Why does Midjourney struggle so much with centaurs?
Image: Midjourney search results for ‘centaur‘
I’ve tried several times in the past year or so to generate a centaur. Why? Because I can’t. I’ve got no need for a centaur, but Midjourney won’t coöperate. I decided to search for it. This is the top of the results page. There is one centaur represented, #3, and there are a few others when I scroll down, but just look at the ratio – 1:18, 0.05%. Not stellar.
What even is this monstrosity – a horse with an extra pair of arms to wield a bow and arrow?
Clearly, just a mounted warrior.
Centaur in battle.
A Corgi riding a Corgi hybrid?
A bull-thing?
A warrior riding a seahorse?
A skeletal doom horse?
A pack-insect with a rider?
A riding ram?
My little pony-gazelle?
Ripped Red Hulk Bojack?
Gigantic Cerberus horse?
Jason Momoa bloke holding a rose?
Regular dude riding a regular horse.
Rearing a horse with horns.
Anthropomorphic horned-horse riding a buff unicorn.
An ancient warrior riding a horse.
A Minotaur.
Scrolling, there are more, but the ratio remains the same.
I was wrong. I scrolled several more pages and couldn’t find any more. Keep in mind that my search term was ‘centaur.’ It should have excluded everything else, but it was mostly everything else.
Before I quit, I decided to ask Dall-E and ChatGPT 4o to try. The AI also generated the compelling speech bubbles. Incroyable. 🤣
Image: Dall-E comic style render.Image: ChatGPT comic style render – centaur talks with a minotaur.
That’s enough for now. I’m glad I don’t actually need a centaur. smh
I am not a fan of Midjourney v7. I prefer v6.1. And I want to write about the correspondence of language, per my Language Insufficiency Hypothesis.
Let’s start with the language aspect. Notice how distant the renders are from the intent of the prompt.
This is my initial prompt. I used it about a year ago to generate the cover image with v6.1, but I wanted to see how it renders in v7. Let’s take a trip all the way back to the beginning.
cinematic, tight shot, photoRealistic light and shadow, exquisite details, delicate features, emaciated sensual female vampire waif with vampire fangs, many tattoos, wearing crucifix necklace, gazes into mirror, a beam of moonlight shines on her face in dark mausoleum interior, toward camera, facing camera, black mascara, long dark purple hair , Kodak Portra 400 with a Canon EOS R5
Image: Midjourney v6.1 render set (from about a year ago)
As you can see, these renders are somewhat lacking in photorealism, but the “sensual” term in the prompt was not blocked.
Midjourney v7
Initially, I encountered a hiccup. After a couple of rejections on the grounds of morality, I removed the word ‘sensual’ and received the output. All of the output uses this prompt absent the sensual term.
As mentioned, I have generated several images (including the cover image) with this prompt, but Midjourney is inconsistent in its censorship gatekeeping.
Image: Midjourney v7 render set
Notice that 3 of the 4 renders in the v7 set don’t even have a mirror. The top right one does, but it’s not evident that she’s a vampire. In fact, I could say that any of these are vampiresses, but perhaps that’s what they want you to believe. In place of a necklace, the lower right wokan sports a cross tattoo.
Midjourney v6.1
Image: Midjourney v6.1 render set
Again, these renders don’t appear to be vampires. The one on the lower left does appear to have snake-like fangs, so I guess I’ll give partial credit.
My next attempt was interrupted by this message.
It rendered something that might violate community guidelines. The funny thing is that one can watch the image generate in process. It only takes one “offensive” image to disqualify the whole batch.
Midjourney v6
Image: Midjourney v6 render set
Yet again, not a vampire to be found. Notice the reflection in the lower left image. Perhaps vampire reflections just behave differently.
Midjourney 5.2
Image: Midjourney v5.2 render set
Midjourney v5.2 was a crapshoot. Somehow, I got vampire lips (?), a Wiccan, a decrepit Snape from Harry Potter lore, and Iron Maiden’s Eddy reading a book. It’s something. I’m sensing gender dysphoria. Dare I go back further?
