AI detector
In plain wordsSoftware that scores how likely it is that an AI, not a person, produced some text or media; for writing, its verdicts are wrong often enough that many schools no longer trust them.
A classifier or statistical test that estimates whether content was produced by a generative model. Text detectors either train a classifier on human and machine samples or use signals from a language model, such as how predictable each word is (perplexityperplexity, n.: A score for how surprised a language model is by real text; lower is better. A perplexity of 10 means guessing each next word is, on average, as hard for the model as guessing a ten-sided die's roll.), how much predictability varies across sentences ('burstiness'), or how a passage's probability changes under small rewrites, as in DetectGPT. Accuracy degrades on short texts, newer models, edited or paraphrased output and non-English text, and studies have found high false-positive rates on writing by non-native English speakers. Because a false accusation of cheating is costly, many universities have disabled or discouraged text detectors. Provenance methods, such as statistical watermarksAI watermarking, n.: Hiding an invisible signal in AI-generated text, images, audio or video so that software can later tell the content was made by AI — like a banknote's watermark, but undetectable to the eye. embedded at generation time or signed content credentials, are generally considered more reliable than after-the-fact detection, which is easily evaded.
The detector rated the essay 98% likely AI, but the student's drafts and version history showed otherwise.
Origin
Statistical detection of machine-written text predates chatbots (for example the GLTR tool, 2019); detectors multiplied after ChatGPT's release, notably GPTZero (January 2023) and Turnitin's detector (April 2023). OpenAI released its own AI Text Classifier in January 2023 and withdrew it that July, citing its low rate of accuracy.
Usage note
Image and audio detectors face a similar arms race; see deepfake detectiondeepfake detection, n.: Using software, often itself AI, to tell whether a video, photo or voice recording has been faked or generated by AI rather than captured from reality.. Detector scores are probabilities, not proof, and should not be the sole basis for disciplinary decisions.
Related
- AI watermarkingAI watermarking, n.: Hiding an invisible signal in AI-generated text, images, audio or video so that software can later tell the content was made by AI — like a banknote's watermark, but undetectable to the eye.
- deepfake detectiondeepfake detection, n.: Using software, often itself AI, to tell whether a video, photo or voice recording has been faked or generated by AI rather than captured from reality.
- perplexityperplexity, n.: A score for how surprised a language model is by real text; lower is better. A perplexity of 10 means guessing each next word is, on average, as hard for the model as guessing a ten-sided die's roll.
- synthetic mediasynthetic media, n.: Images, video, audio or text created or substantially altered by AI rather than recorded or written directly by people, whether for art, entertainment, business or deception.
- SynthIDSynthID, prop. n.: Google's system for hiding an invisible watermark in pictures, video, audio and text made by its AI tools, so they can later be recognized as AI-generated.
- LLM-ismsLLM-isms, n.: Words, phrases and habits of style that AI chatbots overuse — certain fancy verbs, tidy lists of three, 'it's not just X, it's Y' — which make writing recognizable as machine-made.
Sources
- Liang, W., Yuksekgonul, M., Mao, Y., Wu, E. & Zou, J. (2023). GPT detectors are biased against non-native English writers. Patterns 4(7).
- Mitchell, E., Lee, Y., Khazatsky, A., Manning, C. D. & Finn, C. (2023). DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature. Proc. ICML 2023.
Mentioned in
AI watermarkingAI watermarking, n.: Hiding an invisible signal in AI-generated text, images, audio or video so that software can later tell the content was made by AI — like a banknote's watermark, but undetectable to the eye., classifierclassifier, n.: A program or model that sorts inputs into categories, for example flagging an email as spam or a chatbot message as harmful. Many AI safety filters are classifiers., delvedelve, v.: An ordinary English verb meaning to dig into a subject, which AI chatbots used so often in 2023–2024 that it became a joke and a supposed telltale sign of machine-written text.