protonmail-verification=b75cd457bf1b40b74b4827e4c6e319fc096789b3
top of page
mie128

“From Seeing to Understanding”: How Public Surveillance Targets Uyghurs with AI Cameras Throughout China


Author: Björn Alpermann

2 July 2024

 

It has long been known that China’s security apparatus uses facial recognition cameras and AI-driven software in its surveillance of public space in the Xinjiang Uyghur Autonomous Region. But it has been less recognized that these systems – which specifically target Uyghurs – are to this day widely used throughout the country. In April 2019, a report by the New York Times used public procurement documents published by local governments and their security bureaus, as well as interviews with some insiders, to demonstrate the extent and systematic approach to monitoring the Uyghur population far away from their home region. However, since then, to my knowledge, no update on these strategies has been published. Recently collected official procurement documents for AI-driven camera and software systems show that this rollout of surveillance particularly targeting Uyghurs still continues.


Where do the documents come from?


As first reported in the German online media China.Table by Marcel Grzanna, these documents shed light on the way the system operates. The journalist was given access to one tender document from the Xuhui city district of Shanghai[1], which he shared with me. Doing follow-up research, I then additionally collected eight more such documents mostly from east China, but also the southwest (Sichuan and Chongqing) – regions far away from the XUAR where the majority of Uyghurs reside. All these documents are (until now) publicly available on the Chinese internet and copies are now provided on this website.


Specifically, these procurement documents hail from Dongbu New Area of Chengdu city[2], (Sichuan Province), Bishan city district, Chongqing Municipality[3], Chengyang city district, Qingdao city[4],(Shandong Province), and Guli street office, Nanjing city[5] (Jiangsu Province) as well as four places in Zhejiang Province: the cities of Shaoxing[6] and Wenling[7] and the two counties of Kaihua[8] and Xianju[9]. In total, I now hold nine tender documents of which five were discussed in the China.Table report. 


In terms of timing, the oldest in this set is from 2018 (Kaihua), while the Shanghai document is the most recent (2023). The others date from 2019 (Qingdao, Wenling), 2020 (Chongqing, Nanjing, Shaoxing) and 2022 (Chengdu, Xianju).



A = Qingdao    B = Nanjing      C = Shanghai   D = Shaoxing   E = Kaihua   F = Xianju

G = Wenling  H = Chengdu I = Chongqing 

Dark color = Xinjiang Uyghur Autonomous Region (XUAR)

Map based on ©OpenStreetMap Contributors



What do the documents say?


The AI-driven camera and software systems described in the documents are meant to enhance the surveillance capacities of local security organs in public spaces. In the words used in the Chengdu document, the goal is to transform their capacity “from seeing clearly to understanding” (由“看得清“向”看得懂“的转变). Providing the basis for public project tenders, these documents stipulate in great technical detail which functions the AI-assisted camera and software systems have to fulfil. For all the above cases, identifying Uyghurs using facial recognition is listed as one of the criteria – besides categories such as age, gender, and wearing glasses. The surveillance is explicitly and exclusively directed at Uyghurs, not other ethnic groups (the systems are required to determine whether a person is or is not a Uyghur: “是否维族人”). The Qingdao document is even more explicit in requiring a successful detection rate of Uyghurs at no less than 97 percent (“维族人识别准确率不低于97%”). This rate of precision is slightly less than that for correctly identifying other criteria such as age, gender and wearing glasses.


The documents also reveal a lot about how the systems to be purchased are linked with big data analyses. The Shanghai case, for instance, requires that the principle “one person, one file” (一人,一档) is implemented. Persons should be identified with their real names, addresses, ID numbers etc. in order to track their movements. Though not all documents explicitly refer to it, the local systems appear to be part of the national video surveillance system called “sky net” (天网).


What is the meaning of these documents?


For those who follow the repressive policies against Uyghurs and other minoritized ethnic groups in Xinjiang, the documents will not come as a surprise. Yet they provide evidence for a number of observations. First, the documents show that Uyghurs are still being seen as potential criminals or “terrorists” by security agencies throughout China. They suggest that this classification warrants them taking extra precautions and specifically identifying Uyghurs when they move through public spaces. Note that none of the other 54 non-Han ethnic groups officially recognized in China are mentioned in any of these files (although the Chengdu document additionally mentions “racial attributes” 种族属性 as another criterion to watch out for). Other procurement documents reviewed for this blogpost are more circumspect. For instance, one issued by Beijing[10] municipality in 2019 requires the system to manage “identification of non-Han persons including independent alarm” (非汉族人员识别以及独立报警). And another one published by Gansu[11] province in 2018 first requests the function to identify “ethnicity” but later specifies that this means Han and Uyghurs only: 民族(汉族和维族). That said, the 2019 NYT article mentioned above already quoted three insiders who reported that terms such as “minority identification” are nothing but “a euphemism for a tool that sought to identify Uighurs exclusively“.


There is additional evidence to back up this claim of ethnic targeting directed against Uyghurs alone. For instance, another police document, this time from the Wuchang district of the central Chinese metropolis of Wuhan[12] (Hubei Province) dated 2021, singles them out for registration under the heading “Constantly improving the capacity to deal with terrorism and to maintain stability”. Alongside the registration of Uyghurs, other preventive measures mentioned in this section are “the treatment of mentally ill persons who have been involved in accidents, and for the rehabilitation of persons suffering from drug addiction”. Similarly, a procurement document of Baoding[13] city, Hebei province, as recently as February 2023 lists Uyghurs among “persons of concern” (关注的人员(如维族人员)). It requests that the AI system it wants to purchase conducts an analysis to evaluate the likelihood that these persons stayed in the locality even after they disappeared from camera tracking (轨迹消失后仍在本地留存的可能性进行评估分析). These examples underscore that Uyghurs are generally distrusted by Chinese security agencies and treated as source of trouble. In a similar vein, a procurement document for a video image system published by the Minhang district[14] of Shanghai city stipulates against which databases images taken should be compared: Besides databases for fugitives and ex-convicts it lists a “database of high-risk Uyghurs” (维族高危人员库). The mere existence of such a database specifically designed just for one of China’s officially recognized 56 ethnic groups is reason for concern.


Second, the wide geographic spread of the documents (six different provincial-level units) shows that this targeting of Uyghurs is not just a local anomaly but a regular feature of the surveillance state that the CCP is building in China. In fact, Uyghurs have reported for decades of the difficulties they face to stay in hotels or access public services in other parts of China. Third, that the documents have been issued in various parts of China for at least six years now, suggests that this system of surveillance is being rolled out systematically – though further evidence is needed to conclude this with higher certainty. Taken together, these findings indicate that the party-state promoted narrative that Uyghurs in China are free to enjoy their lives since “counterterrorism measures” like reeducation camps have improved the security situation cannot be trusted. Instead, they are still being collectively treated as security threat and therefore as second-class citizens.




Documents


[2] 


[3] 


[4] 


[5] 


[6] 


[7] 


[8] 


[9] 







301 views0 comments

Comentarios

Obtuvo 0 de 5 estrellas.
Aún no hay calificaciones

Agrega una calificación
bottom of page