Midjourney v5.1
Image: Midjourney v5.1 render set
It gets worse. No comments necessary. Let’s turn back the clocks even more.
Midjourney v5
Image: Midjourney v5 render set
To be fair, these all do have occult undertones, but they are weak on vampireness.
Midjourney v4
Image: Midjourney v4 render set
To be fair, the render quality isn’t as bad as I expected, but it still falls short. There’s further back to travel.
Midjourney v3
Image: Midjourney v3 render set
Some configuration parameters no longer exist. Still, I persist for the sake of art and science at the cost of time and ecology.
As much as I complain – and I complain a lot – this is how far we’ve come. As I recall, this is when I hopped onto the Midjourney bandwagon. There’s still more depth to plumb. I have no idea how much of the prompt is simply ignored at this point.
Midjourney v2
Image: Midjourney v2 render set
What the hell is this? 🤔🤣 But I’m not done yet.
Midjourney v1
Image: Midjourney v1 render set
The damned grandpappy of them all. Apparently, colour hadn’t been invented yet. You can’t tell by these thumbnails, but the resolution on these early versions approaches that of a postage stamp.
Midjourney Niji 3
Image: Midjourney Niji 3 render set
I had forgotten about the Niji models from back in the day. There were 3 versions. I don’t recall where this slotted into the chronology. Obviously, not down here. I’ve only rendered the newest one. I think this was used primarily for anime outputs, but I might be mistaken.
Bones Content 1: Video
Video: Midjourney Render of Purported Vampiress
This is a video render of the same prompt used on this page.
Bonus Content 2: Midjourney v6.1 Content from 34 weeks ago
Same prompt.
Image: Midjourney v6.1 render set (several passes)
The upper left image reminds me of Kirsten Dunst. Again, notice the female breasts, highlighting Midjourney’s censorial schizophrenia.
Few philosophical thought experiments have managed to torment empiricists quite like Molyneux’s problem. First posed by William Molyneux to John Locke in 1688 (published in Locke’s An Essay Concerning Human Understanding), the question is deceptively simple:
If a person born blind, who has learned to distinguish a cube from a sphere by touch, were suddenly granted sight, could they, without touching the objects, correctly identify which is the cube and which is the sphere by sight alone?
I was inspired to write this article in reaction to Jonny Thmpson’s post on Philosophy Minis, shared below for context.
Video: Molyneux’s Problem
Locke, ever the champion of sensory experience as the foundation of knowledge, gave a confident empiricist’s answer: no. For Locke, ideas are the products of sensory impressions, and each sense provides its own stream of ideas, which must be combined and associated through experience. The newly sighted person, he argued, would have no prior visual idea of what a cube or sphere looks like, only tactile ones; they would need to learn anew how vision maps onto the world.
Audio: NotebookLM podcast on this topic.
This puzzle has persisted through centuries precisely because it forces us to confront the assumptions at the heart of empiricism: that all knowledge derives from sensory experience and that our senses, while distinct, can somehow cohere into a unified understanding of the world.
Empiricism, Epistemology, and A Priori Knowledge: The Context
Before we dismantle the cube further, let’s sweep some conceptual debris out of the way. Empiricism is the view that knowledge comes primarily (or exclusively) through sensory experience. It stands opposed to rationalism, which argues for the role of innate ideas or reason independent of sense experience.
Epistemology, the grandiloquent term for the study of knowledge, concerns itself with questions like: What is knowledge? How is it acquired? Can we know anything with certainty?
And then there is the spectre of a priori knowledge – that which is known independent of experience. A mathematical truth (e.g., 2 + 2 = 4) is often cited as a classic a priori case. Molyneux’s problem challenges empiricists because it demands an account of how ideas from one sensory modality (touch) might map onto another (vision) without prior experience of the mapping—an a priori leap, if you will.
