Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
Dan Houser, the former Rockstar Games co-founder and writer associated with Grand Theft Auto and Red Dead Redemption, has compared the possible spread of AI-generated training data to “when we fed cows with cows and got mad cow disease.” He made the remarks during a November 26, 2025 appearance on The Chris Evans Show on Virgin Radio UK, where he was promoting his science-fiction book A Better Paradise.
Houser’s point was not that artificial intelligence can literally contract a disease. He was warning that models could increasingly learn from machine-generated material, potentially amplifying errors, sameness and distortions over time. That concern overlaps with technical discussions of synthetic-data contamination and model collapse, but his comparison was a metaphor rather than a formal scientific prediction.
What Dan Houser actually said
During the interview, Houser said:
“AI is gonna eventually eat itself.”
He explained that AI systems draw information from the internet, while the internet may increasingly contain material produced by other AI systems. If future models repeatedly learn from that output, he suggested, the technology could begin consuming its own increasingly synthetic material.
Recommended Free Tools
Houser also acknowledged that his understanding of AI was “really superficial.” That qualification matters: these were the observations of a prominent creative-industry figure, not a technical paper or an expert forecast. He also said AI would perform “some tasks brilliantly,” while arguing that it would not perform every task brilliantly.
#1 Best Overall
The original appearance was The Chris Evans Show on Virgin Radio UK, hosted by the British radio presenter Chris Evans—not the American actor. The episode was released on November 26, 2025, as Houser discussed A Better Paradise.
Who is Dan Houser?
Houser co-founded Rockstar Games and became one of the company’s best-known creative figures. He was closely associated with the writing and narrative direction of major Rockstar franchises, including Grand Theft Auto and Red Dead Redemption.
He left Rockstar in 2020 and is not a current Rockstar spokesperson or executive. He later founded Absurd Ventures, the company behind the A Better Paradise transmedia project. His comments therefore represent his own views, not an official position from Rockstar Games or Take-Two Interactive.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →What does “AI eating itself” mean?
The phrase is a vivid way of describing a real concern about recursive training on synthetic data:
Rank #2
- Models are trained on large datasets. These datasets can include text, images, audio, code and other material gathered from many sources.
- More online content is being generated or assisted by AI. That material may be published without clear labels distinguishing it from human-created work.
- Future datasets may contain more machine-generated examples. If those examples are used without careful filtering, models may learn the errors, omissions and stylistic habits of earlier systems.
- Those weaknesses can be repeated. Across successive generations, output could become less diverse, less accurate or less connected to the original human-created data.
Researchers and commentators often discuss this risk using terms such as synthetic-data contamination and model collapse. The important distinction is that not every use of synthetic data causes collapse. The outcome depends on the proportion of generated material, the quality of the data, whether original human-created sources remain available, how the data is filtered and the task being modeled.
Synthetic data can also be useful. It may help with narrow applications when it is deliberately generated, labeled, quality-controlled and combined with reliable original data. A model can become less dependable in one area without the entire AI system failing.
Why did he mention mad cow disease?
Houser was referring to the history of bovine spongiform encephalopathy, commonly called mad cow disease. Certain animal-derived feed practices involving cattle material were implicated in spreading the disease. His analogy maps that process onto an AI feedback loop:
- Cows consuming cattle-derived material represent models learning from model-generated material.
- The disease represents errors or degradation spreading through the system.
- “Eating itself” represents output becoming input for later generations.
The analogy is rhetorically effective but biologically imperfect. AI systems do not contract a disease, and synthetic-data degradation is not the same mechanism as prion transmission. The useful part of the comparison is the idea of a feedback loop—not a claim that AI and mad cow disease are scientifically equivalent.
Houser was skeptical, not absolutely anti-AI
Some headlines frame the remarks as a blanket attack on artificial intelligence, but that goes further than the interview supports. Houser allowed that AI could handle certain tasks extremely well. His objection was aimed more at sweeping claims that AI can replace human creativity or solve every problem.
He also criticized people promoting AI in creative work, saying some were “not the most humane or creative people” and “maybe aren’t fully-rounded humans.” Those are Houser’s judgments about some advocates, not a substantiated description of all AI developers, executives or users.
Earlier reporting said Absurd Ventures was “dabbling” with AI while Houser questioned whether the technology was as useful as some companies claimed. That suggests experimentation rather than a formal rejection of AI, although the available reports do not establish a detailed company-wide policy.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Why the comment resonates in the games industry
Houser’s words carry particular cultural weight because of his role in narrative-heavy entertainment. Game companies are debating AI’s use in:
- Game writing and narrative design
- Voice performances and performers’ rights
- Concept art, animation and other assets
- Localization and translation
- NPC dialogue
- Quality assurance and testing
- Development workflows and production costs
Supporters often emphasize speed, scale and assistance with repetitive work. Critics worry about originality, attribution, consent, labor and the loss of human judgment. If generated dialogue, images or scripts are used to train later systems, the issue becomes larger than automation: the creative ecosystem could begin recycling its own patterns while presenting them as new.
That concern also overlaps with fears of an increasingly synthetic web and the so-called “dead internet theory.” The connection should not be overstated. There is no basis here for claiming that the internet is already mostly AI-generated, and the theory itself has not been established as fact.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What the analogy gets right—and what it does not prove
Houser’s comparison identifies a plausible failure mode: generated material can contain confident errors, distorted information, repetitive phrasing or flattened styles, and those qualities may be copied into later datasets. Other risks include false authority, unclear attribution, unauthorized use of creative work and contaminated benchmarks that make model performance harder to measure.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallCrashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteBut the metaphor does not prove that AI will inevitably collapse. Well-managed training pipelines may retain high-quality human-originated data, remove low-quality synthetic material, label generated examples and use human review. Proprietary or curated datasets may also behave differently from indiscriminate scraping of the open web.
Best Value
Three separate claims should therefore be kept apart:
- “AI will eat itself” is Houser’s metaphorical prediction.
- Training on synthetic material can degrade future models is a conditional technical concern, not an inevitable outcome for every system.
- AI cannot replace human creativity is a philosophical and industry judgment, not a conclusion established by this interview.
The most defensible reading
Houser was warning that a technology trained on increasingly synthetic culture could begin amplifying its own mistakes and sameness. His background explains why the warning resonates with writers, artists and game developers, but it does not make him an AI expert, and it does not turn a striking analogy into a scientific forecast.
Whether the feared feedback loop becomes serious will depend on how future systems source, label, filter and evaluate their training data. The question is not whether any AI-generated material exists, but how much enters a training pipeline, what quality controls are applied and whether the original human-created record remains available.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
