Chinese educational authorities have teamed up with tech companies to develop and test new AI technologies that can score standardized tests even better than human graders.

The AI project, which was jointly launched by China’s Ministry of Education and China’s Siri-like tech maker iFLYTEK in 2016, has been tested in national exams, including academic proficiency tests and the national college entrance exams, across the country, and with accurate results. The project aims to understand how AI can be used to assess and score academic performance in order to reduce man-made errors and increase efficiency.

The technology was used this year in senior high school entrance exams in Xiangyang, Hubei province, and it received positive feedback from the examiners. Liu Chaozhi, director of the Xiangyang Education and Examination Institute, told Thepaper.cn that the AI system shows great advantages over its human counterparts.

“The AI grading system can find out if an examinee’s essay is plagiarized by using big data. It can also generate a report, indicating the weak points of a student’s knowledge in an effort to help both students and teachers improve,” said Liu.

According to iFLYTEK, by using the deep learning technologies of the multi-layer artificial neural network, the AI grading system can accurately score essays, avoiding errors caused by human emotions and fatigue.

What separates iFLYTEK’s AI from other foreign robo-grading systems, including e-raters used in the Test of English as a Foreign Language (TEFL), is that it not only attaches importance to grammar and wordings, but also focuses on literary grace and structure.

After reading 500 to 10,000 papers marked by human graders, the system can analyze the scoring features and create a statistical model to produce a comprehensive and accurate final score estimate.

According to iFLYTEK, the new technology can be used to score subjective questions in future exams, covering a wide range of disciplines, such as politics, history, and geography. The AI system will not replace human graders, but will be used to help them focus more on teaching and educational research.