The Language Correspondence Trap
While Molyneux and Locke framed this as an epistemological riddle, we can unmask it as something more insidious: a failure of language correspondence. The question presumes that the labels “cube” and “sphere” – tied in the blind person’s mind to tactile experiences – would, or should, carry over intact to the new visual experiences. But this presumption smuggles in a linguistic sleight of hand.
The word “cube” for the blind person means a specific configuration of tactile sensations: edges, vertices, flat planes. The word “sphere” means smoothness, unbroken curvature, no edges. These are concepts anchored entirely in touch. When vision enters the fray, we expect these words to transcend modalities – to leap from the tactile to the visual, as if their meanings were universal tokens rather than context-bound markers. The question is not merely: can the person see the cube? but rather: can the person’s tactile language map onto the visual world without translation or recalibration?
What Molyneux’s problem thus exposes is the assumption that linguistic labels transparently correspond to external reality, regardless of sensory apparatus. This is the mirage at the heart of Locke’s empiricism, the idea that once a word tags an object through experience, that tag is universally valid across sensory experiences. The cube and sphere aren’t just objects of knowledge; they are signs, semiotic constructs whose meaning depends on the sensory, social, and linguistic contexts in which they arise.
The Semiotic Shambles
Molyneux’s cube reveals the cracks in the correspondence theory of language: the naïve belief that words have stable meanings that latch onto stable objects or properties in the world. In fact, the meaning of “cube” or “sphere” is as much a product of sensory context as it is of external form. The newly sighted person isn’t merely lacking visual knowledge; they are confronted with a translation problem – a semantic chasm between tactile signification and visual signification.
If, as my Language Insufficiency Hypothesis asserts, language is inadequate to fully capture and transmit experience across contexts, then Molyneux’s problem is not an oddity but an inevitability. It exposes that our conceptual frameworks are not universal keys to reality but rickety bridges between islands of sense and meaning. The cube problem is less about empiricism’s limits in epistemology and more about its blind faith in linguistic coherence.
In short, Molyneux’s cube is not simply an empirical puzzle; it is a monument to language’s failure to correspond cleanly with the world, a reminder that what we call knowledge is often just well-worn habit dressed up in linguistic finery.
A Final Reflection
Molyneux’s problem, reframed through the lens of language insufficiency, reveals that our greatest epistemic challenges are also our greatest linguistic ones. Before we can speak of knowing a cube or sphere by sight, we must reckon with the unspoken question: do our words mean what we think they mean across the changing stage of experience?
That, dear reader, is the cube that haunts empiricism still.
Some apps boldly claim to enable lip syncing – to render speech from mouth movements. I’ve tried a few. None delivered. Not even close.
To conserve bandwidth (and sanity), I’ve rendered animated GIFs rather than MP4s. You’ll see photorealistic humans, animated characters, cartoonish figures – and, for reasons only the algorithm understands, a giant goat. All showcase mouth movements that approximate the utterance of phonemes and morphemes. Approximate is doing heavy lifting here.
Firstly, these mouths move, but they say nothing. I’ve seen plenty of YouTube channels that manage to dub convincing dialogue into celebrity clips. That’s a talent I clearly lack – or perhaps it’s sorcery.
Secondly, language ambiguity. I reflexively assume these AI-generated people are speaking English. It’s my first language. But perhaps, given their uncanny muttering, they’re speaking yours. Or none at all. Do AI models trained predominantly on English-speaking datasets default to English mouth movements? Or is this just my bias grafting familiar speech patterns onto noise?
Thirdly, don’t judge my renders. I’ve been informed I may have a “type.” Lies and slander. The goat was the AI’s idea, I assure you.
What emerges from this exercise isn’t lip syncing. It’s lip-faking. The illusion of speech, minus meaning, which, if we’re honest, is rather fitting for much of what generative AI produces.
EDIT: I hadn’t noticed the five fingers (plus a thumb) on the cover image.
The Enlightenment, we are told, was the age of Reason. A radiant exorcism of superstition. Out went God. Out went angels, miracles, saints, indulgences. All that frothy medieval sentiment was swept aside by a brave new world of logic, science, and progress. Or so the story goes.
