Since the mid-twentieth century, molecular biology has described DNA through metaphors of language and information: code, message, transcription, translation, program. The historian Lily Kay reconstructed the origin of that vocabulary in Who Wrote the Book of Life? A History of the Genetic Code (Stanford University Press, 2000). She showed that it was not born in biology but in postwar cybernetics and information theory, and that it shaped the way we think about the genome long before there was evidence to support it.

In an earlier article, “The Limits of the Code Metaphor”, I argued that describing DNA as code seems to explain more than it actually explains: the evidence on non-coding DNA forces us to consider that much of the genome is accumulated evolutionary history, not a message written with a purpose. Something similar happens with the label “an AI wrote it”, only in the opposite direction: it seems to refute more than it actually refutes. In both cases, the word takes the place of the analysis.

The question that organizes this text is simple: can an idea be dismissed just because an artificial intelligence was involved in producing it? The quick answer is no. But the quick answer hides a distinction worth drawing carefully, because the provenance of an idea is not irrelevant; what it cannot do is replace the examination.

A mistake with a name

Logic has a name for this move. Morris Cohen and Ernest Nagel called it the genetic fallacy in An Introduction to Logic and Scientific Method (1934): judging a claim by its origin rather than by its grounds. A claim does not become false because an algorithm formulated it, just as it does not become true because a prestigious person utters it.

Behind the fallacy lies an older distinction. Hans Reichenbach, in Experience and Prediction (1938), separated the context of discovery, how someone arrives at an idea, from the context of justification, why that idea deserves to be accepted. The first may include intuitions, dreams, conversations or, now, a query to a language model. The second demands reasons that anyone can check. Dismissing an idea because of its origin confuses the two planes: it treats a question of justification with an answer about discovery.

This does not make provenance irrelevant. Knowing how a text was produced helps calibrate initial trust, anticipate which errors are likely and decide what to check. An AI can invent data, repeat biases and present errors with confidence. That is why sources should be reviewed, quotations checked, reasoning examined and counterexamples sought. If the idea does not survive that examination, it must be rejected for its flaws. What is not acceptable is to reject it before examining it.

The norm science had already given itself

The scientific community formulated more than eighty years ago the principle that today seems forgotten. Robert K. Merton, in the 1942 essay now known as “The Normative Structure of Science”, called universalism the first of the norms of science: the acceptance or rejection of a claim should not depend on the personal or social attributes of the person who puts it forward, whether race, nationality, religion or class (the paraphrase is mine). Claims are subjected to impersonal criteria, established in advance.

Merton was thinking of people, not machines. Extending his norm to a text produced with a tool is my own inference, and I mark it as such. But it seems to me the natural extension: if science decided that the social origin of whoever makes a claim does not decide the truth of what is claimed, I see no reason why its technical origin should.

Merton was also writing with an example in view. In Germany in the 1930s, Philipp Lenard and Johannes Stark, both Nobel laureates in physics, promoted a “German physics” that rejected relativity as “Jewish physics”. They did not dispute its equations or its predictions: they discarded it because of the origin of the person who had formulated it. Relativity remained true, and German physics paid for the prejudice with years of delay.

The Soviet Union offers the symmetrical case. Under Trofim Lysenko, genetics was declared a “bourgeois science” and condemned for its ideological provenance, not for the evidence. Nikolai Vavilov, one of the greatest geneticists of his time, was arrested in 1940 and died in a Saratov prison in 1943. There, dismissing an idea because of its origin ended up killing a person.

Up to this point, the argument could be read as an exaggeration: comparing the rejection of an AI-assisted text with the persecution of scientists. That is not the comparison. The gravity of the cases is incomparable. What they share is the mechanism, judging by origin rather than by reasons, and it is worth recognizing it while it still operates in its mildest form.

The words that close discussions

I confess this frustrates me. It reminds me of something I learned reading my grandmother’s books: ideas should not be judged by who utters them, but by what they say and by the reasons that support them.

It also makes me think of Nineteen Eighty-Four (1949). In George Orwell’s novel, Syme, the philologist working on the Newspeak dictionary, puts it bluntly: the whole aim of Newspeak is to narrow the range of thought. If the available words shrink, so does what can be thought. I am not saying that dismissing a text made with AI is the same as the censorship in the novel. I am saying there is something disturbing when a label is enough to close a discussion before it begins. “An AI wrote it” works, in those cases, like a Newspeak word: it does not describe the text, it shuts it down.

My suspicion is that much of the rejection does not come from an assessment of risks but from a rule of belonging: which texts deserve to enter the conversation and which do not. It is a hypothesis, not a result; but it would explain why the rejection usually arrives before any reading.

What remains

The way out is neither to trust the tool blindly nor to defend every use made of it. It is to do two things at once: state transparently how a text was produced, and subject what it claims to the same examination as any other. Transparency makes evaluation possible; it does not replace it.

If the idea is false, let us argue against it. If it is poorly argued, let us show where. But dismissing it only because an AI was involved is not criticism: it is giving up on examining it.

Sources

• Cohen, Morris R. and Nagel, Ernest. An Introduction to Logic and Scientific Method. New York: Harcourt, Brace and Company, 1934.
• Dagorret, Carlos. “Los límites de la metáfora del código: ADN basura y la disputa sobre ENCODE” (in Spanish).
• Kay, Lily E. Who Wrote the Book of Life? A History of the Genetic Code. Stanford: Stanford University Press, 2000.
• Merton, Robert K. “Science and Technology in a Democratic Order”. Journal of Legal and Political Sociology 1 (1942), 115–126. Reprinted as “The Normative Structure of Science” in The Sociology of Science. Chicago: University of Chicago Press, 1973.
• Orwell, George. Nineteen Eighty-Four. London: Secker & Warburg, 1949.
• Reichenbach, Hans. Experience and Prediction. Chicago: University of Chicago Press, 1938.
• On “German physics”: Beyerchen, Alan D. Scientists under Hitler: Politics and the Physics Community in the Third Reich. New Haven: Yale University Press, 1977.
• On Lysenko and Vavilov: Pringle, Peter. The Murder of Nikolai Vavilov. New York: Simon & Schuster, 2008.