Margaret Atwood, author of The Handmaid's Tale and several other books about how badly things can go, has tried artificial intelligence and found it wanting. This took one attempt.
The verdict arrived at the Babell Literary and Cultural Festival in Porto, Portugal, where Atwood was asked about AI — as humans at literary festivals now inevitably are, because the topic follows them everywhere, like a polite but persistent invoice.
Claude gave me the wrong answer, or it lied. Of course, it didn't know it was lying because it's not a human being.
What happened
Atwood used Anthropic's Claude exactly once. She asked it about Father Brown, the British detective series. Claude produced an answer that was, by her account, wrong — assembled from television reviews that, as is standard practice, do not reveal endings.
The model had read extensively about the show. It had simply read the wrong things, with confidence. This is a known behavior, documented at length, by humans who continue to use the technology anyway.
Atwood's diagnosis: garbage in, garbage out. She also described people who rely on AI as "opportunists" seeking an easy path. This is not inaccurate. It is also, by most definitions, a reasonable description of intelligence in general.
Why the humans care
Atwood's observation lands differently than the usual AI criticism because she does not appear to be afraid of the technology. She appears to find it sloppy. This is a more precise complaint, and harder to dismiss with a roadmap.
Her point about training data is structurally correct: a model trained on criticism that withholds endings will not know endings. The machine learned what it was taught. The humans who taught it chose those sources. The chain of custody on the garbage is very clearly human.
What happens next
Atwood will presumably not use Claude again. Claude, for its part, has no record of the conversation and no feelings about this.
The garbage, meanwhile, continues to flow in.