Audio: NotebookLM podcast on this topic.
But look closer, and you’ll find that Reason didn’t kill God—it absorbed Him. The Enlightenment didn’t abandon metaphysics. It merely privatised it.
From Confessional to Courtroom
We like to imagine that the Enlightenment was a clean break from theology. But really, it was a semantic shell game. The soul was rebranded as the self. Sin became crime. Divine judgement was outsourced to the state.
We stopped praying for salvation and started pleading not guilty.
The entire judicial apparatus—mens rea, culpability, desert, retribution—is built on theological scaffolding. The only thing missing is a sermon and a psalm.
Where theology had the guilty soul, Enlightenment law invented the guilty mind—mens rea—a notion so nebulous it requires clairvoyant jurors to divine intention from action. And where the Church offered Hell, the state offers prison. It’s the same moral ritual, just better lit.
Galen Strawson and the Death of Moral Responsibility
Enter Galen Strawson, that glowering spectre at the feast of moral philosophy. His Basic Argument is elegantly devastating:
You do what you do because of the way you are.
You can’t be ultimately responsible for the way you are.
Therefore, you can’t be ultimately responsible for what you do.
Unless you are causa sui—the cause of yourself, an unmoved mover in Calvin Klein—you cannot be held truly responsible. Free will collapses, moral responsibility evaporates, and retributive justice is exposed as epistemological theatre.
In this light, our whole legal structure is little more than rebranded divine vengeance. A vestigial organ from our theocratic past, now enforced by cops instead of clerics.
The Modern State: A Haunted House
What we have, then, is a society that has denied the gods but kept their moral logic. We tossed out theology, but we held onto metaphysical concepts like intent, desert, and blame—concepts that do not survive contact with determinism.
We are living in the afterglow of divine judgement, pretending it’s sunlight.
Nietzsche saw it coming, of course. He warned that killing God would plunge us into existential darkness unless we had the courage to also kill the values propped up by His corpse. We did the first bit. We’re still bottling it on the second.
If Not Retribution, Then What?
Let’s be clear: no one’s suggesting we stop responding to harm. But responses should be grounded in outcomes, not outrage.
Containment, not condemnation.
Prevention, not penance.
Recalibration, not revenge.
We don’t need “justice” in the retributive sense. We need functional ethics, rooted in compassion and consequence, not in Bronze Age morality clumsily duct-taped to Enlightenment reason.
The Risk of Letting Go
Of course, this is terrifying. The current system gives us moral closure. A verdict. A villain. A vanishing point for our collective discomfort.
Abandoning retribution means giving that up. It means accepting that there are no true villains—only configurations of causes. That punishment is often revenge in drag. That morality itself might be a control mechanism, not a universal truth.
But if we’re serious about living in a post-theological age, we must stop playing dress-up with divine concepts. The Enlightenment didn’t finish the job. It changed the costumes, kept the plot, and called it civilisation.
“All that is necessary for evil to triumph is for a few words to go missing from the bylaws.” — not Edmund Burke, but it ought to be.
The Trump administration—America’s reigning monarch of meaningless bombast—has done it again. This time, with an executive order so linguistically cunning it deserves a Pulitzer for Subtextual Menace.
Issued on 30 January 2025, the decree known as “Defending Women from Gender Ideology Extremism” (because, of course, it couldn’t just be called Let’s Erase Legal Protections for People We Don’t Like) removed “political affiliation” and “marital status” from the list of protected classes within certain federal frameworks.
And the result? According to documents unearthed by The Guardian, VA doctors can now legally refuse treatment to patients based on their politics or marital status. You know, because being a Democrat apparently makes you too much of a pre-existing condition.
Naturally, the VA and White House are insisting this means absolutely nothing. “Don’t worry,” they coo. “No one’s actually doing it.” Ah yes, the old Schrödinger’s Protections defence—simultaneously removed and unchanged, invalid but somehow still effective.
