The Center for Security and Emerging Technology (CSET) has published a new study that suggests that China is disproportionately invested in surveillance technology. The findings are based on a survey of publicly published academic research papers that focus on different kinds of AI-powered computer vision applications.
In raw numerical terms, researchers associated with Chinese organizations accounted for nearly half of the spoof detection and crowd counting studies (42 and 46 percent, respectively) and more than half of the person re-identification studies (56 percent). The country also accounted for a disproportionate number of the studies on emotion recognition (32 percent), action recognition (35 percent), and facial recognition (39 percent) more generally.
Those figures outstrip the collective output of the western world, and may even underestimate the scope of China’s commitment to AI surveillance. The CSET report (titled Trends in AI Research for the Visual Surveillance of Populations) only looked at scientific papers that were published in English, and therefore may have overlooked papers that were published in Chinese. The authors could similarly not factor corporate or classified state research into their final results, and they acknowledged that other data points (such as patents and government policy) could paint a more complete picture of the current state of surveillance research.
Even so, the findings would seem to suggest that China is far more committed to surveillance infrastructure than the rest of the world. The country has frequently been criticized for its invasive surveillance practices, whether facial recognition is being used to monitor children playing video games, or to shame people for wearing pajamas in public. In that regard, the misuse of surveillance technology represents a major human rights concern. The Chinese companies Huawei and Megvii have come under fire for allegedly developing a surveillance system that was designed to track members of China’s minority Uighur population, while China itself has tried using emotion detection technology to try to predict criminal behavior.
It is worth noting that surveillance only represents a small fraction (less than 10 percent) of all computer vision research. The report was based on papers that were published between 2015 and 2019 in six leading academic datasets, with facial recognition standing as the single most investigated application.
January 11, 2022 – by Eric Weiss