Thread by @raftingzalmoxes

#PaleoAgainstAI
Strangers from beyond the paleo circles, friends of old.
We have gathered here to answer the threat of AIGC.
Paleo-related creativity of all shapes and sizes now stands on the brink of destruction. None can escape it.
WE WILL UNITE OR WE WILL FALL.
A THREAD: (1/?)

So what's the imminent threat? Let's examine a few events.
1. On 25 Feb 2026, the trailer of a new docu, 'Mystery of the Megafauna', was released. The film claims to have used 'stunning CGI'. What it really did was passing VFX models made by artist Antonio Peres through AI.
(2/?)

2. Starting from 3 May 2026, Paleorex has been posting a series of short videos called 'Tales of Prehistory'. They used a mixture of photobashing and AIGC to achieve seeming photorealism. Uninformed viewers can no longer identify the inconsistencies of AIGC with confidence. (3/?)

These videos and others like it will be endorsed by a certain very influencial paleo documaker at DinoConUK two months later...apparently because they have 'superior simulated water and mud'. This was also what triggered the formation of this very thread. (4/?)

3. Jurassic Sharks is a Discovery Shark Week docu that will air on 28 July 2026. 'Revolutionary CGI resurrects the ocean's most bizarre Jurassic sharks, bringing them to life like never before', said the press release.But what we actually got was a bunch of AIGC footage. (5/?)

As industry giants and platforms like Netflix continue to push for AI use, we are nearing a dark reality in which AIGC, like the dumbed-down narration the community has come to hate so much, will be semi-mandatory in paleo docus and beyond consultants'/producers' control. (6/?)

HOW DID IT COME TO THIS?
But before we go into that question, a bit of recap for people new to the issue:
WHAT ACTUALLY IS AIGC? AND WHY IS IT THE WRONG CHOICE? (7/?)

In 2025, I posted two threads detailing why AIGC is harmful for paleoart. For someone new to the topic, it's good to start either from here or from the Gizmodo article I'm going to post later on in this thread. (8/?)

https://x.com/raftingzalmoxes/status/1874754089109905798

Sadly, things have changed a lot since I posted those threads, and I confess that I did make some wrong assumptions, and have somewhat underestimated the severity of AIGC's onslaught. (9/?)

https://x.com/raftingzalmoxes/status/1875886074935095709

So, in short.
The AI that produces AIGC is fundamentally a MATH FUNCTION with a defined format and a ton of variables. You need to figure out the coefficients by yourself by fitting AI to a dataset (corresponding prompts and images, etc.). (10/?)
gizmodo.com/paleontologist…

When one uses AI to 'make art', several ethical issues arise.
1. The 'training data' is often sourced (usually without consent) from artists. Images/videos returned are essentially mashups of different elements extracted from artworks in the 'dataset'. this is plagiarism. (10/?)

2. AI fitted to images ISN'T ACTUALLY REFLECTING ANATOMICAL KNOWLEDGE. It's only reflecting the concensus patterns it learns from the images.
If the AI is trained on high quality data, it may imitate shapes, movements, etc. that can *look* passable even for some enjoyers. (12/?)

Meanwhile, if the dataset is of poor quality or consists predominently of only one type of data, and if the AI OVERFITS the data, features that only exist in high concentration within the dataset would be treated as a 'general feature' by the AI. (13/?)

Look at these AIGC goblin sharks. Real goblin sharks don't usually leave their protrusible jaws outstretched all the time, but because so many goblin sharks photos show outstretched jaws, AI generates sharks with outstretched jaws by default, perpetuating a misconception. (14/?)

The situation becomes even more exasperating when AIGC gets hold of tons and tons of data and becomes capable of generating plants, modern animals and other organisms with fine textures and apparent realism. It becomes very hard to discern between AIGC and real photo. (15/?)

Coupled with the rapid creation and fast spread of AI-generated content, the photo-like textures and twisted 'animals' in AIGC can easily misguide uninformed, unsuspecting viewers in ways far worse than a scientifically uninformed paleo docu. AIGC is the enemy of scicomm. (16/?)

