, aided by technological developments that we have long thought of as progressive. If the kids today began to show signs of a return to sobriety, religion, monogamy, and so forth, that could easily be understood as a lamentable return to reaction. You could say, “Ah, the pendulum has swung; the squares are back,” and remain secure in the belief that you still understand the map. But what happens when you become the person saying, “That’s not a real relationship, that’s not authentic intimacy, you’re supposed to want human connection, this technology is degrading something essential”? You may possibly be correct, but you can no longer pretend to be Dionysian.

For people whose cultural identity was formed around novelty, liberation, experimentation, transgression, technological optimism, or being on “the right side of history,” AI can produce an almost metaphysical embarrassment. It threatens to reveal that progress was not a destination at which they arrived. It was a train they happened to ride for a while. And trains keep going.

Claudemarking, Claudewashing and Nudereading

By in ai, jests japes jokes jollies, the feeling of absurdity, writing

Join me, if you will, as I bow my head and give thanks to the trickster god who loves us and wants us to be amused. What’s that? You’re agnostic? Well, let’s see…

I wonder how long it will be until some enterprising coder—oh, really? That was fast.

Let’s sum this up. Claude is beginning to watermark text, and others seem likely to follow. I don’t know if this has an official name yet, so let’s call it “Claudemarking.” If you copy and paste text that has been “processed” by an LLM, it will bear the devil’s clanker’s mark, which others will be able to detect. But the human entrepreneurial spirit being what it is, this will immediately be accompanied by means for removing said clanker’s mark. Let’s call this “Claudewashing” (which will quickly become an epithet for those suspected of evading justice). Inevitably, an online service will appear with a name like “HumanizeMyClaude.ai” and a landing page showing a smiling undergraduate giving a thumbs-up. Philosophically, this is funny because if a mark can be stripped, absence proves nothing. If a mark survives editing, presence proves only that the text passed through Claude at some point. That leaves us exactly where we were: arguing about what “authorship” means, except now with cryptographic confetti scattered over the argument.

So we already have Pangram; soon we’ll have Claudemarking. I’m usually not one for predictions, but I’m going to take a stab at the likely next innovation in digital prophylactics: SynthScreen. I envision this being a new version of that wearable tech that has never quite caught on, like smart glasses. SynthScreen would have a full Iron Man HUD for prose. You’re reading an essay and little red brackets suddenly snap around a sentence:

SYNTHETIC-LIKELIHOOD EVENT DETECTED
Triadic construction: 87%
Em-dash residue: elevated
Metaphorical symmetry: abnormal
Recommended action: discontinue reading and unsubscribe immediately

Naturally there would be different protection levels. Normal web browsing requires only the glasses. High-risk environments — corporate LinkedIn, university DEI statements, Substack during a discourse outbreak — require full Narrative-95 filtration. Then comes the social etiquette. Someone sends you an article and you ask, “Was this SynthScreened?” They say no. “You mean you just sent me raw text?” “They told me it was human-written.” “Did you see the test results?” “No, but I trust them.” “Jesus Christ, man!” Offices establish designated clean-reading zones. Libraries put signs on the door: THIS FACILITY MAINTAINS A LOW-SYNTHETIC ENVIRONMENT.

Eventually, all this fussy hypochondria around reading as a biosecurity procedure produces a counterculture that glamorizes risk and danger. A new (old) practice emerges. Let’s call this “nudereading.” People. Text. No intermediaries. You just read and use your judgment, like they did in the Before Times. Underground book clubs meet without scanners. A dissident magazine announces that it accepts submissions in whatever state they arrive and refuses all provenance testing. I’m picturing scenes like Dead Poets Society, where the students slip away to a cave in the night to read and chant text with no heed of provenance. The elders are scandalized. “You don’t know where that text has been!” “Do you have any idea how many models may have touched it?” “Your cousin’s friend’s co-worker nuderead one newsletter in 2029 and never recovered!” School assemblies feature reformed clankertext addicts describing how it began innocently with a few Claude sonnets and ended with them reading anonymous machine-generated erotica in unlicensed prompt dens. Clankertext culture takes on the most subversive connotations, like gay bathhouse culture of past decades. Soon, somebody, probably the Atlantic, publishes a stern public-health article entitled “The Dangerous Romanticization of Unprotected Reading.”

