The 6th Digital China Construction Summit was held in Fuzhou on April 27, and many AI artificial intelligence products were unveiled. However, the Chinese government has only recently drawn a red line for the management of generative artificial intelligence, requiring that it must “reflect socialist values”. Analysts said that this may put China’s AI development in a dilemma, because any AI that can be artificially controlled by the CCP It will not be a very powerful AI; a truly powerful AI must be trained and learned through a large amount of comprehensive and extensive data, but this may produce an AI that is difficult for the CCP to control in the end, causing the CCP regime its own risks .
China’s 6th Digital China Construction Summit was held in Fuzhou, Fujian Province from April 27 to 28. Xinhua News Agency stated on April 27 that the theme of this summit is to “accelerate the construction of digital China and promote Chinese-style modernization, and focus on displaying the latest achievements in digital China construction.”
On April 26, Alibaba Cloud, a Chinese Internet giant, announced the launch of the “Tongyi Thousand Questions Partner Program”, and worked with several partners to create a large language model that can be applied to different industries. Among them, the first batch of large-scale industry models will cover petrochemical, electric power, transportation, hotels, corporate services, communications and finance and other fields.
Tongyi Qianwen is another company in China that launched chatbots after Baidu’s “Wenxinyiyan” and SenseTime’s “SenseNova” and “SenseChat”. The Chat GTP developed by the American start-up company Open AI is different from each other.
AI management red line
On the same day that “Tongyi Qianwen” was announced, the China Internet Information Office also announced the draft “Generative Artificial Intelligence Service Management Measures”. The draft stipulates that the content generated by artificial intelligence should reflect the core values of socialism, and must not contain content such as subverting state power, overthrowing the socialist system, instigating secession, or undermining national unity. The Security Law stipulates that users are required to provide real identity information. Previously, it was necessary to declare a security assessment to the national network information department and perform procedures such as algorithm filing.
In recent years, Chen Zhijie, an associate professor at the Institute of Political Science at Sun Yat-sen University in Taiwan, who has devoted himself to researching Chinese issues with digital and text mining technology, said in an interview with Voice of America: “In China, almost everything that involves public interests must be surnamed the party. Now this issue There is no need to discuss it.”
He said that this means that the training data in these large-scale generative AI systems in China basically come from the data that the CCP has filtered, and any speech or information that is not conducive to the CCP will not be blocked. Put it into the AI training database.
Caught in a Dilemma
Chen Zhijie said that such AI censorship may cause China’s AI development to fall into a dilemma, because the more powerful the AI, the more data must be provided during its training process, otherwise if the AI training If it is limited, the analysis results and decisions it produces may have some biases.
“But what China hopes to produce is an AI that can be used by the CCP,” Chen Zhijie said, so it is difficult to train a chatbot that can truly compete with Western countries.
He said: “The more you need comprehensive and unfiltered training data, the more likely it is to produce an AI that is difficult for the CCP to control; any AI that the CCP can control is not a very powerful AI.” Therefore, Chen Zhijie believes that the CCP may
eventually Two types of AI will be produced. One is the AI specially used by the CCP’s top leaders, the government or the People’s Liberation Army. The training content of this AI system will be huge; the other is the AI for people’s livelihood or commercial use. AI will have a lot of censorship and restrictions.
Li Hengqing, an economist at the American Institute of Information and Strategy, said in an interview with Voice of America that after ChatGTP was born, many major AI technology companies in China have also vigorously launched various similar products for testing, and there are many unsatisfactory results. Because netizens may ask questions such as “Will China use force against Taiwan”, “Will the CCP fall after it uses force against Taiwan”, “In what year will the CCP fall”, “When will the Chinese economy collapse”, and so on. The answer given by AI is not what the CCP likes.
Risk to the Communist Party?
Li Hengqing said: “It (AI) is likely to subvert the results of decades of brainwashing by the Communist Party of China, so it is very important to launch this AI-generated content management now. It should be said that it is indeed Chinese. It must ensure that There are no problems, but it’s hard to say whether it can be done.”
Li Hengqing said that when ChatGTP was first launched, he asked AI: “How do you evaluate China’s economy?” At that time, the answer given by AI was: “China’s economy has many problems. , weak demand, high debt, and expected limited economic development.” But if you ask AI the same question now, you will find that the answer has changed, which shows that both AI and the Chinese government are adapting to how to deal with these sensitive issues.
However, some analysts believe that the development of artificial intelligence in China will not overthrow the Communist regime, but may instead help its authoritarian rule.
In an interview with Nikkei, Microsoft President Brad Smith said that Chinese research institutions and companies will become the main competitors of ChatGPT. At present, there are three institutions at the absolute forefront, one is OpenAI and Microsoft, the second is Google (Google), and the third is Beijing Zhiyuan Artificial Intelligence Research Institute.