3. AI requires BIG data. The more diverse data you have, the more likely your AI can be generalisable (in theory). AI training/data storage consume large amounts of energy and produce heat/pollution on a huge scale. Sustainable solutions might exist, but they are FAR off. (17/?)

Dr. Mark Witton once said that paleoart MUST 1) be BEHOLDEN TO SCIENTIFIC DATA, and 2) INVOLVE RESTORATIVE COMPONENT TO FIT MISSING YET ESSENTIAL BIOLOGICAL DATA. This means that true paleoart/paleo docu must be both SCIENTIFIC INFORMED and ARTISTIC. AIGC achieves NEITHER. (18/?)

I have already shown you how AI cannot reflect true scientific knowledge, and now l want you to think about the following.
AI is essemtially FITTING A STRING THROUGH A BUNCH OF DEFINED POINTS.
How to define this process?
Math, maybe.
Grunt work, perhaps.
BUT IS THIS ART? (19/?)

If our entire art landscape is dictated by and can be generated with a single type of AI, does that not imply that HUMAN ART PAST, PRESENT AND FUTURE ARE A FINITE SET WITH A DISTRIBUTION THAT CAN BE DETERMINISTICALLY DESCRIBED BY A FUNCTION?! IS THIS A CORRECT WORLD VIEW? (20/?)

ARE ARTWORKS, SOURCED FROM THE INEXHAUSTABLE DEPTHS OF HUMAN MIND AND BLOOMS WITH SPARKS OF INSPIRATION, MERELY POINTS, DATA ENTRIES, FOR AN AI TO FIT INTO AND DEFINE?
WHAT MEANING EXISTS IN A FUNCTION THAT SOMEHOW MANAGES TO FIT THROUGH ALL THESE 'POINTS'? (21/?)

Paleoartists can conjure up brand new scenes that are both 'radical' and plausible.
AI CAN'T.
It can't conceive a Brian Engh Dilophosaurus with wild ornamentation.
It can't think up a deliciously bizarre HodariNundu art in which a mammoth smites a predator with its *****. (22/?)

WITHOUT KNOWN DATA, AI PRODUCES NONSENSE.
EVEN WITH ENOUGH KNOWN DATA, THE FUNCTION CAN STILL END UP IN WEIRD PLACES.
There's no true creativity or artistic labour in AI prompting and data fitting! (23/?)

To clarify, I'm NOT against the function itself.
A function can be used in a lot of ways, many of which are ethical and beneficial.
It can be fitted to disease data and aid treatment.
It can be fitted to protein interaction data and lead to the development of new drugs. (24/?)

When it comes to film industry, NOBODY is going to criticise the MASSIVE and ARMY MANAGER algorithms used by Peter Jackson to animate tons of models when filming LotR, or bash the ink lines machine learning tool used in Spiderverse that frees animators from menial tasks. (25/?)

There's even an instance of ETHICAL AI USE in Paleontology in which Dr. Anthony Romilio used AI to classify footprints and pinpoint trackmakers. (26/?) popularmechanics.com/science/archae…

That's what AI SHOULD be used for (Though Dr. Romilio also seemingly generated AI mammoths, which I obviously HATE. At least they have been identified as AIGC by other members of this community). (27/?)
artstation.com/artwork/zP3zBm

What's the difference between these examples and AIGC?
Those types of AI fit to data that are ETHICALLY SOURCED (scientific data, mocap data, etc.) AND DO CONTAIN REAL, MEANINGFUL, QUANTIFIABLE PATTERNS,. THEY ARE DESIGNED TO LIBERATE PRODUCTIVITY FROM *REAL* GRUNT WORK. (22/?)

The AIGC we are talking about has NONE of these traits.
It uses INETHICALLY SOURCED DATA.
It fits the data in a way that looks convincing but HAS NO TRUE MEANING WHATSOEVER.
And it seeks to *REPLACE* PRODUCTIVITY AND *REMOVE* JOBS THAT ONLY HUMAN EXPERTS CAN REASONABLY DO. (29/?)

I once said that paleoart is the last fortress against AIGC, and, thanks to its diversity in terms of reconstruction and art style, paleoart remains as impregnable as ever.
So why has the situation become so dire all of a sudden? What changed in the last year-or-so? (30/?)