Then you get the full subculture apparatus: clandestine reading rooms, coded symbols, pseudonyms, bars where the house rule is that nobody asks whether the prose is human or machine. “Don’t provenance me” becomes an assertion of privacy. People wear pendants shaped like em dashes. There are digital zines with manifestos about textual autonomy and the right to unreadability of origin. Meanwhile clankertext itself develops an aesthetic of deliberate excess. Why pretend to be human? Let the machine write six-page sentences, hallucinated metaphors, impossible registers, prose no sane MFA workshop would permit. What began as stigma becomes style. “Human-passing” is now the insult. Young writers sneer at clean, tasteful, provenance-certified prose as “meatsack realism.” You get the idea.

Anyway, the “effort-to-result” ratio of all this bot-sniffing reminds me of nothing so much as watching a dog chase its tail. I think I previously referred to this farce as an arms race, but if so, I retract that. An arms race at least suggests two sides accumulating genuinely greater capabilities. Here a lot of the motion is just circular, like the dog and his tail. We devise a technique for identifying synthetic prose; therefore we create a technique for making synthetic prose harder to identify; therefore we devise a technique for identifying prose that has been deliberately made harder to identify, and round and round we go…until somebody finally publishes a piece in the Atlantic titled, “How I Learned to Stop Worrying and Love Nudereading,” culminating in a revelation: “I had spent years trying to determine whether the intelligence behind a sentence was artificial. It had never occurred to me to ask whether the sentence was intelligent.” Cue the podcast appearances, and an eventual book titled Raw: Reclaiming Reading in the Age of Synthetic Text.

So, then, as I was saying: Would you like to join me in a prayer of thanks?

Why Y?

By in ai, foolosophy, technology, the geist of the zeit

Gregory Conti:

AI means the death of authorship—and of the inspiration that work which you know reflects human achievement, and human competence, provides to all who read it. We will increasingly become mere “prompters,” at best “collaborators” with the machine. As sci-fi as it sounds, we are already well on our way to becoming mere interchangeable vessels for ChatGPT’s accomplishments and insights. The awe that we feel in reflecting on an Einstein or a Dostoevsky, we will feel toward no one again; Claude cannot be someone’s hero. The whole concept of a “life’s work” that belongs to a great progenitor of ideas is under threat; soon we may have less reason to care who “wrote” a paper with an AI’s help or who inaugurated a theory by prompting Claude than we do about who was riding Secretariat when he won his triple crown.

Conti is reflecting on what AI means for human achievement in the big scheme of things. I don’t have strong feelings about the majority of his points, but this one stands out like a sore, dumb thumb to me.

The crucial word is interchangeable. Give the same powerful tool to ten different people and they do not become interchangeable unless the only thing they contribute is pressing Enter. They bring different questions, accumulated knowledge, obsessions, standards of evidence, aesthetic preferences, capacities for judgment, willingness to reject attractive nonsense, and, perhaps most importantly, different ideas of what is worth doing in the first place. AI can supply an enormous amount downstream of those choices without making the choices cease to matter.

In fact, Conti’s own essay inadvertently demonstrates this. Claude apparently got close to an argument he was already developing, and he found that “phenomenally deflating.” But why was that particular argument significant to him? Why had he chosen those primary sources? Why was he interested in the question? Why does he accept some conclusions and reject others? Claude did not spontaneously decide one morning that technological succession was a threat to human flourishing and pitch an essay to Compact. Gregory Conti did. The machine’s competence can potentially alter how much credit he deserves for particular inferential moves, but it doesn’t turn him into a fungible nozzle through which Claude squirts ideas.

The machine, unlike Secretariat, doesn’t want to run anywhere. Secretariat was an organism pursuing the same immediate end as Ron Turcotte: get around the track ahead of the other horses. Claude has no intellectual project corresponding to Conti’s. It doesn’t care whether AI destroys authorship, whether the argument is published, whether it is true, whether it contradicts something it said yesterday, or whether anybody reads it. It can perform cognitive labor inside a project, sometimes astonishing amounts of it, but the project still has to come from somewhere.

“Claude cannot be someone’s hero” is therefore true and almost completely beside the point. Neither can a forklift. Yet the existence of forklifts has not destroyed our ability to admire someone who deadlifts 1,000 pounds. A forklift can make the human achievement look absurd if the only thing that matters is “how many pounds moved vertically?” But that was never what the human achievement meant. We admire a human being’s ability and desire to do it.

Cars didn’t destroy running. Calculators didn’t destroy mathematical virtuosity. Cameras didn’t eliminate painting. Recorded music didn’t make live musicians pointless. Mechanical advantage doesn’t automatically annihilate human accomplishment because human accomplishment has never consisted simply in maximizing output irrespective of the means. By Conti’s logic, chess should have become meaningless sometime around Deep Blue, because there was “one fewer arena in which a person, through their diligence and hard work and intellect, can shine.” Instead, millions of amateurs remain devoted to the game, and Magnus Carlsen remains famous precisely because nobody gives a damn that Stockfish could annihilate him.

