2018-01-17 Wen·Peng Zishan
12353 China’s AI unicorn SenseTime, which just received US$600 million in Series C financing in April 2018, has set a record for the highest investment amount in the history of the AI industry. This company, invested by Alibaba, Singapore’s Temasek Holdings and others, not only focuses on facial recognition, but also provides technology and accumulates facial information for China’s “Skynet” surveillance project, which uses 20 million cameras to monitor 1.4 billion citizens. The commuting, financial, and online activity information of residents in various cities in China will be more transparent. Since then, does personal privacy still exist? Xu Li, co-founder and CEO of SenseTime, accepted an exclusive interview with “World” in Taiwan in mid-January. How does he see the impact of this AI technology on urban life?
Minister of Science and Technology Chen Liangji wrote a long post on Facebook on January 14, saying that he was “deeply touched” by meeting China’s largest AI unicorn, SenseTime. Because “SenseTime has been very ambitious since the beginning of its business, and all Chinese companies have continued to introduce high-tech applications without hesitation.”
In fact, the Chinese government is probably the most active and most proactive in introducing AI technology in China. The “Skynet” surveillance system, which has recently caused an international sensation and attracted extensive coverage by the BBC, Reuters and other foreign media, has more than 20 million surveillance cameras across China. The main provider of the AI technology behind it is SenseTime. According to Reuters, SenseTime’s customers include multiple Chinese public security departments.
SenseTime also confirmed to “Tianxia” that they “provided some technology” for China’s “Skynet Project” monitoring system. A senior executive in Taiwan’s security control industry, who did not want to be named, said that SenseTime should be ranked as the technology supplier of Skynet alongside the Chinese security control giants Hikvision and Dahua.
When “Tianxia” interviewed Xu Li, co-founder and CEO of SenseTime, and asked about his views on the government’s pervasive surveillance of the people, this fast-talking, fast-thinking computer doctor suddenly slowed down his tone and firmly stated that this would have the opposite effect. Security, “Do you think it’s safe if the security guard sees your video, or if the computer (computer) sees your video?”
Relying on world-class computer vision technology, it only took more than three years for SenseTime to be established, and it has become a company with a valuation of more than 1.5 billion US dollars. Today, China may even be the world’s largest AI unicorn.
SenseTime’s legendary deeds began in the second year after its establishment. The founder, Professor Tang Xiaoou, head of the Department of Information Engineering at the Chinese University of Hong Kong, led his students to win the video object detection championship in the 2015 global image recognition and classification competition ImageNet, becoming the first Chinese company to win the championship, shocking the global artificial intelligence industry and academia. advertise
In 2016, it once again won three championships in film object detection, object detection, and scene analysis. Professor Tang Xiaoou’s laboratory has since become famous both at home and abroad, earning the name “Whampoa Military Academy” of computer vision.
Investors also came one after another. In the middle of last year, SenseTime raised US$410 million in Series B financing, setting a record for the highest single-round financing in the global AI field. In the next Series C round of financing, Qualcomm led the investment of US$500 million. Caijing reported that Alibaba also invested 1.5 billion yuan in SenseTime.
The company is named Shangtang, which is derived from the name of Shangtang, the king of the ancient Shang Dynasty in China, in order to recreate the flourishing period of Chinese civilization and expand outward in the Shangtang era in the fields of deep learning and computer vision. Less than 4 years after starting its business, SenseTime is no longer just a technology leader in AI visual recognition, but has 800 employees. It is actively developing a business model and burying deep learning in vertical applications such as finance, security control, and retail.
SenseTime’s technical feature is that it can continuously compress the algorithm, so that different devices, such as surveillance cameras, mobile phones, and Internet of Things devices, can run and complete face or object recognition directly on the central office, so it is faster and more accurate. Save electricity.
In the daily interactions of Chinese netizens, whether it is news-related Toutiao, Weibo, or social networking sites such as Snow, FaceU, etc., SenseTime’s technology is used behind the AR (Augmented Reality) special effects.
Xu Li, 37, received his master’s degree from the Department of Computer Science and Engineering at Shanghai Jiao Tong University and then moved to the same department at the Chinese University of Hong Kong to obtain his doctorate. After gaining experience at Microsoft Research Asia and Lenovo Research Institute, Xu Li met Tang Xiaoou’s research team in Hong Kong and then joined SenseTime to start a business together.
Last year, Xu Li was listed on Fortune’s 2017 list of China’s 40 business elites under the age of 40. On January 11, he accepted an exclusive interview with “World” during his spare time when he came to Taiwan to attend a forum speech. advertise
When AI facial recognition meets China’s demand for security surveillance, Xu Li believes that it will be very easy to arrest people in the future. “The problem to be solved in the future is not to solve crimes, but to prevent them.”
The following is a summary of the interview:
Q: China’s application of facial recognition technology, especially the ubiquitous camera surveillance under the government’s “Skynet Project”, has attracted special attention. Many people are worried that excessive government surveillance will affect personal privacy. What do you think?
Answer: I think the existence of cameras is inevitable. Without facial recognition technology, cameras would rely on people to watch. I remember when I was a kid, there was a public phone downstairs, and someone would call and say, what time did your mother tell you to go there? The news is spread downstairs, and everyone in the whole building knows about it.
