Although China has been closely surveilling its citizens’ internet usage for the last decade, the Chinese government only recently began constructing a “smart surveillance” CCTV system that matches the extent of existing online surveillance tools. The smart surveillance system is currently being piloted for several causes, from suppressing uprisings in the far western Xinjiang Autonomous Region, to policing jaywalking in the urban areas of Shenzhen in Guangdong. Chinese companies produce much of the smart surveillance technology, setting this new system apart from older imported online security hardware. But what makes smart surveillance more dystopian than earlier systems is the integration with centralized databases and artificial intelligence (AI) image recognition technology.
Smart surveillance’s first strength is in its ubiquity: China already has more than 175 million CCTV cameras, and plans to have a network of more than 450 million by 2020. The other key features of smart surveillance are its integration with databases and the application of AI. China has digitized census information linked with photographs, fingerprints and other records, creating a citizen database of unmatched quality and comprehension. But until recently, even excellent databases were useless without humans policing video footage for wanted individuals. Now, surveillance is automated. Substantially improved facial recognition and object tracking algorithms allow live video analysis tools to algorithmically search a database and notify authorities nearest to a target’s location. Smart surveillance connects AI and image databases, allowing it to achieve unprecedented results.
Thus far smart surveillance is primarily deployed locally and, in a few cases, regionally. The most extensive network is located in Xinjiang, a province home to the Uighurs, a predominantly Muslim ethnic minority in western China. After major protests in Xinjiang in 2009, Beijing dedicated more resources to domestic security there. More recently, Xinjiang’s new security chief, Chen Quanguo, implemented a system of CCTV cameras while simultaneously forcing Uighurs to install spyware on their cellphones, swabbing citizens for DNA, and creating a database of “risky” individuals. Although Xinjiang’s system seems like dystopia realized, smart surveillance has other, more benign uses. In both Shenzhen and Shenyang, new facial recognition cameras identify jaywalkers, and digital signs subsequently display violators’ faces. In Shenzhen, police have connected a citywide population database with cellular networks to send fines in real time via text message as offenders are identified.
Another unique aspect of smart surveillance is that much of the equipment is designed and manufactured within China, and the technology is even being exported abroad. Contrast that with the the Great Firewall and other online surveillance tools’ network management equipment, which foreign companies helped provide. Not only the hardware is made in China, the software database systems and facial recognition algorithms are likely homegrown as well. While the providers of government software sometimes remain private, given that Baidu has achieved world-record-setting facial recognition results and has often collaborated with the Chinese government in the past, it is feasible that Baidu may have contributed its algorithms.
So, the key question is: can China actually implement a nationwide system with these new technologies? The short answer is only with serious effort. Implementing a nationwide system is simply too costly. Since 2015, China’s domestic security spending, including smart surveillance, has nearly breached $150 billion, exceeding China’s military budget. And the system is only implemented fully in one region and partially in a handful of cities. Beijing could take a “per capita” approach and only surveil densely populated areas, which might serve party authorities well should they want to monitor Hong Kong. The central government could also specifically target Shanghai, Beijing, Tianjin and Chongqing. These four provincial-level cities fit the per capita logic and are also more likely to be targeted by threats to CCP authority, such as protestors or even terrorists. But unless cameras and networking technology drop substantially in price, or artificial intelligence systems learn to surveil more without additional camera installations (perhaps cellular networks could play a larger role), China’s system will remain a powerful panoptic force in Xinjiang (and perhaps the most densely populated cities) but will likely stay restricted to general law enforcement in already low-threat areas.
Avery Jordan is a senior at Rice University majoring in Asian studies and computer science and working for the China Studies Program. He received a Huayu Enrichment Scholarship for the 2018-2019 school year and is currently studying Mandarin Chinese at the International Chinese Language Program at the National Taiwan University. His interests include artificial intelligence policy, LGBTQ rights, and social media in Asian contexts.