Colleges struggle with AI writing detection: tech limits and errors

It's graduation season, and Chinese universities face a new challenge: papers that sound too much like AI wrote them. Besides checking for plagiarism, blind review, and defense, graduates now have to pass AIGC detection. Some students found that after using AI to rewrite their papers, the AI detection score actually went up, from 62% to 94%. This is not a rare case.

Cai Hailong, from Capital Normal University, explained that AI detection is based on probability, not certainty. The core issue is that we're using AI to detect AI, making it hard to tell if a human or machine wrote something. Chinese language's rich meanings also cause more errors.

Experts suggest schools should have clear rules on AI use and rely more on human review than just AI detection. Currently, most universities use systems like CNKI for AIGC checks, which look at 'perplexity' and 'burstiness' to spot AI text. But none of these methods are 100% accurate, and mistakes are common.