Many Chinese provinces are using artificial intelligence to monitor the national college entrance exams. Gaokaoand crack down on fraud and other violations.
A record 13.42 million students have registered for this year's exams, which begin on Saturday, according to the Education Ministry.
According to domestic media, provinces such as Guangdong and Hainan in southern China and Shandong in the east have introduced “artificial intelligence patrol systems” to monitor exams to step up monitoring and ensure the fairness of the tests.
04:32
What is the Gaokao? A look into China's terrifying college entrance exam
What is the Gaokao? A look into China's terrifying college entrance exam
“[We should] “We will actively promote real-time smart inspection of examination rooms and secret rooms and further strengthen our high-tech system to prevent cheating,” the ministry said on its website on June 1.
In China, Gaokao The university entrance exam is widely considered to be the most crucial test that will determine a young person's future. Competition to get into the country's top universities remains fierce, leading authorities to take extra measures to prevent cheating.
Regulators have turned to a range of technology in recent years to detect fraud, including the deployment of drones in 2015.
And since 2016, Gaokao It may be treated as a criminal offence.
According to the Guangdong Daily, the ministry has deployed AI in 386 exam centres to detect cheating, plagiarism and other abnormal behaviour through image and video data. If anomalies are detected, the system will immediately sound an alarm, alerting supervisors to take action.
But the AI system will not fully replace human eyes: Any warning information provided by the AI will be manually verified, with human inspectors making the final decision in case of disputes, Chinese media outlet Hongxing News reported.
The test lasts between 75 and 150 minutes, and according to provincial government regulations, there are usually two proctors in each testing room monitoring 30 candidates.
Security checks and online monitoring have been implemented at exam centres since 2013, but cheating still occurs. In 2021, a test-taker uploaded photos of math problems to a search app during the exam.
Through repeated training, the AI now far surpasses the accuracy of human monitors, according to a paper analyzing the use of AI test monitoring, written by a team of researchers from the Sichuan Provincial Education Testing Bureau and led by Yang Xiaolin, the bureau's IT director, and published in the Journal of Educational Testing and Evaluation in March.
AI can process large volumes of surveillance video and reduce review times, and its learning capabilities can turn large datasets into learning materials to help improve future test surveillance.
To develop an AI-based fraud detection system, the Sichuan research team said, they must first collect and annotate a large dataset of normal and anomalous behavior during testing.
Computer vision technology is used to capture and process images of the testing environment to extract visual features. Machine learning algorithms are then trained on these image features to establish models that identify anomalous testing behaviors, which can then be applied for real-time testing monitoring.
But the paper also notes challenges with AI systems: While they can monitor in real time and report any anomalies to various levels of management, standardized exam rooms tend to have tightly controlled wireless communications, meaning that in some cases patients may have to rely on manual reporting.
