Detecting AI

AI Detectors are themselves AI, and thus are not 100% accurate.

GenAI detection tools privilege students who are English first language, have access to paid large language models/applications, and are more digitally literate. 

-Leon Furze  (source)

AI detectors are more likely to give false positives to writing by:

• ESL students

• Students with low digital literacy

• Students who use simple vocabulary or repetitive sentence structures

These students may be unfairly penalized while more privileged students remain undetected.

As AI detectors become more common, students are getting smarter about avoid detection, and it doesn’t take much to trick a detector:

• Paraphrasing tools

• Manual rewrites

• Sentence restructuring

• Using multiple AI tools in sequence

Because of this, an AI detector score cannot be your primary line of defense when initiating an academic misconduct claim.

Some instructors have started embedding hidden “gotcha” instructions in assignments, such as white text that tells students to include a specific phrase or detail as a test for AI use. While well-intentioned, these tactics can cause serious harm:

• Confuses students who are using screen readers, screen readers will detect and read out all text, including hidden white text

• Erodes student-teacher trust

1. Why AI writing detectors don’t work (Ars Technica) 

2. Large Language Models can be Guided to Evade AI-Generated Text Detection 

3. AI Detection in Education is a Dead End 

4. GPT detectors are biased against non-native English writers 

5. Turnitin's AI writing detection capabilities FAQs 

 Next: Protecting Privacy