Authors Guild test finds some AI detectors perfectly identify human writing while others fail on every single text

Insight summary

• •The Authors Guild tested several AI detectors on ten human-written articles from 2020-2022 and found Pangram and Grammarly correctly identified all as human-written.
• •Originality.ai also performed well, while Sidekicker and ZeroGPT often falsely flagged human texts as AI-generated.
• •False positives in AI detection can jeopardize authors' contracts and reputations, prompting calls for transparency and author defense opportunities.
• •Pangram's CEO highlighted that AI detection tools operate as black boxes and struggle to distinguish highly skilled human writing from AI-generated text.
• •These tools prioritize minimizing false positives, so their accuracy in detecting actual AI-generated text remains uncertain.
• •The article emphasizes the ongoing cultural debate about the value and impact of AI writing tools and the limitations of current detection technologies.
• •The debate includes concerns about AI potentially flooding the internet with low-quality content and the social contract between writers and readers.