by Giorgio Cuscito
In just eight years, SenseTime, a start-up that sprouted in Hong Kong, has become a pillar of Beijing’s AI strategies and a target for America. The uses of facial recognition in Xinjiang, the Sky Net project and ties with Saudi Arabia.
1. The epic of SenseTime is the most tangible example of the progress made by the People’s Republic of China in the field of artificial intelligence (AI) over the last ten years. The focus of this company born in Hong Kong in 2014 is on computer vision, i.e. the branch of AI that enables it to collect, classify, and ‘understand’ information from digital images, videos, and other visual inputs.
Thanks to its unique capabilities, SenseTime has quickly become the world’s most valuable ‘unicorn’ (start-up worth over $1 billion) and has been included along with Baidu, Alibaba, Tencent, and Hikivision in the line-up of national champions in charge of driving Chinese progress in the ‘military-civil fusion’ (ronghe). It is up to them to develop the latest generation of tools with which the People’s Republic wants to compete with the United States on the economic and military fronts. Therefore, it is not surprising that Washington has included this rising AI star among the players to be barred from access to American technology.
SenseTime excels in three interlinked activities: facial recognition, human verification, and crowd analysis. The first allows people’s faces to be captured and identified. Then, the collected images are compared with the data already held by the authorities. This is precisely human verification, which in turn makes it possible to study the behaviour of large crowds. The aim is perhaps to manage the heavy traffic in the megacities on the Chinese coast. Or to understand if a protest is about to break out, and to identify the perpetrators and prepare countermeasures. With the ultimate aim of preserving domestic stability and thus defend Beijing’s sovereignty.
The topic of surveillance is made particularly topical by the demonstrations that took place in November 2022 in several cities in China against the strict zero-Covid tactics. That is, the stringent control and prevention measures applied by the government led by President Xi Jinping to contain the epidemic after its outbreak in Wuhan in 2019. The protests prompted Beijing to speed up the relaxation of these measures. However, a few days after those events, police in the capital summoned several people by telephone to ask them to account for their participation in the gatherings. It is unclear what tools were used to identify the protesters, but the use of geolocation via smartphones and data collection based on photos and videos may have been decisive.
The Chinese authorities make extensive use of such resources. For instance, SenseTime provides technology to various police offices, banks, mobile phone operators, private security companies, and smartphone manufacturers. Above all, the Hong Kong-based company cooperates with the Ministry of Public Security (MPS, responsible for domestic intelligence among many other things) in the Sky Net (Tianwang, Celestial Network), Fox Hunt (Liehu, Fox Hunt) and Sharp Eyes (Ruiyan gongcheng, Sharp Eyes Project) operations. The first two are aimed at fighting crime, corruption, and arresting offenders who have fled abroad. Sharp Eyes consists of a surveillance programme of the Chinese population by means of more than 200 million cameras with facial recognition deployed in the country.
Residents experience this constant monitoring with some anxiety. According to a poll conducted in 2021 by Beijing News Think Tank on a sample of 1,515 people, 87% were against the use of facial recognition in commercial areas. 68% believed it should not be used to access residential areas. Between 43% and 52% thought the technology should not be used in facilities such as hospitals, schools, and offices either. The main concern was related to the risk of data theft (96%), followed by privacy issues (91%).
Surveillance operations extend beyond China’s borders. In 2021, 1,273 fugitives were reportedly captured via Sky Net. This was a triumph for Beijing, which last January promised further expansion of the Sky Net via the Belt and Road Initiative (BRI, New Silk Road), the project with which Xi wants to increase the People’s Republic’s geopolitical projection abroad. This move also coincides with the recent discovery of a hundred or so ‘overseas police stations’ employed by the MPS in at least 53 countries, including Italy. Their purpose is to monitor the Chinese diasporas and their host countries.
SenseTime has several foreign partners and has offices in states such as the US, Japan, Singapore, South Korea, Malaysia, the United Arab Emirates, and Saudi Arabia. Just last September, the joint venture formed with Saudi Company for Artificial Intelligence (SCAI) was commissioned to set up an AI laboratory in the Middle Eastern country. The long-term aim is to contribute to its progress in the context of the strategic partnership signed by Beijing and Riyadh during the meeting between Xi and Crown Prince Muhammad bin Salman al-Saud in December. A memorandum was also signed for the event, according to which Huawei will contribute to the construction of the cloud network of some Saudi cities.
Looking ahead, Beijing wants to use the digital branches of the new silk roads to technologically penetrate the Middle East, increase regional dependence on its infrastructure, collect data on local communities, and monitor the emergence of terrorist entities that could penetrate China via Xinjiang.
The official motivation for Washington’s sanctions against SenseTime is precisely its contribution to the repression and assimilation campaign conducted by Beijing against the Uyghurs, a Muslim and Turkic-speaking minority that inhabits this unstable region of the People’s Republic. Yet for the White House, the stakes are bigger. It is about severing the ties that the company has with technology players in the United States or their satellites, such as the Massachusetts Institute of Technology (MIT), the semiconductor company Qualcomm or the Japanese company Honda. More generally, America wants to prevent China from dominating the supply chain of artificial intelligence as this would give it an advantage in their confrontation.