A person who spends twenty years reading, thinking, developing peculiar preoccupations, deciding what questions matter, interrogating an AI, recognizing which of its suggestions are fertile, rejecting ninety-nine others, connecting one result to something he read fifteen years ago, and shaping all of that into a coherent body of work isn’t an “interchangeable vessel.” He’s doing something much closer to what authors have always done: imposing the form of one mind upon heterogeneous material. Conti sees a real problem — machine contribution can become large enough that conventional human credit becomes fraudulent — and expresses it as the much grander and far less defensible proposition that machine capability renders human individuality irrelevant. Those aren’t remotely the same claim.

AI attacks the scarcity of competence much more directly than it attacks the scarcity of salience. Producing a competent essay, competent illustration, competent computer program, competent critique — these things may become astonishingly cheap. Deciding that this obscure conjunction is interesting, that that apparently trivial experience deserves attention, that everybody else is asking the wrong question, that a ridiculous analogy is actually illuminating: those are different faculties. Weirdness is one of the sources from which importance originates. Dostoevsky is not Dostoevsky because he was extraordinarily proficient at generating grammatical Russian prose. Plenty of people could do that. What mattered was the fantastically peculiar configuration of obsessions that made The Brothers Karamazov seem necessary to him in the first place.

This is also why the “interchangeable vessels” line strikes me as almost exactly backwards. The more abundant generic intellectual competence becomes, the more valuable non-interchangeability may become. If everybody has access to something capable of producing a polished 2,000-word essay on Augustine, then “can produce polished prose about Augustine” stops distinguishing anyone. What remains distinctive is: Why Augustine? Why this passage? Why connect it to something your warehouse supervisor said yesterday? Why does that remind you of an obscure Tolstoy diary entry? Why do you find the conventional interpretation unsatisfactory? Why make a joke about forklifts halfway through?

There was a smart-ass kid in my Philosophy 101 class who argued that science could answer all philosophical questions better than philosophy could. My professor’s quip was that science couldn’t answer the question “Why?” Decades ago, when LLMs were still science fiction, that fault line was already obvious to anyone who looked. Science, technology, AI may quickly and effectively answer the question “Given X, how do I achieve Y?” They may have less to say in response to “Why Y?”

If I had to speculate on future trends, I’d suggest that we are probably headed toward a world in which competence is cheap and personality is expensive. And, amusingly, that may make the intellectual oddball rather more valuable, not less. The person who can write the 400th respectable essay about “AI and the Future of Humanity” is in trouble. The lunatic who suddenly realizes that prednisone, French symbolist decadence, and the ability to easily put on one’s socks belong in the same sonnet still has a job.

Most of these big-picture overviews of AI deal in ventose generalizations rather than the interesting particulars. This makes the genre frustrating and nearly useless to me. A sentence like Conti’s “we are already well on our way to becoming mere interchangeable vessels for ChatGPT’s accomplishments and insights” is nearly evidence-proof. What observation would establish that people had become “interchangeable vessels”? What observation would disprove it? The phrase creates an atmosphere rather than identifying a mechanism.

And because so much AI criticism operates at that altitude, it has developed a remarkably standardized inventory of anxieties: loss of authorship, loss of authenticity, deskilling, homogenization, dependence, replacement, alienation, loss of meaning, death of creativity. The nouns change places, but you can often predict the next paragraph. There is something genuinely funny about essays warning that AI will produce generic thinking while themselves reproducing a highly recognizable genre template.

Ink Libel

By in ai, books, extraordinary popular delusions, media/propaganda, the madness of crowds

Pulping our humanity!

Horrific destruction!

A CULTURAL CRIME AGAINST HUMANITY!

Wait, what’s that?

But contrary to many news reports, the old books bought from these sellers do not appear to be “rare” in the typical sense of the word. According to Becker and the online discussions I read, the books in these orders are almost entirely nonfiction, were mostly published from the 1970s to the 1990s, and were written in a variety of languages. Becker wrote in a blog post that he has received orders for “stuff like The Insider’s Guide to Metro Denver from 1995 or How to Use Corel WordPerfect 1991.” These belong to a category of books that you could call rare because a small number of copies remain, but they may also be out-of-date, undesirable to readers, and not valuable per se. In some cases, booksellers have noted, the cost of shipping exceeds the books’ resale value.

 

Oh.

Boy, do we look foolish! If only some idiot blogger had kept his head while the rest of us were losing ours!

