Is Common Sense by Thomas Paine AI-generated? 1 of 7 AI detectors said yes
No. Only 1 of 7 free AI detectors flagged the introduction to Common Sense, Thomas Paine's 1776 pamphlet, as AI-generated, and that was TMR, which flagged 83.3% of human texts in our main test. Desklib gave the passage a 0.1% AI probability and Fakespot 3.7%, and Fast-DetectGPT and MAGE passed it too.
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 | 0.1% |
| Fakespot AI text detector | Human | 3.7% |
| MAGE | Human | raw score only |
| E5-small LoRA detector | Human | 27.0% |
| OpenAI RoBERTa detector | Human | 0.1% |
| TMR (Target Mining RoBERTa) | AI-generated | 98.1% |
The passage we tested
Perhaps the sentiments contained in the following pages, are not yet sufficiently fashionable to procure them general favor; a long habit of not thinking a thing wrong, gives it a superficial appearance of being right, and raises at first a formidable outcry in defense of custom. But the tumult soon subsides. Time makes more converts than reason. As a long...
One flag, from the detector that flags almost everything
TMR was the only detector to call Paine’s introduction AI-generated, with a 98.1% AI probability. That verdict is close to meaningless on its own. TMR flagged 83.3% of the human texts in our main test and 28 of our 30 famous texts, and our paragraph break test showed that line breaks drive most of its AI verdicts. The passage we tested runs over several short paragraphs and ends with a signature and a postscript, which is the kind of layout TMR reacts to.
Every other detector passed it. Desklib and the OpenAI RoBERTa detector both put it at 0.1%, Fakespot at 3.7% and E5-small LoRA at 27.0%. MAGE and Fast-DetectGPT, whose threshold we set on a calibration split, also called it human.
Common Sense and the Declaration, both from 1776
Common Sense came out in 1776, the same year as the Declaration of Independence, and both argue for American independence. The Declaration drew four flags, including 91.9% from Desklib. Common Sense drew one.
The likeliest explanation is how each text is used today. The Declaration’s opening lines are quoted constantly, so language models have probably seen them far more often than Paine’s introduction. Paine’s prose is probably harder to predict as well. He capitalizes nouns at will (“the Means”, “Fire and Sword”, “Lovers of Mankind”), uses “hath” even with a plural subject, and runs one long sentence around an aside in parentheses. Elsewhere he drops in short, plain lines: “But the tumult soon subsides. Time makes more converts than reason.” Those swings between long and short sentences are the kind of variation people say AI writing lacks.
Some features could still make a modern reader suspicious: the text is persuasive, speaks of “the cause of all mankind” and leans on formal, abstract vocabulary. None of that pushed the other detectors past their thresholds.
Reading a lone TMR flag
A single flag means little until you know which detector raised it. Here the flag came from a detector that flagged most of the human writing we gave it, and Desklib, the most careful of the detectors that catch a real share of AI text, put it at 0.1%. Check the TMR results before reading anything into a TMR verdict, and see the famous texts page for how the rest of the texts fared.
Questions people ask
Does any AI detector think Common Sense is AI?
Only TMR, at a 98.1% AI probability. TMR flagged 28 of the 30 famous texts we tested and reacts mainly to paragraph breaks, so its verdict says very little about this text.
Why was the Declaration of Independence flagged more than Common Sense?
We can only guess. The Declaration is quoted far more often, so language models probably know its wording better, and its sentences are more evenly balanced. Paine's text has irregular punctuation, odd capitals and a signature line, which probably make it less predictable.