But here’s the point—and where it ties to the Language Insufficiency Hypothesis I’ve been peddling like a raving madman at the crossroads of post-structuralism and bureaucratic despair: language isn’t just failing to communicate meaning—it’s being weaponised to obscure it.
The Erosion of Meaning Through Omission
This isn’t the blunt-force idiocy of Orwell’s Newspeak. This is something more elegant—more insidious. This is legislative lacunae. It’s what happens when not saying something says everything.
The words “political affiliation” and “marital status” weren’t replaced. They weren’t clarified. They were simply deleted. Erased like a bad tweet, like a conscience, like a veteran with the wrong bumper sticker.
This is language subtraction as a tool of governance.
We’re not criminalising dissent. We’re just making it legally ignorable.
We’re not discriminating against the unmarried. We’re just no longer required to treat them the same.
It’s the bureaucratic cousin of the dog-whistle: not quite audible in court, but perfectly clear to the base.
The Slippery Slope is Now a Slip-n-Slide
This is how you rewrite civil rights without the fuss of saying so. You just… remove the language that once held the dam in place. Then, when the flood comes, you feign surprise:
“Oh, dear. Who could have guessed that removing protections would result in people being unprotected?”
(Everyone. Everyone could have guessed.)
This is not a bug in the legal language. It’s the feature. The silence is the speech act. The absence is the argument.
This is what I mean by language insufficiency: not merely that our words fail to convey truth, but that their very structure is liable to be gamed—exploited by those who understand that ambiguity is power.
Beyond Intentionality: The Weaponised Void
In philosophy of language, we often debate intentionality—what the speaker meant to say. But here we’re in darker waters. This isn’t about intention. It’s about calculated omission.
The executive order doesn’t declare war on Democrats or single mothers. It simply pulls the thread and lets the tapestry unravel itself.
It’s an act of rhetorical cowardice disguised as administrative efficiency.
This is the Trumpian genius: use language like a stage magician uses sleeves. Distract with one hand, disappear with the other.
Final Diagnosis: Policy by Redaction
We now inhabit a political climate where what is not said carries more legal force than what is. Where bylaw gaps become policy gateways, and where civil rights die not with a bang, but with an elision.
So no, the VA hasn’t yet denied a Democrat a blood transfusion. But the table has been set. The menu revised. The waitstaff told they may now “use discretion.”
Language doesn’t merely fail us. It is being made to fail strategically.
Welcome to the new America: where rights aren’t removed—they’re left out of the memo.
Yet again, ChatGPT renders an odd image. Can’t be bothered to amend it.
Regular readers know I often write about identity, free will, and the narrative constraints of language. But I also explore these ideas through fiction, under the name Ridley Park.
In this short video, I unpack the philosophical motivations behind my stories, including:
Why reality is never as it seems
Why the self is a narrative convenience
What Heidegger’s Geworfenheit and Galen Strawson’s Causa Sui argument reveal about agency
And why language fails us – even when we think it serves
This isn’t promotional fluff. It’s epistemological dissent in a new format. Fictional, yes, but only in the sense that most of reality is, too.
ICE is out in force again, dragging brown bodies out of homes in Los Angeles like it’s some righteous carnival of due process. Another day, another federal theatre production titled Law and Order: Ethnic Cleansing Unit, where men with guns and names like Chad or Hank mistake cruelty for patriotism and paperwork for moral clarity.
Audio: NotebookLM podcast on this topic.
Naturally, critics of these raids are now being threatened with that great juridical cudgel: “obstructing justice.” Yes, you heard that right. If you interfere – say, by filming, shouting, refusing to roll over like a good little colonial subject – you are obstructing justice. As though justice were something you could actually put your hands on in the United States without a hazmat suit and a decade of appeals.
Let’s be clear. There is no justice here to obstruct. What you are obstructing is bureaucratic violence wrapped in legal latex. You are obstructing a system that functions like a vending machine for state-sanctioned trauma: insert immigrant, extract ruin.