The People’s Liberation Army has long established the need to move from conducting ‘computerised’ wars (i.e. focused on the enhanced circulation of information) to those in which the use of AI, cloud, and big data affects all dimensions of warfare, starting with the process of command and control. The concept is expressed with the term zhineng hua, translatable into ‘intelligentisation’.
Beijing has only started to really question the quality of its technology and the viability of artificial intelligence in the military field a few years ago. The debate was fuelled by the victory in a game of go (or weiqi) achieved by by Google DeepMind’s Alphago software against Chinese champion Ke Jie in 2017. The weiqi (or game of encirclement) is one of the elements that best describe the strategic culture of the People’s Republic. The aim of this pastime is not to clash head-on with the opponent’s pawns, as in chess. Rather, it is to encircle them on several sides of the game board. At the end of the game, the areas of domination of the two opponents are so intertwined that it is difficult to determine who has won.
The triumph of AlphaGo triggered the SenseTime retaliation, which last August unveiled SenseRobot: a small android with artificial intelligence capable of competing in xiangqi (a hybrid game between chess and weiqi) on 26 difficulty levels and with over 100 endgame scenarios.
After a few months, SenseRobot defeated two Chinese champions in a tournament. It thus confirmed Beijing’s convictions about the strategic use of AI and proved that the People’s Republic is also capable of building a device with the same capabilities as AlphaGo. What’s more, by offering it for sale as a latest generation toy, the Hong Kong company now encourages members of the community (and in particular the younger generation) to improve their decision-making skills, thus preparing themselves to the drastic interpenetration in the making between the real and digital worlds, both at stake between the US and China. It also equips them with the awareness, also expressed by Beijing, that AI can help but not replace human discernment, especially with regard to strategic decisions.
2. The story of SenseTime founder Tang Xiao’ou underlines the impact that interaction with America has had on the technological progress of the People’s Republic. Tang was born in Anshan (Liaoning) in 1968 and fully experienced the Chinese economic boom triggered by the reform and opening-up policy promoted by Deng Xiaoping. By his own admission, frequent reading of picture books during his childhood prompted him to apply artificial intelligence to images. Between 1990 and 1991, Tang graduated in computer science first at the Chinese University of Science and Technology in Anhui and then at the University of Rochester in New York. Five years later, he earned a PhD from the Massachusetts Institute of Technology in Boston, where he worked on submarine robotics.
After finishing his studies, Tang started teaching at the City University of Hong Kong (CityU), where he opened a multimedia lab dedicated to computer vision in 2001. This is the centre where SenseTime later geminated. The Perfumed Harbour had just been returned by the UK to the People’s Republic. Compared to the rest of the country, the local population enjoyed greater freedom of economic, social, cultural, and academic interaction with the West – an interaction which would later be significantly curtailed by Xi from 2013 onwards. However, Tang had few contacts in Hong Kong – AI was not the most followed topic at CityU and the brightest students had already gone to study in the US.
Therefore, the young lecturer reconnected with the University of Science and Technology of China (USTC) in Hefei (Anhui), his alma mater. His objective: to gather the best minds for inclusion in the lab. Among them were Li Xuelong, Wang Xiaongang, and Tao Dacheng, future pillars of SenseTime. Before long the Hong Kong University became a vibrant AI research centre, to the point of being considered the ‘Whampoa Academy’ of the Chinese digital industry. This was a clear reference to the military institute founded in 1924 by Sun Yat-sen, where the leading commanders of the major conflicts in which China was involved in the first half of the 20th century, including the civil war between Communists and Nationalists.
Tang’s work between 2005 and 2008 at Microsoft Research Asia (MSRA, based in Beijing – the company’s largest research lab outside the United States) also stimulated the lab’s growth. Zhang Ya-qin (later chairman of Baidu) and Jack Ma (founder of Alibaba) were also students at the centre, founded in 1998 by the Taiwanese Kai Fu-lee (now an Ai guru). Tang focused on computer vision and developed visual algorithms for products such as Xbox, Kinekt, Windows Hello, and the image search engine Bing.
In 2014 came the breakthrough: the Hong Kong lab produced the DeepId facial recognition software. The programme had an accuracy rate accuracy rate of 98.52%, which is almost one percentage point higher than that recorded by the DeepFace algorithm developed by Facebook. Above all, it was the first time that artificial intelligence showed a higher performance than humans. Within a few months, Tang and eleven other CUHK academics founded Shangtang (SenseTime) with the initial contribution of ten million dollars from Boston-based IDG Capital, one of the most active investment companies in China.
The start-up’s name underlined its founder’s ambitions. The Emperor Shang Tang is famous for starting the Shang dynasty (1600-1045 BC), which was particularly prolific in terms of technology. The first archaeological finds of Chinese-written texts, bronze weapons and horse-drawn carriages date back to this period. These changed the way warfare was conducted and for centuries gave the Shang dynasty a decisive advantage on the battlefield. At least until it was defeated by the Zhou kingdom. As stated by Xu himself, the name was chosen because the company wants to bring the country back to the technological glories of the past. A stylistic decision undoubtedly welcomed by Xi Jinping, who has made the ‘resurgence’ of China the pivot of his geopolitical project.
