Margaret Atwood says the problem with AI is ‘garbage in, garbage out’
TL;DR
Margaret Atwood said at the Babell Literary and Cultural Festival in Porto that she tried Claude exactly once and dropped it after a wrong answer about the British detective series Father Brown. Her point was blunt: an LLM does not know whether it is lying. It samples published material, inherits its gaps, and can turn that into an answer that sounds plausible but is false.
Nauti's Take
Atwood is not offering a technical breakdown, but she gives a useful everyday rule: the person using AI still owns the verification. Treating hallucinations as a temporary nuisance gets weaker once people copy answers into research, work, or school.
The problem is not only poor data quality. It is the temptation to mistake a confident answer for a checked one.
Briefingshow
Atwood's example is small, but it hits a larger issue: AI models cannot reliably separate knowledge from weak source material when the training or retrieval base is incomplete. For users, the lesson is practical: the more specific the question and the higher the cost of being wrong, the less a fluent answer should count as evidence.
Sources