AI Research Monoculture Threatens Scientific Independence
Why Strangers Matter in Science
Science has a quiet rule that most people never hear stated aloud: a finding becomes trustworthy when strangers, working separately, arrive at the same place. The stranger part matters. If the same person checks their own work, you have diligence. If a different person, using different tools, on different data, reaches the same conclusion, you have something closer to knowledge.
Lisa Messeri and Molly Crockett argue in Nature that artificial intelligence can quietly break that rule. Their warning, published in the journal in 2026, is not that AI produces wrong answers. It is that AI produces the same answers. When many researchers route their questions through the same models, the methods may converge, the questions may converge, and the errors may converge too.
The confidence boost that comes from independent replication depends entirely on the independence. Remove it, and you keep the ritual while losing the substance. Two laboratories agreeing means nothing if both asked the same machine the same question.
The First Users Are Already in the Room
Messeri and
Crockett are themselves researchers publishing in the scientific literature, and their warning appears in the same venue where the tools they describe are being adopted Their argument does not require anyone to behave badly. A researcher who adopts a popular AI tool is making a reasonable choice. The tool is fast, widely used, and cited by colleagues.
A monoculture is not a field with one bad crop. It is a field where every plant shares the same vulnerability, so a single stressor can take the whole harvest.
The risk they name is structural: methods and questions converging, with error as the common source rather than the exception.
