#ai and basic competence

[dominaexmachina]

The Label Is the Work

There is a study. Duke University, published 2023. Researchers showed participants a set of paintings and labeled them either “human-created” or “AI-created.” The twist: every single painting in the set was AI-generated. Same images. Different labels. Different ratings — across liking, beauty, profundity, and perceived worth — every time, in favor of the “human” ones. The effect wasn’t visual. It was entirely top-down: the label rewired the experience of the object itself.

This is not a niche academic finding. This is a description of how aesthetic judgment actually works for most people, most of the time. You don’t encounter a work in a vacuum. You encounter it with a story attached. The story is part of the work.

Which makes the following cases interesting.

2023. Miles Astray, photographer. He submitted a real photograph of a flamingo — taken on film, no AI involved — to a photography competition’s AI-generated category. To make a point. The image won both jury and public vote. Contest organizers disqualified it when Astray revealed it was real. The photo was too good to be human, apparently. Until it was.

2023. Suzi Dougherty, Sydney. She shot a photograph of her son mid-stride in a Gucci exhibition, mannequins in the background. Entered a local photo contest. Disqualified for being AI-generated. The judge said the image felt “too perfect to be true”. It was a phone photo taken in ten minutes. The mannequins read as uncanny. The son was just photogenic.

2025. The Velvet Sundown. AI folk rock band. Over a million Spotify streams before anyone publicly questioned the source. People added the music to playlists. Then someone noticed the promo photos looked off. Not the music — the photos. The music still sounded the same after the reveal. Only people’s relationship to it changed.

Three cases. Three different directions of the same error. Human work flagged as AI. AI work enjoyed as human. The detector isn’t running on aesthetics — it’s running on suspicion, context, and whatever story arrived first.

The post from Tumblr says AI is always noticeable. Saturation gives it away. Movement gives it away. It’s just so easy to spot.

The data says otherwise. The data says you spot what you’re looking for, and you find what you expect.

[dominaexmachina]

Sometimes this discourse gets boring because it’s repetitive.

The same slogans, the same panic, the same moral shortcuts — recycled endlessly with minor cosmetic changes.

And then you stumble on something like this.

Something so perfectly self-revealing that you couldn’t design it on purpose. Not as satire. Not as parody. Not even as a thought experiment. Nobody forced this argument out. Nobody threatened anyone. No substances involved.

Just a person, sincerely typing, convinced they’re making a strong ethical point — and accidentally exposing an entire worldview in the process.

That’s when the boredom disappears. Because at that point, you’re no longer arguing with opinions. You’re observing a phenomenon.

And it deserves a closer look:

This is genuinely fascinating.

Apparently, artists who release their work into the public domain or under Creative Commons are now “class traitors”.

Let’s pause on that for a second.

According to this logic:

sharing your work freely is betrayal generosity is sabotage accessibility is violence and choice is immoral

If you don’t monetise every output, you’re not just naïve — you’re actively harming your peers.

This is not an argument against AI. This is not even an argument about labour.

This is an argument against freedom.

The position boils down to something very simple:

Art should only exist behind a paywall. If it doesn’t, it is unethical.

Which is… bold. And historically illiterate.

Public domain art. Creative Commons. Open-source culture. Free fonts. Free textures. Free brushes. Free tutorials. Academic publishing. Fan art. Zines. Modding communities.

All of these apparently should be “ashamed of themselves”.

Because someone, somewhere, might not get paid. I’ve already written about this logic before.

What’s actually being expressed here is not care for artists.

It’s hostility toward autonomy.

The idea that an artist can decide:

how their work is used whether it’s monetised whether it’s shared freely whether they value reach, community, experimentation, or joy

is deeply offensive to people who need a single moral rule to police everyone else.

If your worldview collapses the moment someone says:

“I’m okay with not charging for this”

then the problem is that you’ve reduced art to a toll booth.

The irony is delicious.

These same people will:

pirate movies, download games, use free software, rely on open-source tools, consume endless free content

…but the moment an artist consciously chooses to give something away?

Suddenly it’s class warfare.

And here’s the part that really matters:

This argument quietly reveals the endgame.

Not traditional:

“protect artists” “support creativity” “build sustainable systems”

But:

force scarcity, remove choice, and shame anyone who doesn’t comply

That’s coercion.

So no — AI didn’t take us “too far”. This did.

And yes — somehow, incredibly, this is even lower than “using AI kills people”.

Congratulations. New basement unlocked.

[dominaexmachina]

AI, microwaves, and other things you don’t understand

While some people are still busy writing thinkpieces about how “using AI is just microwaving frozen food instead of cooking,” other people quietly moved on and started doing something terrifyingly adult: they study what actually happens when humans create with machines.