Well, mistakes were certainly made — and amplified, and magnified — but extremism in defense of The 1995 Insider’s Guide to Metro Denver is no vice, as I believe Erasmus might have said. At any rate, I’m sure we’ve all learned just enough to save face without forcing us to question our priors — so, no harm, no foul; all’s well that ends well!

I don’t want to let the Atlantic fully off the hook, though:

Regardless, the idea of an AI behemoth destroying books has sent a bolt through the collective psyche. Book destruction is historically linked to authoritarian regimes and attempts to control public discourse, as well as ancient tragedies—the burning of the Library of Alexandria comes to mind. The encroachment of AI, particularly in the realm of books, is now everyday news: Last month it emerged that a top-selling Kindle book may have been partly written by AI, and grandparents are buying children’s books filled with AI-generated slop.

Even if a literal book burning is not under way, the process of ingesting millions of books, stripping them of authorship, and blending them into a homogenous “intelligence” branded with the name of a chatbot certainly feels destructive. It masks the hard work and collaboration that goes into knowledge creation, undermines incentives for authors to write books, prevents experts from finding one another, and gives AI companies tremendous power over what information people can access. Given this scale of cultural damage, the physical destruction of books seems almost quaint.

“Regardless” here is doing the work of an emergency generator after the factual power supply has failed. The current it supplies is too weak to illuminate anything, but strong enough to animate a corpse. Yes, a “bolt” did indeed shoot through the collective psyche: the mad scientists of viral outrage sent it there. They had stitched together a monster from misshapen rumors, decaying conspiracies, rotting scraps of metaphor, and adjectives scavenged from dead headlines; once electrified, it lumbered off through the global village, bellowing about antique books, pulped humanity, and crimes against mankind.

This is an embarrassing effort to salvage a virtue out of a moral panic. If I may take some creative liberties in paraphrasing, “We investigated the burning of Alexandria and found that Alexandria probably isn’t burning. But the fact that people imagined it burning reveals a deeper fire.” No, what it reveals is that your audience consists of gullible cretins who should be ashamed of themselves. I’d like to see a Venn diagram of the people forever warning us about “misinformation” and the people who instantly, uncritically believed that AI was devouring “ultra-rare” books. I suspect it would actually be a perfect circle.

(Misinformation, n. The stupid things other people believe. The corresponding belief among one’s friends is called context, raising awareness, or an important conversation.)

I’m talking about you, 404media. I’m talking about you, Ted “The Honest Broker” Gioia. I’m talking about you, Futurism. I’m talking about you, TechRadar. And you, Ryan Fleck. You, Guardian. You, Hyperallergic. Oh, there were so many of you. And let us not forget the supporting cast among the Bandar-log: the minor imbeciles, volunteer panic merchants, ambulatory rumor mills, and unpaid interns of hysteria sharing their infectious outrage on Substack Notes, TikTok, Facebook Reels, and sundry other sewage dispensers. Your appalling ignorance is rivaled only by your overweening self-righteousness. You squeezed out fake tears over library books that had not been checked out since the Reagan administration. You rent your garments over worthless monographs that had been abandoned at Goodwill and remained unbought. Were you fooled, or were you doing the fooling? Did you mistake wishful thinking for corroboration, or merely trust that your readers would? (Take your time. Neither answer helps.) The slightest ounce of skepticism or simple patience would have sufficed; instead, you mistook your priors for evidence, your indignation for inquiry, your metaphors for facts, and one another’s repetition for independent confirmation. You stand amidst a mess of fabrications and aver that your heart, at least, was in the right place, even if your brain was AWOL. You embroidered rumor, embellished speculation, and passed the resulting monstrosity down the line with the solemnity of monks preserving the Gospels, only you produced something more like The Protocols of the Techbros of Silicon Valley. The online environment is, incredibly, an even stupider, uglier place today because of your malign efforts.

In a just world, you would cancel your internet service, surrender your passwords, and sew your mouths shut. If you possessed even the vestigial stump of integrity, you would renounce all “hot takes” and wander for years like so many Ancient Mariners of epistemic disgrace, buttonholing wedding guests, commuters, grocery clerks, and innocent diners, forcing each in turn to hear the tale of your sordid imbecility, your appalling groupthink. You would conclude by saying, plainly and without qualification: “I was wrong because I was too stupid to question a story telling me exactly what I wanted to hear.” But this is not a just world. So tomorrow you will return, chastened by nothing, confidence undiminished, to pronounce upon some fresh catastrophe you first encountered nine minutes earlier. I see you, though, and I know what you are.

1 2 3 4›»