Is On the Origin of Species AI-generated? 2 of 7 AI detectors said yes
Darwin's introduction to On the Origin of Species dates from 1859, and only 2 of 7 free AI detectors flagged it as AI-generated. TMR gave it a 98.2% AI probability, and Fakespot flagged it at 50.4%, barely over its 50% threshold. Desklib, Fast-DetectGPT, MAGE, E5-small LoRA and the OpenAI RoBERTa detector all passed it.
Tested
What each AI detector said
| Detector | Verdict | AI probability |
|---|---|---|
| Fast-DetectGPT (Qwen2.5-1.5B) | Human | raw score only |
| Desklib AI text detector | Human | 1.2% |
| Fakespot AI text detector | AI-generated | 50.4% |
| MAGE | Human | raw score only |
| E5-small LoRA detector | Human | 28.6% |
| OpenAI RoBERTa detector | Human | 0.2% |
| TMR (Target Mining RoBERTa) | AI-generated | 98.2% |
The passage we tested
When on board H.M.S. ‘Beagle,’ as naturalist, I was much struck with certain facts in the distribution of the inhabitants of South America, and in the geological relations of the present to the past inhabitants of that continent. These facts seemed to me to throw some light on the origin of species—that mystery of mysteries, as it has been called...
A verdict decided by a fraction of a percent
Two detectors flagged the start of Darwin’s introduction. TMR was confident at 98.2%, but TMR flagged 83.3% of the human texts in our main test, and 28 of our 30 famous texts, so its verdict on Darwin carries almost no information.
Fakespot’s flag was a near tie. It gave the passage a 50.4% AI probability, and its cut-off is 50%. The verdict is still “AI-generated”, the same label Fakespot gave Machiavelli’s The Prince at 99.8%, but the score itself says Fakespot could not tell. Fakespot flagged 34.9% of human texts in our main test and 13 of the 30 famous ones.
Everything else called it human. Desklib gave it 1.2%, E5-small LoRA 28.6% and the OpenAI RoBERTa detector 0.2%, and MAGE and Fast-DetectGPT passed it too.
Personal details and hedged claims
The passage is about as personal as scientific writing gets. Darwin recalls his time as naturalist on the Beagle, the notes and sketch he worked on for years, his poor health, and the memoir from Alfred Russel Wallace that pushed him to publish. Those concrete, dated details are what you would expect a classifier to read as human, and most of the classifiers did.
What may have nudged Fakespot is the register. Darwin hedges again and again: facts “seemed to me to throw some light”, something “might perhaps be made out”, the work “must necessarily be imperfect”. Careful qualification and formal, Latinate vocabulary are traits people often associate with AI text, and a classifier trained on modern model output could plausibly have learned the same association. That is a likely explanation, and the score suggests the effect was small.
Treat scores near the threshold as noise
A yes or no verdict hides how close the call was. Before trusting a flag, look for the probability behind it: Fakespot’s 50.4% and TMR’s 98.2% on this passage are very different statements, even though both come out as “AI”. If a tool only shows the label, assume some of its flags are this marginal. The Fakespot results show how often it flags human writing by genre, and The Wealth of Nations is another Fakespot flag that came in just over the line. All 30 results are on the famous texts page.
Questions people ask
Why was Darwin flagged as AI?
TMR flags almost every text with several paragraphs, so its verdict says little. Fakespot's flag came at 50.4%, a near tie. Darwin's careful, hedged scientific style, with phrases like "seemed to me" and "might perhaps", is a likely reason it scored as high as it did.
What does a score of 50.4% mean?
It means Fakespot was almost exactly undecided. Its threshold is 50%, so the passage counted as AI by a sliver. A slightly different excerpt might well have landed on the other side.