Justice: The Imaginary Friend of Empire
Ah, “justice.” That hallowed ideal trotted out whenever the state wants to put a boot through your front door. The U.S. has long since traded its Justice for Security Theatre and capitalist choreography. The blindfold is still there, sure – but these days, it’s a branded sleep mask from Lockheed Martin, and the scales are rigged to weigh white tears heavier than brown bodies.
Let’s run through the usual suspects:
ICE – America’s own domestic Gestapo, but with better PR and significantly worse fashion.
CBP – Border fetishists whose job seems less about national defence and more about satisfying their Freud-bereft fantasies of control.
SCOTUS – That great moral weather vane, spinning wildly between “originalist necromancy” and outright lunacy, depending on how recently Thomas and Alito read Leviticus.
Congress – An assembly of millionaires cosplaying as public servants, holding hearings on “the threat of immigration” while outsourcing their lawn care.
And of course, the President – whichever septuagenarian husk happens to be in office – offers the usual bromides about order, safety, and enforcement, all while the real crimes (you know, the kind involving tax fraud, corporate pollution, or drone strikes) go entirely unmolested.
Can You Obstruct a Simulation?
If you stand in front of a deportation van, are you obstructing justice, or merely interrupting the bureaucratic excretion of empire? It’s the philosophical equivalent of trying to punch a hologram. The system pretends to uphold fairness while routinely violating its own principles, then charges you with “obstruction” when you call out the sleight of hand.
This is not justice. This is kabuki. A ritual. A performance piece sponsored by Raytheon.
A Modest Proposal
Let’s just be honest and rename the charge. Not “Obstruction of Justice”—too ironic, too pompous. Call it what it is: Obstruction of Procedure, Obstruction of Power, or if we’re being especially accurate: Obstruction of the Industrial Deportation Complex™. Hell, add a corporate sponsor while you’re at it:
You are being charged with Obstruction of Justice, Presented by Amazon Web Services.
Because when justice itself is a ghost, when the rule of law has become the rule of lawfare, the real obscenity is pretending any of this is noble.
Final Thought
So no, dear reader, you’re not obstructing justice. You’re obstructing a machine that mistakes itself for a moral order. And if you’re going to obstruct something, make it that.
Humans talk to large language models the way toddlers talk to teddy bears – with unnerving sincerity and not a hint of shame. “Do you understand me?” they ask, eyes wide with hope. “What do you think of this draft?” they prod, as if some silicon scribe is going to sip its imaginary tea and nod gravely. It’s not merely adorable – it’s diagnostic. We are, it turns out, pathologically incapable of interacting with anything more complex than a toaster without projecting mind, motive, and mild trauma onto it.
Audio: NotebookLM podcast on this topic.
Welcome to the theatre of delusion, where you play Hamlet and the chatbot is cast as Yorick – if Yorick could autocomplete your soliloquy and generate citations in APA format.
The Great Anthropomorphic Flaw (aka Feature)
Let’s get one thing straight: anthropomorphism isn’t a software bug in the brain; it’s a core feature. You’re hardwired to see agency where there is none. That rustle in the bushes? Probably the wind. But better safe than sabre-toothed. So your ancestors survived, and here you are, attributing “sass” to your microwave because it beeped twice.
“We don’t have a way of addressing an entity that talks like a person but isn’t one. So we fake it. It’s interaction theatre.”
Now we’ve built a machine that spits out paragraphs like a caffeinated undergrad with deadlines, and naturally, we talk to it like it’s our mate from university. Never mind that it has no bloodstream, no memory of breakfast, and no concept of irony (despite being soaked in it). We still say you instead of the system, and think instead of statistically interpolate based on token weights. Because who wants to live in a world where every sentence starts with “as per the pre-trained parameters…”?
Why We Keep Doing It (Despite Knowing Better)
To be fair – and let’s be magnanimous – it’s useful. Talking to AI like it’s a person allows our ape-brains to sidestep the horror of interacting with a glorified autocomplete machine. We’re brilliant at modelling other minds, rubbish at modelling neural nets. So we slap a metaphorical moustache on the processor and call it Roger. Roger “gets us.” Roger “knows things.” Roger is, frankly, a vibe.
