Margaret Atwood has never been one to mince words. The author of The Handmaid's Tale and The Blind Assassin has spent decades dissecting power structures, human folly, and the quiet horrors that bloom from our own inventions. So when she sat down for an interview at the Babell Literary and Cultural Festival in Porto, Portugal, and the conversation inevitably turned to artificial intelligence, she didn't reach for the usual techno-utopian platitudes. She reached for a computing axiom that's older than most of the startups in the room: "garbage in, garbage out."
According to www.theverge.com, Atwood was blunt about what she sees as the fundamental problem with current AI systems. It's not that they're too smart. It's that they're trained on the messiest, most contradictory, and often outright toxic data set we've ever assembled: the entire internet. "If you feed it all of human production, including the worst of it, what do you expect?" she reportedly said. It's a line that stuck with me, because I've been thinking about it all week while testing the latest wave of AI writing tools.
The Problem Isn't the Machine—It's the Training Data
Here's the thing about "garbage in, garbage out." It's not just a technical complaint. It's a philosophical one. Atwood, who has written about everything from speculative dystopias to historical fiction, understands that language is never neutral. Every word carries context, bias, history. So when you scrape the entire open web—Reddit threads, poorly moderated forums, biased news articles, fan fiction, and genuine hate speech—and feed it into a statistical model, you're not creating a neutral oracle. You're creating a mirror of our worst impulses, polished to a deceptive sheen.
I've been playing with several popular AI writing assistants for the past month, and I can tell you: the results are often impressive in a technical sense and terrifying in a human one. I asked one tool to write a short story about a mother and daughter. It produced something grammatically perfect, structurally sound, and utterly hollow. The emotional beats were there, but they felt like they'd been assembled from a checklist. The characters were archetypes, not people. It was, in Atwood's framing, the literary equivalent of a fast-food burger: technically food, but missing everything that makes a meal memorable.
And that's when the model isn't actively reproducing harmful stereotypes. According to www.theverge.com, Atwood's comments came during a broader discussion about creativity and technology, but she zeroed in on the ethical dimensions. She pointed out that AI doesn't "know" anything. It predicts. It strings together words based on statistical probability, not understanding. So when you ask it to write a poem about love, it's not drawing from experience or emotion. It's drawing from every sappy greeting card, every breakup text, every overwritten novel that's ever been digitized. No wonder the results feel like a parody of human expression.
The Handmaid's Tale Was Never Really About the Future
It's worth remembering that Atwood has been warning us about the misuse of technology for decades. The Handmaid's Tale isn't a science fiction novel about a distant future. It's a novel about the present, exaggerated just enough to make us uncomfortable. The Republic of Gilead didn't need new weapons or alien invasions to become a totalitarian nightmare. It needed people who were willing to use existing tools—surveillance, propaganda, religious zealotry—to strip away rights. Atwood has always understood that the most dangerous technologies are the ones we already have, deployed by people who think they're doing good.
AI fits that pattern perfectly. We're rushing to deploy large language models in everything from customer service to healthcare, and we're not asking the hard questions. Who trained the model? On what data? What biases are baked in? Who gets to decide what constitutes "harmful" output? Atwood's "garbage in, garbage out" is a reminder that we're building systems on a foundation of human garbage, and then acting surprised when they produce garbage results.
I think about this every time I see a company announce a new "AI assistant" for therapists or teachers. The pitch is always the same: it'll save time, reduce burnout, improve outcomes. But what happens when the model gives bad advice because it was trained on a forum where people were arguing about whether therapy actually works? What happens when a student gets a an assignment back from an AI grader that penalizes them for using a dialect that differs from the model's training data? The problems aren't hypothetical. They're happening right now.
The Creativity Question: Can AI Actually Make Art?
Atwood didn't just critique the input. She questioned the output's status as art. And this is where I think she's most interesting—and most controversial. At the festival, she apparently argued that AI-generated text lacks the intentionality that defines genuine creative work. A novelist chooses every word, sometimes agonizing over a single sentence. An AI generates the most probable next token. The result might look like a poem, but it's not poetry. It's a statistical artifact.
I used to be more sympathetic to the pro-AI art crowd. I thought, maybe it's just a new tool, like photography was for painting. But the more I use these tools, the more I think Atwood is right. Photography captured light and shadow. It required composition, timing, an eye. AI text generation requires a prompt. It's not a collaboration. It's a delegation. You're outsourcing the creative act to a machine that has no stake in the outcome. No ego. No desire to be understood. No fear of failure. And without that human stakes, can it really be art?
I'll give you an example. Last week, I tried to get an AI to write a opening paragraph for a story about a man who returns to his hometown after twenty years. I wanted something that evoked the specific smell of rain on concrete and the weird tension of seeing a place that's both familiar and alien. The AI gave me: "John hadn't been back to Millbrook in two decades. The town had changed, but so had he." Technically fine. But it's the kind of writing that disappears from your memory the moment you finish reading it. It's placeholder text. It's the literary equivalent of elevator music.
What Atwood Gets Right About the Silicon Valley Mindset
There's a through line here that connects Atwood's critique of AI to her broader skepticism of technological solutionism. The people building these systems tend to believe that every problem has a technical fix. Loneliness? Build a chatbot. Writer's block? Generate text. Democracy under threat? Deploy algorithmic content moderation. But Atwood has always understood that some problems are fundamentally human, and they require human solutions.
I think about the current wave of AI hype and how it mirrors the crypto boom of a few years ago. Same breathless promises. Same evasion of hard questions. Same tendency to blame users when the technology fails. Atwood's "garbage in, garbage out" is the kind of common sense that gets lost in the frenzy. It's a reminder that you can't build a cathedral on a swamp. You have to drain the swamp first.
And let's be clear: the swamp is us. The internet is not a pristine library of human knowledge. It's a firehose of everything we've ever said, thought, or typed—including the stuff we're embarrassed about. Including the stuff that's actively harmful. Including the stuff that's just plain wrong. And we're training our most powerful tools on this mess and expecting wisdom.
The Practical Implications for Writers and Readers
So what does this mean for someone who writes for a living? For someone who reads for pleasure? It means we need to be more discerning about how we use these tools. I'm not saying all AI use is bad. I use it for brainstorming sometimes. For generating alternative headlines. For summarizing dense research. But I don't let it write the actual article. That's my job. Because I have something that the model doesn't: a perspective. A voice. A willingness to be wrong.
Atwood's interview should serve as a cautionary tale for every company rushing to replace human creativity with statistical prediction. The output will always be limited by the input. And if the input is the chaotic, beautiful, terrible mess of human expression, then the output will be a reflection of that mess. That might be useful for some things. But it's not a substitute for the real thing.
The Bottom Line (No Bullets, I Promise)
Margaret Atwood has been right about a lot of things over the years. She was right about the fragility of democratic institutions. She was right about the danger of charismatic authoritarians. And I think she's right about AI. The problem isn't that the technology is too powerful. It's that we're feeding it the worst of ourselves and expecting the best. Garbage in, garbage out. It's as simple and as devastating as that.
I'll leave you with this: the next time you're impressed by something an AI wrote, ask yourself whether it made you feel anything. Whether it changed how you see the world. Whether you'd remember it a week from now. If the answer is no, then maybe Atwood is right. Maybe we're just polishing garbage. And maybe it's time to start thinking about what we actually want to build, instead of just building what we can.
Originally reported by www.theverge.com. Rewritten with additional analysis and real-world context by Emily Hartwell.
