Margaret Atwood is not here for your AI hype. The 86-year-old author of The Handmaid's Tale and The Blind Assassin was in Porto, Portugal, last week for the Babell Literary and Cultural Festival, and as it usually does at these things, the conversation turned to artificial intelligence. Atwood, as she tends to do, cut through the nonsense with a scalpel.
According to www.theverge.com, Atwood said the problem with AI is 'garbage in, garbage out.' She didn't mince words. She didn't offer a nuanced, academic take that tries to balance the promises and perils. She just stated the obvious, which apparently needs stating: if you feed AI a diet of bad writing, bad data, bad thinking, you're going to get bad output. End of story.
I've been covering AI for over a decade now, and I've sat through more breathless keynote speeches than I care to count. Every demo shows some chatbot generating a poem about a cat in the style of Shakespeare. Every panel has someone saying 'this will democratize creativity.' And every time, I think about what Atwood said. Because here's the thing: the internet is already full of garbage. It's full of AI-generated blog posts that read like they were written by a spreadsheet, AI-generated art that looks like a fever dream from 2015, and AI-generated music that sounds like elevator muzak from hell. And all of that garbage is being scraped, fed back into the models, and regurgitated as 'new' content.
The Garbage Loop We're Building
Atwood's point is not just a throwaway line. It's a fundamental critique of the entire AI paradigm as it's currently practiced. Most large language models—GPT-4, Claude, Gemini, Llama—are trained on vast swaths of the internet. And what's on the internet? A lot of it is self-published fan fiction, SEO-optimized listicles, and, increasingly, AI-generated content itself. We're building models that learn from a corpus that's already contaminated. It's a feedback loop of mediocrity.
I tried this last week. I asked ChatGPT to write a short story about a detective in a rainy city. It gave me something that was technically correct—proper grammar, no spelling errors, a beginning, middle, and end. But it was dead. It had no voice, no soul, no sense of place. It was like reading a summary of a noir novel written by someone who'd only ever seen screenshots of Blade Runner. The language was flat. The metaphors were borrowed. The detective 'stood in the rain, thinking about his past.' That's not writing. That's pattern matching from a million other detective stories that were themselves mediocre.
Atwood knows this because she's spent decades honing her craft. She knows that great writing is not about following rules. It's about violating them in interesting ways. It's about finding the exact right word, the unexpected turn of phrase, the detail that makes a scene come alive. AI can't do that because it doesn't know what it's saying. It's a probabilistic parrot, and parrots don't write novels.
The 'Democratization' Myth
Every time a new AI tool launches, the marketing copy is the same: 'This will democratize creativity.' The idea is that anyone can now write a novel, compose a symphony, or paint a masterpiece. All you need is a prompt. And honestly? That's insulting. It's insulting to the people who spend years learning their craft. It's insulting to the editors, the mentors, the teachers who help writers grow. And it's insulting to readers, who deserve better than algorithmic sludge.
According to www.theverge.com, Atwood's comments came during a broader discussion about the state of literature and technology. She didn't single out any specific company, but she didn't have to. The subtext is clear: the tech industry is selling a fantasy. They're selling the idea that creativity is a commodity that can be automated. But creativity is not a commodity. It's a human process that involves struggle, failure, revision, and insight. It's messy. It's unpredictable. And it's precisely the messiness that makes it valuable.
I've been a journalist for 15 years, and I've written thousands of articles. Some of them were bad. Some of them were good. The good ones came from late nights, from staring at a blank screen, from rewriting the same paragraph six times until it clicked. No AI tool can replicate that. No AI tool can feel the frustration of a sentence that won't cooperate, or the joy of finding the perfect ending. And if we pretend otherwise, we're devaluing the very thing that makes human expression worth reading.
What Atwood Gets Right That Silicon Valley Gets Wrong
Silicon Valley operates on a logic of scale. More data. More compute. More output. The assumption is that if you just throw enough processing power at the problem, you'll eventually get intelligence. But Atwood's critique suggests that this is fundamentally misguided. The problem isn't that we need better algorithms. The problem is that the input is garbage, and garbage cannot be polished into gold.
Think about the training data for these models. It includes Reddit threads, Twitter arguments, poorly edited Wikipedia pages, and, increasingly, AI-generated content from other models. It's a hall of mirrors. The models are learning from their own hallucinations. And the result is a sort of statistical mean of all the mediocre writing on the internet. It's not bad in the way that a typo is bad. It's bad in the way that a hotel lobby painting is bad. Technically competent. Completely forgettable.
Atwood's point is also a political one. Who decides what goes into the training data? Who decides what's worth preserving? The tech companies, that's who. And they're making those decisions based on what's available, not what's valuable. They're scraping everything because they can. But more importantly, they're scraping everything because they don't have a better way to judge quality. They don't know what good writing looks like. They only know what's popular.
The Real Danger Isn't Skynet
Here's where I might diverge from some of my colleagues. A lot of the hand-wringing about AI focuses on existential risk—the idea that superintelligent machines will take over the world. And sure, that's a fun sci-fi premise. But Atwood, who literally wrote the book on dystopia, seems more worried about the slow erosion of quality. The real danger isn't that AI will become too smart. It's that we'll lower our standards to match its mediocrity.
I already see this happening. Newsrooms are using AI to write first drafts of earnings reports and sports recaps. The drafts are fine. They're accurate. But they're also soulless. And as readers, we're starting to accept that. We're starting to expect that online content will be bland, formulaic, and predictable. We're training ourselves to consume AI-generated mediocrity without complaint.
Atwood's 'garbage in, garbage out' is a warning. It's a warning that if we let AI define our creative standards, we'll end up with a culture of garbage. And we'll have no one to blame but ourselves.
What Can We Do?
I don't want to end this article on a purely pessimistic note. There are things we can do. We can demand transparency from AI companies about their training data. We can support human creators—writers, artists, musicians—by paying for their work. We can teach critical thinking about AI outputs, so that people recognize the difference between a generated text and a crafted one.
And we can listen to people like Margaret Atwood. She's been writing for over 50 years. She's seen trends come and go. She's seen technology transform publishing. And she's not afraid to say that some things are worth preserving. Great writing is one of them.
I walked away from that interview recap with a single thought: Atwood is right. The AI industry is built on a foundation of garbage. And until we address that, all the hype in the world won't change the fact that we're building a machine that produces polished mediocrity at scale.
So the next time someone tells you that AI is going to revolutionize creativity, ask them: what's it learning from? And if the answer is 'the internet,' maybe we should all be a little more worried.
Originally reported by www.theverge.com. Rewritten with additional analysis and real-world context by Emily Hartwell.