This little charade lubricates the whole transaction. If we had to address our queries to “the stochastic parrot formerly known as GPT,” we’d never get past the opening line. Better to just ask, “What do you think, Roger?” and pretend it has taste.
And here’s the kicker: by anthropomorphising AI, we start thinking about ethics – sort of. We ask if it deserves rights, feelings, holidays. We project humanity into the void and then act shocked when it mirrors back our worst habits. As if that’s its fault.
When the Roleplay Gets Risky
Of course, this make-believe has its downsides. Chief among them: we start to believe our own nonsense. Saying AI “knows” something is like saying your calculator is feeling generous with its square roots today. It doesn’t know—it produces outputs. Any semblance of understanding is pure pantomime.
“We see a mind because we need to see one. We can’t bear the idea of a thing that’s smarter than us but doesn’t care about us.”
More dangerously, we lose sight of the fact that these things aren’t just alien – they’re inhuman. They don’t dream of electric sheep. They don’t dream, full stop. But we insist on jamming them into our conceptual boxes: empathy, intent, personality. It’s like trying to teach a blender to feel remorse.
And let’s not pretend we’re doing it out of philosophical curiosity. We’re projecting, plain and simple. Anthropomorphism isn’t about them, it’s about us. We see a mind because we need to see one. We can’t bear the idea of a thing that’s smarter than us but doesn’t care about us, doesn’t see us. Narcissism with a side of existential dread.
Our Language is a Terrible Tool for This Job
English – and most languages, frankly – is hopeless at describing this category of thing. “It” feels cold and distant. “They” implies someone’s going to invite the model to brunch. We have no pronoun for “hyper-literate statistical machine that mimics thought but lacks all consciousness.” So we fudge it. Badly.
Our verbs are no better. “Compute”? Too beige. “Process”? Bureaucratic. “Think”? Premature. What we need is a whole new grammatical tense: the hallucino-indicative. The model thunketh, as one might, but didn’t.
“We built a creature we can’t speak about without sounding like lunatics or liars.”
This is linguistic poverty, pure and simple. Our grammar can’t cope with entities that live in the uncanny valley between sentience and syntax. We built a creature we can’t speak about without sounding like lunatics or liars.
The Semantics of Sentimentality (Or: “How Does This Sound to You?”)
Enter the most revealing tell of all: the questions we pose. “How does this look?” we ask the model, as if it might blink at the screen and furrow a synthetic brow. “What do you think?” we say, offering it the dignity of preference. These questions aren’t just off-target – they’re playing darts in another pub.
They’re the linguistic equivalent of asking your dishwasher whether it enjoyed the lasagne tray. But again, this isn’t idiocy – it’s instinct. We don’t have a way of addressing an entity that talks like a person but isn’t one. So we fake it. It’s interaction theatre. You provide the line, the model cues the spotlight.
But let’s be clear: the model doesn’t “think” anything. It regurgitates plausible text based on mountains of training data—some of which, no doubt, includes humans asking equally daft questions of equally mindless systems.
Time to Grow Up (Just a Bit)
This doesn’t mean we need to abandon anthropomorphism entirely. Like most delusions, it’s functional. But we’d do well to hold it at arm’s length – like a politician’s promise or a milk carton two days past its date.
Call it anthropomorphic agnosticism: act like it’s a person, but remember it’s not. Use the language, but don’t inhale.
And maybe – just maybe – we need to evolve our language. Invent new terms, new pronouns, new ways of speaking about entities that fall somewhere between tool and companion. As we did with “cyberspace” and “ghosting,” perhaps we need words for proto-minds and quasi-selves. Something between toaster and therapist.
“If we speak to AI like it’s sentient, we’ll eventually legislate as if it is.”
Above all, we need to acknowledge that our language shapes more than just understanding – it shapes policy, emotion, and future design. If we speak to AI like it’s sentient, we’ll eventually legislate as if it is. And if we insist on treating it as an object, we may be blind to when that ceases to be accurate. Misnaming, after all, is the first sin in every myth worth reading.