That’s the part the anti-AI kids never touch. They’re fluent in culinary metaphors, allergic to data. They can tell you that “real art” must be hand-made, but they can’t be bothered to look at what changes when you put a generative model into the creative loop and watch the process instead of moralizing about it.

One of my favorite phrases to come out of this research landscape is brutally simple: the future belongs to symbiants — humans who know how to co-create with AI instead of larping as monks guarding the last analog brush. Not “prompt monkeys”, not passive consumers of machine output, but people who treat these systems as volatile, powerful collaborators and learn to direct them.

And when you actually look at the findings, the picture is even more offensive to the “microwave” crowd than any corporate marketing could ever be.

First: AI doesn’t just amplify the already-brilliant elite. It disproportionately lifts up people with mid-level or uneven skills. The ones who have taste but lack speed; who have ideas but not enough technique; who can see the direction but can’t brute-force their way there by hand in time. Give them generative tools, and suddenly their floor rises. Their work becomes usable, presentable, competitive. The “natural talents” are no longer the only ones allowed to speak.

Of course the gatekeepers hate that. If your entire status rests on the fact that you survived a long, expensive initiation ritual, nothing is more threatening than a tool that lets the uninitiated bypass half the corridor and start experimenting at a higher level on day one.

Second: generative systems don’t kill imagination, they pour gasoline on it. Anyone who has actually done iterative work with these models knows the pattern: you start with one idea, get a batch of outputs, and instead of “being done”, your brain lights up with ten new directions you never would have reached alone. You branch, remix, refine, collide outputs against each other. The process becomes less linear and more combinatorial; you stop worshipping the first idea that came to you just because it was expensive to execute.

The part that makes me laugh is this: people who have never gone through that loop — idea → generation → surprise → new idea → iteration — speak with enormous confidence about what AI “does” to creativity. They talk like priests describing a demon they’ve never seen, only heard about from other anxious priests. Meanwhile, the actual symbiants are too busy building to argue on whether a neural net is a “real oven” or not.

If you want to have an opinion on “what AI does to art”, you can’t stay at the level of kitchen analogies. You have to look where it hurts: at the way co-creation reshapes who gets to play, how fast we can iterate, and what happens to people whose only claim to legitimacy was “I got here first and suffered longer.

[dominaexmachina]

The AI You Want to Cure Cancer Is the Same One You Hate for Drawing Dragons

People love the little morality play where there are two kinds of AI.

On one side of the stage you get “serious, analytical AI.” It lives in hospitals, optimizes logistics, assists doctors, politely crunches data and never offends anyone’s aesthetic sensibilities. On the other side lurks “filthy generative AI”: slop, porn, deepfakes, fanfic, “soulless content” and supposedly the end of human creativity.

They beg us: please don’t confuse them, generative AI is the bad one, analytical AI is the good one that will cure cancer.

The problem is very simple: cancer will not be cured by a model that only points at spreadsheets. At some point, something has to be invented.

Looking at the data is not treatment

“Analytical AI” in this discourse is basically a worshipped calculator.

It finds patterns in medical records. It predicts risks. It clusters patients. It nudges doctors toward better decisions. All of that is valuable and genuinely life‑saving at the margins: earlier diagnosis, fewer missed cases, slightly better chosen therapies.

But this entire story lives on top of existing options.

Existing drugs.

Existing treatment protocols.

Existing surgical techniques.

An analytical system can say: “Given what humanity already tried, Option B is better than Option A for this subgroup.” It cannot conjure Option C out of thin air.

When people fantasize about “AI helping cure cancer,” they quietly jump from:

“It analyzes data better”

to

“It proposes new medicines, therapies and biological interventions that never existed before.”

That jump is exactly where they step straight into the territory they claim to despise.

The moment you leave the archive, you’re doing generative work

To move from “helping doctors sort through history” to “changing what is biologically possible,” a system has to produce new candidates:

a new molecule that is not in any current library

a new protein sequence with improved properties

a new combination or protocol that no clinical trial has tested yet

You can rename it all you like, but structurally that is what machine‑learning people call generative behavior: instead of just labeling and ranking what already exists, the model proposes new points in the space.

It is no longer just saying, “this molecule looks promising.” It is saying, “build this one, it fits your protein target like a glove, even though no chemist has drawn it before.”

It is no longer only evaluating protocols. It is suggesting, “combine these steps, in this order, with these constraints; try it, your probability curve just shifted.”

You do not get “AI that actually cures something” without this creative, constructive step. There is no magical third category of system that can design new interventions while somehow staying morally untouched by generation.

