The United States’ most important document, the US constitution, has been flagged as AI generated in a worrying test of anti-AI software.
As ChatGPT becomes one of the most widely adopted internet tools in history, many are using the service for university essays, schoolwork and more. As the tool becomes more prominent by those looking to quickly complete work, AI detection software is quickly coming to market.
However, it seems that the current generation of AI detection software is far from perfect as a number of perfectly legitimate documents are being flagged as AI-generated works.
As pointed out by Twitter user Gaut, current-gen AI detection tools flagged the US constitution as AI-generated content. In a worrying test of AI text detectors, the historical document is flagged as being filled with AI text.
According to the detection software, the IS constitution is 92.26% AI generated. The detector shows that around 30% AI content is an acceptable level for a document. However, the detection software shows just how ill-prepared companies are to deal with artificial intelligence.
AI-detection services such as ZeroGPT and TurnItIn are built on pre-existing plagiarism checkers, all of which would pick up the American constitution as stolen copy if it was posted verbatim. What’s more intriguing is that the 8% of the US constitution that isn’t detected as generative work, despite being in circulation for over 230 years.
As universities, colleges and schools fail to regulate AI written essays and tools, false positives within AI detection software could become a huge issue. Students already have issues with modern plagiarism checkers over-analysing citations and quotes, but with ChatGPT detection on top, this could become a huge detriment to essay work.
Of course, AI-generated works do need to be detectable in order to properly vet content online as well as in education. However, as long as they are as inaccurate as they currently are, it seems they cannot be trusted.