Transcript
Video duration: 1:05
My name is Yang Jiang.
I am a research scientist at ETS.
With the rapid development of large language models such as ChatGPT,
researchers and educators have raised concerns about students using these tools
to cheat and its threats to test security and academic integrity.
In our study,
we developed machine learning models to automatically detect ChatGPT generated
essays in GRE, a large scale educational assessment.
Our detectors were able to identify
AI-generated essays with an accuracy close to 100 percent.
We further explore the fairness of these
detectors when they are applied to essays written by native English speakers as
opposed to written by non-native English speakers.
Our findings provide guidelines
for developing both high performing and fair AI models in education