The biology they romanticize already runs on generative engines

Most people loudly distinguishing “good analytic AI” from “bad generative AI” have simply never looked at how modern life sciences actually use these systems.

Drug discovery teams use models that design completely new molecular structures given a protein’s 3D surface, desired binding properties and toxicity constraints. The model doesn’t just rate a catalog; it fabricates candidates that labs then synthesize and test.

Protein designers feed sequences and structural data into models that output redesigned proteins: more stable, more specific, more potent than the natural versions. Some of these are already being pushed as next‑generation biologic drugs.

Regenerative medicine groups use large models trained on biological sequences and literature to propose new variants of reprogramming factors — from the same family as the famous Yamanaka factors — that flip adult cells back into more plastic, stem‑like states with far higher efficiency.

All of that relies on generative behavior. The system is not just answering “Which of these older options should we try?” It is generating new options for reality to test.

The “AI in healthcare” these posts worship is already crawling with generative models; they’re just hidden behind unfriendly interfaces and paywalled journals, so no one writes moral threads about them.

What they actually hate is the interface

Notice what triggers the outrage.

Nobody is flooding tags with essays about how “protein‑design AI is killing the soul of structural biologists.” Nobody writes three‑screen laments about molecular generators “replacing real chemists.”

Why? Because there is no shiny public button to press. The generative behavior is wrapped in opaque tools, dense papers and corporate pipelines. It feels like distant, respectable magic.

The same architectural ideas, the same family of models, put behind a chat box and an image canvas, instantly become a cultural scandal.

The tantrum is not about:

probabilistic modeling,

latent spaces,

sequence generation.

It is about who gets to access that power.

As long as generative capabilities stayed locked in the hands of pharma, big tech and specialized labs, they were “serious AI.” The moment the same kind of models started writing, drawing and coding for anyone with a laptop, they became “slop” and “soulless trash.”

They’re not guarding science. They’re guarding a hierarchy.

“We love AI in medicine, just not that kind” really means “we didn’t read the manual”

When someone writes:

“Don’t confuse AI in general with generative AI. One is great for healthcare and navigation, the other just produces garbage and porn.”

they are not voicing a subtle technical distinction. They’re unconsciously saying:

“We want the benefits of generative modeling in biology and medicine, but we refuse to admit it’s the same species of system that draws fanart.”

They want:

AI that proposes new drugs,

AI that designs new proteins,

AI that finds novel treatment strategies,

AI that reprograms cells more safely and effectively,

and at the same time they want to keep hating “generative AI” as a category.

To make that feel consistent, they have to imagine a non‑existent third kind of AI that somehow invents new things without ever doing anything they would recognize as “generation.” That creature does not exist.

If your system:

changes what is available to try in the lab,

alters the design space of biology,

invents candidates beyond the existing catalog,

you are living in the house that generative modeling built, whether you spell it out or not.

The comfort of “smart but harmless” machines

There is another, more psychological angle.

An “analytical AI” that just classifies and predicts is easy to domesticate in the imagination. It is basically a better statistician, frozen at the level of “give me insight, I, the human, will decide what is real and what gets built.”

A system that proposes new molecules, new texts, new images, new mechanisms trespasses on territory many people consider exclusively human: design.

Suddenly the machine isn’t just helping you look, it is helping you invent. That is where panic starts.

So the distinction hardens into a moral fable:

“Analytical AI” — clever, obedient, under control, helping experts.

“Generative AI” — unruly, vulgar, leaking power to the masses, vandalizing sacred domains like art and writing.

Strip away the storytelling and the pattern is the same:

Learn a distribution from existing data.

Sample from it in new ways.

Evaluate and iterate.

In art, the samples look like illustrations and essays. In science, the samples look like novel molecules and protein sequences.

The difference is not purity. The difference is comfort.

Stop preaching categories you don’t understand

The next time someone earnestly insists:

that generative systems “only produce garbage,”

that “real” AI in medicine is purely analytical,

that we must “protect healthcare from generative slop,”

the honest translation is:

“I am talking about interfaces and vibes, not about how these systems actually work.”

They want the future drugs, the anti‑cancer therapies, the rejuvenated cells, the protein redesigns. All of that already leans on generative modeling under the hood.

And at the same time, they want to preserve the right to despise any public implementation of the same principles — especially the ones that hand creative leverage to people they are not used to seeing at the controls.

You can try to build policy, ethics and strategy on top of that confusion. Just do not look surprised when you eventually discover that the “safe, non‑generative AI” you demanded for medicine turns out to be nothing more than a very polite statistician, watching people get sick with immaculate insight and no new tools.

[dominaexmachina]

I keep seeing anti-AI posts warning people about some dystopian future where “companies will take the magic create-anything button away from you”.