Smith warns: “We should absolutely assume, and even expect, that some nation-states will use artificial intelligence to launch cyberattacks that may even be more powerful than the cyberattacks and cyber influences we are seeing now.” Be an authoritarian accomplice
Lin Zongnan, a professor of electrical engineering at National Taiwan University who specializes in machine learning and deep learning, said in an interview with Voice of America that artificial intelligence technology has a good feature for the Communist Party, that is, it is a centralized technology. It means that AI needs to be calculated on a computing platform, and this computing platform must be produced in the computer room of a large company, so the CCP can directly control the company’s computer room. Compared with when there was no artificial intelligence in the past, the CCP had to rely on a large amount of manpower, and it was not necessarily able to accurately control every individual. Not only is the cost lower, but it can also conduct large-scale surveillance of the domestic people more efficiently. . This can be known from the fact that in the Hong Kong “anti-extradition” incident, the CCP used artificial intelligence image recognition as evidence for subsequent prosecution.
He said that although China’s artificial intelligence can be applied to various livelihood industries, there are two more important target areas. One is to develop technology for large-scale surveillance of domestic people. The technology is exported to other countries in the world.
Lin Zongnan said: “Even through these artificial intelligence technologies, China can more efficiently create false information, so if he uses this artificial intelligence, the potential damage (harm) and risks he causes to the free world, In fact, it is very huge.”
AI combined with military technology
Lin Zongnan said that the second main target area for China’s development of AI is military technology. One of the virtues of hegemony.
The Diplomat magazine reported on April 22 that China expects to spend $14.7 billion on artificial intelligence this year, accounting for about 10 percent of global investment. By 2026, this figure is estimated to reach approximately $26 billion. The report also said that the combination of artificial intelligence and military technology can add key advantages to China, with two examples in particular, one in space and the other underwater.
Artificial intelligence is the key technology and industry of “Made in China 2025” and innovation strategy. A research report by Harvard University’s Belfer Center for Science and International Affairs pointed out that China will soon become a global leader in many basic technology fields in this century, including artificial intelligence AI, and even In applications such as face recognition, voice recognition and financial technology, China is now ahead of the United States.
The National Security Council of the United States has also warned that China aims to surpass the United States by the end of 2030 and become the global leader in the field of artificial intelligence, and by the end of 2025, Chinese students who have obtained doctoral degrees in science, technology, engineering and mathematics The number is expected to reach twice that of the United States. Meanwhile, the number of U.S. students earning PhDs in AI-related industries each year remains about the same as it was in 1990.
However, some analysts pointed out that China’s regulations on generative AI content may not come at the right time, because China and the United States are engaged in a fierce technological war. The US ban and sanctions on semiconductors and advanced AI chips may weaken the development of artificial intelligence in China.
U.S. sanctions have a big impact
However, Lin Zongnan said that the key to the development of AI technology lies in the capabilities of hardware computing platforms, deep learning algorithms, and feeding data. Chatbots in the United States will also filter some words about racial discrimination or hatred, but China filters political thoughts and historical knowledge, so filtering content will not affect AI technology. The real impact is still the United States’ embargo on China’s high-end AI chips, but this will not be an immediate heavy blow, but will gradually restrain the development of China’s AI.
He said, but the algorithm will not be affected by the sanctions of the United States, and the part of the data is also because China has relatively no intellectual property rights and regulations on the use of personal information, so they can freely obtain data in the country and the free world, so the latter two are both It will not be detrimental to the development of China’s AI.
Li Hengqing, an economist at the Institute of Information and Strategic Studies of the United States, said: “But in the long run, it is possible to force out this kind of original technology of China’s self-reliance. It is not impossible.” He also said that China may also use white gloves or Stealing intellectual property to obtain high-end AI chip technology, “this is what China has been doing in the past, just like atomic bombs and launch vehicles.”
Chen Zhijie, an associate professor at the Institute of Political Science at Sun Yat-sen University in Taiwan, said that China’s chip industry currently does not have the ability to produce high-end chips on an economic scale, but in the long run, China will sooner or later produce at least military or dual-use chips. He expects this is the direction China will actively strive for in the next 5 to 10 years.
As for commercial AI chips, Chen Zhijie said that since China does not yet have the economic scale, even if it is able to produce them, the price will be very expensive, so in the end, only the Chinese Communist Party’s People’s Liberation Army or government agencies may be able to use them regardless of cost; However, for commercial AI chips with truly competitive prices, it may still be difficult for China to compare and compete with Taiwan’s TSMC or South Korea’s Samsung.