It is no coincidence that the rise of SenseTime is collimated with the president’s plan to transform the People’s Republic into an AI power by 2030 and with the tough anti-terrorism and repression campaign in Xinjiang. This tactic involves the notorious ‘vocational education camps’, detention centres where impose ethnic Han customs on Uyghurs. To this end, the government employs sophisticated technologies – including SenseTime – to identify the minority and monitor its members’ behaviour.
In 2018, the Hong Kong company was listed as one of the ‘national AI champions’ and funnelled capital from entities such as the UK’s Fidelity International, the US-based Silver Lake Partners, Tiger Global Management and Qualcomm 10. It has also signed collaboration agreements with several Singaporean players: the Temasek company, Nanyang University, the National Supercomputing Centre, Singapore Telecommunication Limited. Above all, SenseTime signed an ‘alliance’ with the Massachusetts Institute of Technology, with which Tang had contacts through his doctorate. The result was the launch of 27 research projects, involving some 50 faculties of the American university.
Then, during the second New Silk Road Forum in Beijing, the unicorn announced with Malaysian G3 Global and the port giant China Harbour Engineering the construction of an AI park near Kuala Lumpur. It is a perfect example of how the People’s Republic is simultaneously investing in its transformation into a maritime and digital powerhouse.
However, in 2019, SenseTime’s activities within Sky Net came to the fore. Researcher Victor Gervers discovered that SenseNets (supported by Tang’s company until 2018) had taken in almost seven million GPS satellite locations and the private information of 2.6 million people, mostly located in Xinjiang. A few months after these revelations, the Hong Kong company also sold its stake in Tangli Technologies, through which it provided monitoring equipment to the Xinjiang police in cooperation with Leon Technology.
The move was not enough to avoid being placed on the US Department of Commerce’s list of entities prohibited from purchasing US-made technology unless they hold special licences. However, a detail in the official document allowed the company to partially dodge the sanction mechanism: the entity called into question is only Sense-Time Beijing, i.e. the centre located in the capital, formally responsible for the sale of software.
Despite this, Beijing chose Tang’s company as the leader of the working group for the definition of standards in the field of facial recognition in China. At the same time, the company intensified its activities in the healthcare field, stemming from the need to make the management of the coronavirus epidemic more efficient, and strengthened its operations in the Middle East, particularly in Saudi Arabia.
Washington delivered a new blow to SenseTime in December 2022, when the Treasury Department listed it as part of the Chinese military complex and responsible for the repression in Xinjiang. Thus it has prevented American counterparts from investing further in the Chinese unicorn, which postponed its expected debut on the Hong Kong stock exchange by two weeks.
Despite everything, Tang’s company withstood all adversities. The collaboration with MIT also continued, at least until February 2022. In addition, Japan Computer Vision (JCV), owned by SoftBank, offered together with SenseTime its biometric technologies to Visa and Mastercard. Incidentally, JCV is also one of the players that have invested in the identity authentication system of PopId, a rapidly expanding Pasadena-based company. Its story symbolises the difficulties encountered in unravelling the complex production chain of AI.
3. Despite being declared a US target, SenseTime is now expanding its activities into other sectors. For instance the automotive sector, through the development of intelligent in-car devices and the marketing of a system of automated engine fault detection. This is used by Foton Cummins of Beijing, which is the world’s leading independent manufacturer in this sector. Moreover, the debut in the automobile industry was coupled with the collaboration with the Alfa Romeo Orlen Formula 1 team, operated by the Swiss Sauber under the Italian brand name. This was a direct consequence of China’s desire to position itself as the driving force behind the boom in smart and electric cars.
The final frontier is the Metaverse. SenseTime aims to create urban contexts that fuse physical reality with digital reality using artificial intelligence. In short, a virtual world without barriers only in appearance, since establishing its boundaries and monitoring the flow of data will be the actors who dictate its rules. For this reason, the Hong Kong company’s plan is to position itself as a point of reference in this technological segment on a par with Baidu, Tencent, and TikTok (Douyin, also in the American crosshairs) in the ambit of Beijing’s ‘cybernetic sovereignty’ over the Chinese Web.
The first theatre of application of such projects outside the People’s Republic is the ‘Asia-Pacific’, as can be seen from the plan published by SenseTime last October 17. The terminological choice did not happen by chance. It serves to distinguish the Sinic vision of geographical surroundings from the American and Japanese one, which envisages the inclusion of India in the ‘Indo-Pacific’ regional quadrant in order to contain Beijing’s ambitions. In short, Tang intends to contribute to China’s strenuous attempts to shape its sphere of influence in the riparian waters and push back the presence – physical and virtual – of the United States at its doorstep. These are the first signs of the role that the Metaverse and SenseTime could play in the digital evolution of the new silk roads. That is, if Washington does not succeed in interrupting the gallop of the unicorn watching China.
The article appeared on Limes 12/2022 L’intelligenza non è artificiale.
Translated into English by Mark Sammut Sassi.