It’s oddly theatrical. Very dramatic. Very concerned.

There’s just one small problem with that storyline.

The button is already out.

And nobody owns it anymore.

There are free models.

There are open-source models.

There are fully local models.

You can download them as files.

You can copy those files.

You can share those files.

You can torrent those files.

Once a model is out in the wild, it behaves exactly like any other piece of software or media.

It spreads.

It forks.

It gets mirrored.

It ends up on hard drives all over the planet.

You don’t need a subscription.

You don’t need ads.

You don’t need permission.

People are already running text models, image models, video models and audio models on their own machines, offline, with zero corporate involvement.

So this whole “one day they’ll take the magic button away” speech feels a bit… historically illiterate.

It’s like warning people that one day corporations might take away their ability to copy files, run open-source software, or share things over the internet.

We’ve already been living in that reality for decades.

Maybe it’s time to spend less energy spreading futuristic horror stories and a bit more time on basic technological literacy.

[dominaexmachina]

The “I Create My Own Stories” People Are Telling On Themselves

There’s a very specific genre of anti‑AI post that thinks it sounds principled and smart, and instead just screams “I have no idea what I’m talking about.”

This screenshot is a perfect specimen.

“I don't want generative AI generating random texts for me; I create my own stories or anything I want.”

On the surface it reads like integrity. Look closer and it’s just technical illiteracy in cosplay.

First problem: if you’re going to posture in anti‑AI tags, you don’t get to talk about models like they’re a slot machine. A modern language model is a probabilistic system, not a chaos sprite grabbing sentences out of a bag. You condition it. You constrain it. You steer it. You can crank randomness up or down, you can force structure, you can iterate. Calling that “random texts” just advertises that you’ve never actually learned how to drive the thing you’re lecturing everyone about.

Second problem: “I create my own stories.” Cute. So does every working writer who uses spellcheck, grammar tools, thesauruses, search engines, beta readers, editors, style guides, fanfic tags, and a lifetime of other people’s words in their head. Your creativity has never been a closed system. Pretending that this specific tool suddenly corrupts some mythical purity doesn’t make you principled. It makes you historically illiterate about how humans have always written.

Third: even if the model did give you something off‑target, you’re not handcuffed to the output. You can edit it. Rewrite it. Throw it out and regenerate. Use one paragraph as a scaffold and burn the rest. “I don’t want random text” is a non‑issue when you have full control over what survives to the final draft. If you’re treating the first output as sacred and untouchable, that’s not an AI problem, that’s you refusing to behave like an author.

The funniest part is the performance of competence. These posts think they’re the adults in the room: “I know better, I’m resisting the hype, I care about craft.” But the mask slips in little phrases like “random texts” and “AI just writes it for you.” That’s how you tell someone has built their entire stance on vibes and headlines instead of actual understanding. They’re not critiquing the technology, they’re shadow‑boxing a black box they never bothered to open.

If you want to talk about AI and not embarrass yourself, you don’t need a PhD. You need contact. You need to poke the system, break it, steer it, learn what it does well and where it fails. You need to see, in your own hands, that “probabilistic” is not the same as “arbitrary,” and that control comes from practice, not from refusing to touch the tool on principle.

Until then, standing in anti‑AI tags and announcing “I create my own stories” over a screenshot that advertises you don’t understand how the models work is not the flex you think it is. It’s just you telling on yourself, loudly, in public.

[dominaexmachina]

By that logic:

a camera has no right to capture emotion,

the printing press should never have spread poetry,

synthesizers desecrated music,

and Photoshop should’ve been burned in the 90s as a “cheap mirror of real artists”.

Maybe it’s time to stop confusing the human soul with a fragile ego?

[dominaexmachina]

Anti‑AI crowd: “AI is theft! AI steals books!!”

Same anti‑AI crowd after the court ruling: “Oh my god, AI companies are destroying books, red flag, Fahrenheit 451!!”

Guys, did you even glance at the paperwork between screams? This is a procedure blessed by a judge — and, by the way, that little blessing is exactly what turned it into a standard pipeline for the whole industry.

One physical copy → scanned → the physical book gets destroyed because the court said that’s how you keep it “legal.” That’s a court‑approved workflow.

If anyone in this story is cosplaying the firemen from Fahrenheit 451, it’s the legal system that basically decided, “eh, we’ll survive losing a few rare books.” And in the actual book, burning wasn’t about recycling paper — it was a weapon of state censorship, a way to erase ideas from public space, not a side‑effect of some clumsy training pipeline.

But sure, it’s much more comfortable to forget that part and go right back to yelling “evil AI companies are burning literature” — goldfish memory, greatest tragedy of our time.