Despite an impressive vocabulary, artificial intelligence often fails to accurately analyze scientific statements. A new study reveals that popular text generators frequently contradict themselves when asked the exact same question multiple times.