Dan Houser on AI in creative work: "It's being steered by people who aren't the most humane or creative"
Grand Theft Auto co-creator Dan Houser thinks a lot of AI hype is coming from the wrong people. Speaking on Virgin Radio UK, he questioned whether the loudest voices pushing AI into creative fields actually understand creativity or humanity in the first place.
"Some of these people trying to define the future of humanity, creativity or whatever it is using AI are not the most humane or creative people," he said. He framed it as a power grab dressed up as progress: a claim that machines do "being human" better than humans-an idea he called obviously false.
He also echoed recent pushback against well-known tech boosters. Figures like Mark Zuckerberg, Sam Altman, and Elon Musk have faced criticism from artists and writers, including Joyce Carol Oates, for being out of touch with culture.
The model collapse problem
Houser warned that today's AI has hard limits, especially as models begin to train on their own outputs. He likened it to feeding cows to cows-eventually you get a sickness in the system. His take: AI will "eat itself."
This issue-often called model collapse or the "Habsburg AI" idea-shows up when the web fills with AI-made content and models learn from that synthetic data. Over time, quality degrades, errors compound, and the outputs become bland or broken. If you want the research view, this paper is a solid overview: The Curse of Recursion: Training on Generated Data.
What this means for working creatives
Used well, AI can speed up grunt work and help with options. Used blindly, it pulls your taste toward the average and hands your edge to the crowd. The point isn't to avoid AI-it's to keep your taste, judgment, and process intact.
- Keep a human reference library: books, films, galleries, interviews, your field notes. Train your taste on real culture, not just prompts.
- Protect your inputs: collect and tag your own sketches, drafts, mood boards, and client feedback. Build a private corpus to reference.
- Prompt for divergence: set weird constraints, ban clichés, force "show, don't tell," and request multiple contradictory directions.
- Mix mediums: start analog when it matters (thumbnail, outline), then use AI for variants and cleanup, then finalize by hand.
- Audit voice: compare AI drafts to your last three strong pieces. If it reads generic, throw it out or rewrite from scratch.
- Timebox the tool: let AI help with the last 20% (polish, alt lines, alt comps), not the first 80% (concept and taste).
- Mind rights and credit: know what you can use, what you can't, and where attribution is required.
Houser's line in the sand
Houser stressed that AI already does some tasks brilliantly-but not everything. He's previously called the output "generic" and said the tech is "not as useful as some companies would have you believe." The takeaway: great art still needs a point of view.
For creatives, treat AI like a fast intern with an infinite memory-useful, never the author. Your job is direction, taste, and judgment. That's the moat.
If you're sharpening your workflow
Want structured ways to use AI without losing your voice? See role-based learning paths here: Complete AI Training - Courses by Job. If prompts are your bottleneck, you can also explore practical prompt patterns: Prompt Engineering resources.
Houser's debut novel, A Better Paradise, also leans into this tension-an AI gone rogue and the cost of outsourcing human judgment. The warning is clear: keep the human at the center, or the tool will set the taste for you.
Your membership also unlocks:
