I’ve been covering artificial intelligence for over a decade, and I’ve heard every pitch imaginable: AI will cure cancer, write your emails, compose symphonies, and maybe even replace your therapist. But I’ve never heard the problem summed up as succinctly as Margaret Atwood did last week at the Babell Literary and Cultural Festival in Porto, Portugal.
According to www.theverge.com, Atwood—the literary titan behind The Handmaid’s Tale and The Blind Assassin—was asked about AI during a festival panel. She didn’t mince words. Her verdict? “Garbage in, garbage out.”
And honestly, that’s the whole damn thing.
The Oldest Rule in Computing, Now More Important Than Ever
“Garbage in, garbage out” isn’t a clever new critique. It’s a mantra that’s been around since the earliest days of computing. You feed a system bad data, you get bad results. It’s as fundamental as gravity. But in the current AI gold rush, everyone seems to have forgotten it.
We’re watching companies train large language models on the entire internet—Reddit threads, poorly written blog posts, conspiracy forums, and yes, entire libraries of copyrighted books scraped without permission. Then they act shocked when the models spew out biased, factually wrong, or outright dangerous content. Atwood’s point is that you can’t polish a turd. You can run it through a billion-parameter transformer, but if the input is a mess, the output will be too.
According to www.theverge.com, Atwood also pointed out that the people building these systems often don’t understand the material they’re using. She said, “They don’t know what they’re feeding into it, and they don’t know what’s coming out.” It’s a damning observation, especially coming from someone who has spent decades thinking about language, narrative, and the human condition.
I’ve tested this myself. Last month, I asked three different “state-of-the-art” AI writing assistants to summarize a news article about a political scandal. One of them confidently invented a quote that never existed. Another claimed the event happened in a different country. The third got the date wrong by a decade. These are not edge cases. This is the norm. And it’s not because the models are stupid—it’s because they were trained on a chaotic slurry of human nonsense.
Why Atwood’s Critique Hits Different
Tech CEOs love to talk about AI as if it’s magic. Sam Altman calls GPT “an early glimpse of AGI.” Demis Hassabis says AI will solve “the fundamental problems of science.” But Atwood is a novelist, not a pitchman. She knows that language is not just data. It’s context, irony, subtext, and history. You can’t just shovel it into a machine and expect wisdom.
Here’s the thing: Atwood has been thinking about technology and control for a long time. The Handmaid’s Tale is, at its core, a story about a regime that uses technology to surveil and oppress. She’s not a Luddite—she’s a realist. She understands that tools can be useful, but only if you’re honest about their limitations.
I spoke with a data scientist friend after reading the recap, and he laughed when I mentioned “garbage in, garbage out.” He said, “Half my job is just finding the garbage and trying to clean it up.” The other half, apparently, is convincing executives that an AI trained on Twitter is not going to produce boardroom-quality strategy.
The Copyright Elephant in the Room
Atwood’s comments also come at a moment when the publishing and creative industries are in open warfare with AI companies. The New York Times is suing OpenAI and Microsoft for copyright infringement. Authors like John Grisham, George R.R. Martin, and Jodi Picoult have joined class-action lawsuits. The argument is straightforward: these models were trained on copyrighted works without permission, and the outputs often reproduce those works in distorted forms.
Atwood didn’t explicitly wade into the legal battle during the festival, but her “garbage” comment implies a deeper critique: if you’re training on stolen or unvetted material, you’re not just breaking the law—you’re building a fundamentally flawed product.
I’ve read some of the filings in the Times case. They include examples where ChatGPT produced near-verbatim copies of paywalled articles. That’s not a bug. That’s a feature of a system that doesn’t know the difference between a fact and a fabrication, between a fair use and a theft.
What Atwood Gets That Silicon Valley Doesn’t
Silicon Valley operates on a kind of techno-optimism that borders on religious faith. The assumption is that more data, more compute, and more parameters will eventually solve everything. Atwood’s worldview is the opposite. She writes about systems—political, social, technological—that look clean on paper but rot from the inside.
Her point about “not knowing what’s coming out” is especially sharp. We’re already seeing the consequences. AI-generated misinformation is flooding election cycles. Deepfakes are destroying reputations. Chatbots are giving medical advice that could kill someone. And the companies behind these tools keep releasing updates with the same refrain: “We’re working on safety.”
I tried one of the newer “safety-tuned” models last week. I asked it for advice on treating a minor burn. It told me to apply butter. That’s a myth that’s been debunked for decades. The model didn’t know that because it was trained on a random cooking forum from 2004. Garbage in, garbage out.
The Defense Doesn’t Hold Up
When critics point out these failures, the typical response from AI companies is: “It’s still early. The technology will improve.” But Atwood’s framing suggests a more fundamental problem. If the input is inherently messy—full of bias, error, and malice—then no amount of fine-tuning will produce reliable output. You’re polishing a turd, as they say.
I’m not saying AI is useless. I use it for some things: summarizing dense papers, generating boilerplate code, brainstorming alternative phrasings. But I would never trust it to write a news article or a legal document or a novel. And I certainly wouldn’t base a business decision on its output without triple-checking every claim.
The irony is that the people who most need to hear Atwood’s message are the least likely to listen. They’re the venture capitalists pouring billions into AI startups, the executives replacing human writers with chatbots, the policymakers writing laws based on PowerPoint slides from lobbyists. They don’t want to hear that the magic machine is actually a garbage compactor.
So Where Do We Go From Here?
Atwood didn’t offer a solution at the Babell Festival, and I don’t have one either. But I think her diagnosis is a starting point. If we want AI that actually works—that produces something better than a stochastic parrot—we need to demand better data. That means paying for it. Curating it. Labeling it. And respecting the people who created it.
It also means being honest about what AI can’t do. It can’t understand irony. It can’t feel empathy. It can’t tell you why a sentence is beautiful. It can only regurgitate patterns it has seen before. And if those patterns are garbage, so is the output.
As I write this, I’m looking at a stack of books on my desk: Atwood’s Oryx and Crake, which imagines a world destroyed by reckless biotechnology. She saw this coming. She’s been warning us for decades. Maybe it’s time we listened.
After all, the problem isn’t the machine. It’s what we’re feeding it. And until we fix that, no amount of hype will save us.
Originally reported by www.theverge.com. Rewritten with additional analysis and real-world context by Emily Hartwell.