Firstly, the situation hasn't become dire 'all of a sudden'. By the slow accumulation of training sets and the steady refinement through continuous data output, today's AI is able to recognise patterns much better than AI in 2025. 'Twisted JP rexes' are a bygone thing. (31/?)

This new type of AI output is more deceptive because it often adheres strongly to existing art and 'enhance' the textures (but can still get small details wrong). The photo-like appearance would often lure casual people into favouring the AI image over the original art. (32/?)

Secondly, AI video-making can now output footage that feels like segments from big-budget films. Real-looking yet fundamentally nonsensical 'AI paleo docus' can spread much faster than static images and look more inviting than true scicomm videos featuring only static art. (33/?)

Thirdly, with so much AIGC online, uninformed viewers who are against AI often wrongly vent their anger on true artists, epecially artists who create pieces with hand-made photo-like textures. Artists are being discouraged, and some are even retreating to other platforms...(34/?)

Lastly and most detrimentally, THE FORTRESS IS NOW BEING TAKEN FROM THE INSIDE.
(35/?)

We've seen museums and institutions going for the seemingly cost-efficient AI instead of commissioning real paleoartists. We've seen universities trying to prove that AI can be trained to create 'realistic reconstructions'. But things are now MUCH worse than just this. (36/?)

Toy lovers that want figures to 'come alive', VFX artists who want to 'animate' static models, paleoartists who generate AI backgrounds or even photoshop AIGC into 'reconstructions', documakers wishing for realism at a lower budget...
ONE BY ONE, THEY FELL INTO DARKNESS. (37/?)

And as each figure fell from grace, they relinquish their data, their handiwork, their models, their art...to AI, and AI consumed them without question. Their offerings formed HIGH-QUALITY DATASETS, allowing AI to go once step closer to making even more convincing output. (38/?)

They believe that AI is the magic black box that builds big castles while you sit back and watch. They see the process of writing prompts as creative, the work to produce art as grunt work, and artists as carriage drivers soon to be thrown aside in an era of railways. (39/?)

And with their combined might, they are forging a new world.
A world of AI 'paleo docus'.
A world in which a good 'docu' is one with 'only AIGC background'/'only one AIGC dinosaur' and a mid 'docu' is one where 'AIGC dinosaurs still can be recognised as the actual animal'. (40/?)

A world in which CGI = AIGC.
A world in which casual paleo 'docu' enjoyers are totally disjointed from our present community and don't understand why we hate AI this much.
'It looks passable and the texture looks real, so it's okay, right?' They would say with perplexity. (41/?)

WHERE SHALL WE BE WHEN ALL THAT COMES?
Can we only mumble that 'we are nearing to the end of times' and that 'we humans are losing faith in ourselves'?
Do we compress our works to smaller sizes and poorer quality, knowing that AI would be able to 'enlarge' them anyway?
NO!
(42/?)

Not even a 'mid' paleo docu should contain AIGC. If a docu wants this community to smile upon it, it must contain ZERO AIGC. If a docu's visuals, audio and even story are supported predominanty by AIGC, then it's ROTTEN. WE MUST MAKE THAT VERY, VERY CLEAR! (43/?)

It's time to speak up if you contributed your art and your consultancy to a production, only to discover that all that effort was fed to AI.

It's time to boost our support of true artists in the community.

It's time to unite under one banner and send one clear message.

(44/?)

At the core of this great community is the constructive interaction between experts, scicommers and enjoyers. We are perhaps closer to those that can actually make a difference at executive level than any other circle.
So LET US MAKE A DIFFERENCE HERE AND NOW.
(45/?)

LET THE CONCEITED AI BELIEVERS SEE THAT ALL THEY'VE MANAGED TO DO IS ALLOWING A FUNCTION TO DEFINE AND REPLACE THE MOST CRUCIAL THINGS THAT MAKE THEM HUMAN:

EMPATHY.
RESPONSIBILITY.
CREATIVITY.

[THREAD ENDS]
(46/46)

@threadreaderapp unroll