In traditional telephones, all connections between different channels are done manually by telephone operators. Many people will feel that the information is not safe, and everyone will know it once it is broadcast. But now everything you say is transmitted to the Internet through channels, through transmissions operated by various telecommunications operators. The telecom operator owns all your information, which is safer because every step is done by machine and no one interferes.
The same goes for facial recognition. Most of it is automated by machines because machines are more accurate than people. You want to check the person in the yellow shirt and which intersection he is at, but the logic behind it will not tell you. This is actually safer. Otherwise, everyone on the building security team will be happy to watch the surveillance video in your building. He can see everything you do in the building. Do you think it is safer when the security guard sees your video, or when the machine sees your video?
Question: According to a BBC report, China’s “Skynet Project” has more than 20 million surveillance cameras. Among them, the “real-time pedestrian detection and identification system” can identify pedestrians’ age, gender, clothing and other information through camera lenses on the road. A BBC reporter did a test. After logging his own photo into the system and walking around on the street, the computer picked it up within 7 minutes. Is this kind of recognition speed the norm in China now? advertise
A: It’s not just a matter of minutes. In China, it is said that in the past two years, there have been no unsolved murder cases, and all cases have been solved using technical investigation methods. Therefore, in the future, what computers will have to solve is not the problem of solving crimes, but the problem of prevention. How to predict that you are going to commit a crime.
Because it has become very easy to catch people, not only with cameras, but also with mobile phones, text messages, etc. As long as you have contact with the world, you will be found.
Question: Since 2014, AI technology has become more and more mature, and the accuracy of face recognition has increased rapidly. Is AI technology mature now and what matters is market application?
Answer: Application is definitely the most important, but this must be a misunderstanding from the outside world. For AI to be implemented, it still needs to “cross the red line” (a mainland term that means “breaking through the technical bottleneck”). For example, face recognition is used in security surveillance and retail applications. It is not only recognition, but also solves technical problems such as lens placement, imaging blur, and connection of image big data. To solve the problem, we have to cross the red line. The red line in many industries is the accuracy of faces. Many serious applications still have very high accuracy requirements.
Q: The most popular application fields of AI currently are Internet finance and security monitoring. Is this also an area that SenseTime focuses on?
Answer: These two fields are slightly higher. The application of artificial intelligence focuses on replacing human labor, because thousands of people are doing certification in finance. With this technology, it is equivalent to laying off thousands of people and embracing artificial labor. The same goes for security control. In the past, surveillance pictures required people to see them, and the police took a long time to check them. Now, machines can see more accurately than people. The first wave of things that can bring industrial value must be artificial labor, which replaces people and does better than others. The second is to change the human-computer interaction model. In the next year or two, new interactive models will generate new games, advertising portals, search engines, etc., which will definitely subvert many existing apps and traffic portals. This is a new portal concept brought by AI. advertise
Q: In terms of revenue, what is the proportion of SenseTime in each vertical field?
Answer: We are divided into four directions. The first is IVA, which is Intelligent Video Analytics, including security, transportation, retail, video, etc. There is already a huge demand for retail and buildings. The second part is identity authentication, including bills, documents, OCR (optical character recognition) authentication, and financial-related applications. This part is growing steadily. The third part is inside and outside the car. The car understands and entertains the driver’s status, including object recognition outside the car, ADAS (advanced driving) analysis, and even driverless driving in the future. This is a new demand point. It will be in 2018. There is a new outbreak. There are also new apps on the Internet. Now Baidu and Toutiao are all provided by our technology, and now they are probably these four cores.
Q: In what ways does the Chinese government assist SenseTime?
Answer: It is unreliable for the government to help technology companies. The investment in science and technology must be huge. Government investment alone will not work.
The correct approach should be when the industry is upgrading. Just like when there were steam locomotives in the past, I told you that if you want to transport, you need to lay railways, from Taipei to Kaohsiung, but you are a horse-drawn carriage, and if you are asked to lay railways, you will say that if there are horse-drawn carriages, then you need to build railways. enough. The government says you lay the railway, and I will give you half of the money. After the railway is laid, you will feel that the train is much more powerful. The same logic applies. Many people want to work on AI. When technology is upgraded, the government has special channels to provide tax cuts and other preferential treatment. The government should help the entire industrial chain, which will greatly promote the development of the industry.
Q: What do you think of the Taiwan market? What applications have great potential?
Answer: Taiwan is small, but its urbanization process is relatively high, and the cost of connecting everything is relatively low. For example, in residential areas, facial recognition can be directly introduced, or retail experiences can be created. Electronic payment in Taiwan is not as popular as in China. Credit cards are still used to swipe the card, but it can be skipped. In Taiwan, it might be possible to jump directly to face payment and skip the process of swiping the QR code. Missing a stage is not a big problem, the question is whether there are conditions for a pre-emptive strike. advertise
Q: What is the current status of cooperation between SenseTime and Taiwan?
A: We are talking about it, including in the retail industry.
Editor in charge: Wu Kailin
(This article is reprinted from ” Tianxia Magazine ” and authorized to be published by “Future City@Tianxia”. Please do not reprint without permission.)
Newsessentials Blog 
Posted on April 30, 2024
0