Margaret Atwood tried Claude and summed up AI's problem as 'garbage in, garbage out'

Margaret Atwood, author of The Handmaid's Tale and The Blind Assassin, used the AI chatbot Claude exactly once. That was enough for her to conclude that artificial intelligence has a fundamental data-quality problem.

One query, one wrong answer

At the Babell literary festival in Porto, Portugal, Atwood described asking Claude about the British detective series Father Brown. She was disappointed with the result.

"Claude gave me a wrong answer — or lied.

Then again, it didn't know it was lying, because it's not a human — it's a large language model," the writer said.

According to Atwood, the model simply skimmed the surface of the available data and produced information that sounded confident but was factually inaccurate. That, she argues, is the core weakness of today's language models: they don't know what they don't know, and they don't warn the user about it.

Garbage in, garbage out

Atwood's diagnosis is concise: "garbage in, garbage out," a principle programmers have known since the 1960s. If training data contains errors, gaps, or bias, the output will be unreliable — no matter how much compute sits behind it.

Hallucinations — when language models generate confident but wrong answers — remain one of the industry's central unsolved problems:

• All major models — GPT, Claude, Gemini — regularly get dates, names, facts, and quotes wrong
• Models are trained on internet text, which contains far more inaccurate and outdated information than it might seem
• The more confident an answer sounds, the harder it is for a non-expert to verify it
• In medicine, law, education, and journalism, these errors create real risks

To counter this, companies are connecting models to external search, building fact-checking systems, and training models to acknowledge uncertainty. But nobody has a complete solution yet.

A cultural voice against tech optimism

Atwood is not the first major author to openly criticize AI. In 2023, thousands of writers signed an open letter demanding that AI companies pay authors for using their work in training. George R.R. Martin, John Grisham, and others filed class-action lawsuits.

The literary community is broadly skeptical: writers see AI as a threat to their livelihoods and object to models being trained on their books without permission or compensation. Atwood went a step further. She didn't just stand in solidarity with her colleagues — she tested the technology herself. The result was telling: even someone skilled at critically evaluating sources got a wrong answer from AI — and spotted it immediately.

What it means

AI companies market their products as tools for working with information: search, analysis, summarization. But that is exactly the role in which they remain unreliable for now. An ordinary user might never notice a single wrong answer. For a writer used to working with primary sources, that one answer becomes a verdict on the entire technology.