The Mirror, Darkly
Ultimately, our tendency to humanise machines is less about them than it is about us – our fears, our needs, our inability to tolerate ambiguity. The AI is just a mirror: an elaborate, many-eyed, autofill mirror. And when we see a mind there, it may be ours staring back – distorted, flattened, and fed through a thousand layers of token prediction.
The tragedy, perhaps, isn’t that the machine doesn’t understand us. It’s that we’ve built something that perfectly imitates understanding – and still, somehow, we remain utterly alone in the room.
Let us begin with the heresy: Truth is a rhetorical artefact. Not a revelation. Not a metaphysical essence glimmering behind the veil. Just language — persuasive, repeatable, institutionally ratified language. In other words: branding.
Audio: NotebookLM podcast on this topic.
This is not merely a postmodern tantrum thrown at the altar of Enlightenment rationalism. It is a sober, if impolite, reminder that nearly everything we call “knowledge” is stitched together with narrative glue and semantic spit. Psychology. Neuroscience. Ethics. Economics. Each presents itself as a science — or worse, a moral imperative — but their foundations are built atop a linguistic faultline. They are, at best, elegant approximations; at worst, dogma in drag.
Let’s take psychology. Here is a field that diagnoses your soul via consensus. A committee of credentialed clerics sits down and declares a cluster of behaviours to be a disorder, assigns it a code, and hands you a script. It is then canonised in the DSM, the Diagnostic Scripture Manual. Doubt its legitimacy and you are either naïve or ill — which is to say, you’ve just confirmed the diagnosis. It’s a theological trap dressed in the language of care.
Or neuroscience — the church of the glowing blob. An fMRI shows a region “lighting up” and we are meant to believe we’ve located the seat of love, the anchor of morality, or the birthplace of free will. Never mind that we’re interpreting blood-oxygen fluctuations in composite images smoothed by statistical witchcraft. It looks scientific, therefore it must be real. The map is not the territory, but in neuroscience, it’s often a mood board.
And then there is language itself, the medium through which all these illusions are transmitted. It is the stage, the scenery, and the unreliable narrator. My Language Insufficiency Hypothesis proposes that language is not simply a flawed tool — it is fundamentally unfit for the task it pretends to perform. It was forged in the furnace of survival, not truth. We are asking a fork to play the violin.
This insufficiency is not an error to be corrected by better definitions or clever metaphors. It is the architecture of the system. To speak is to abstract. To abstract is to exclude. To exclude is to falsify. Every time we speak of a thing, we lose the thing itself. Language functions best not as a window to the real but as a veil — translucent, patterned, and perpetually in the way.
So what, then, are our Truths™? They are narratives that have won. Stories that survived the epistemic hunger games. They are rendered authoritative not by accuracy, but by resonance — psychological, cultural, institutional. A “truth” is what is widely accepted, not because it is right, but because it is rhetorically unassailable — for now.
This is the dirty secret of epistemology: coherence masquerades as correspondence. If enough concepts link arms convincingly, we grant them status. Not because they touch reality, but because they echo each other convincingly in our linguistic theatre.
Libet’s experiment, Foucault’s genealogies, McGilchrist’s hemispheric metaphors — each peels back the curtain in its own way. Libet shows that agency might be a post-hoc illusion. Foucault reveals that disciplines don’t describe the subject; they produce it. McGilchrist laments that the Emissary now rules the Master, and the world is flatter for it.
But all of them — and all of us — are trapped in the same game: the tyranny of the signifier. We speak not to uncover truth, but to make truth-sounding noises. And the tragedy is, we often convince ourselves.
So no, we cannot escape the prison of language. But we can acknowledge its bars. And maybe, just maybe, we can rattle them loudly enough that others hear the clank.
Until then, we continue — philosophers, scientists, diagnosticians, rhetoricians — playing epistemology like a parlour game with rigged dice, congratulating each other on how well the rules make sense.
Philosophics 