Discover more from ChinaTalk
AI Proposals at ‘Two Sessions’: AGI as ‘Two Bombs, One Satellite’?
Resource distribution, cybersecurity risks, legal reform, and a Chinese AGI
The “Two Sessions,” during which the National People’s Congress (NPC) and Chinese People’s Political Consultative Conference (CPPCC) convene, is the most important annual event on China’s political calendar. Delegates descended upon Beijing on March 4, bringing with them a wide range of policy proposals. Here’s a roundup of what has been discussed on the AI front so far, including:
How China can build its own Microsoft + OpenAI coalition;
Why high-quality data is a bottleneck for Chinese AI;
How to encourage nationwide collaboration;
And why a PKU computer scientist believes the CCP should go all-in on making AGI.
ChinaTalk is a reader-supported publication. To receive new posts and support our work, consider becoming a free or paid subscriber.
Professor Zhu Song-Chun is a leading Peking University computer scientist and the director of the Beijing Institute for General Artificial Intelligence. At the CPPCC, he’s cheering for an AGI moonshot to secure China’s global leadership:
“Artificial general intelligence (AGI) is the strategic high ground of international scientific and technological competition in the next ten to twenty years, and its influence is equivalent to the ‘atomic bomb’ in the field of information technology.”
On March 4, the opening ceremony of the 14th National Committee of the Chinese People’s Political Consultative Conference (CPPCC) was held. Zhu Songchun, a member of the CPPCC and director of the Beijing Institute for General Artificial Intelligence, suggested in a proposal that China should elevate the development of AGI to the level of the “Two Bombs, One Satellite” of the contemporary era and seize the global strategic high ground of technology and industrial development. [Jordan: AI racing dynamics will be with us so long as the US and China are locked in strategic competition.]
“Six months ago, many experts in China believed that AGI was an unreachable goal, but the emergence of ChatGPT recently made the public feel that AGI seems to be within reach.” In Zhu Songchun’s view, AGI will have a disruptive impact in many fields, such as military, industry, social governance, and cognitive confrontation, and is a new strategic field in which countries are vying for layout.
“If China can be the first to achieve a truly generally intelligent entity, it will become the ‘ace in the hole’ of international scientific and technological competition.”
Zhu Songchun said, therefore, that China should accelerate the strategic layout for AGI.
According to public reports, as of 2022 Microsoft has invested more than $3 billion in OpenAI and will continue to invest $10 billion in the next few years. And according to a credible public document, the United States has invested more than $2 billion in university laboratories in the field of AGI in recent years.
“The gap between China and the United States in the field of artificial intelligence may continue to widen in the future, and I hope to attract the attention of the country.” Zhu Songchun said that we need to plan and establish domestically controlled AGI research and development technology clusters, industry clusters, and fund clusters that cover the whole chain from the national level.
Zhou Hongyi is the founder of 360, a leading Chinese cybersecurity software firm. At the CPPCC, he’s advocating for an AI investment model copied from the Americans:
Recently, ChatGPT has led to global discussions, and Zhou Hongyi believes that the tremendous leap of artificial intelligence large-scale model technology represented by ChatGPT will usher in a new industrial revolution, and China should catch up with it. While focusing on technological innovation, it is also necessary to pay attention to the innovative models behind the technological breakthroughs. In his proposal, he suggested starting with two points to develop large-scale artificial intelligence model technology. Firstly, establishing a collaborative innovation model between large-scale technology companies and key scientific research institutions to build China’s “Microsoft + OpenAI” combination and lead the research of large-scale model technology. Secondly, supporting the establishment of multiple national-level open-source projects for large-scale artificial intelligence models to create an open innovation ecosystem for open-source crowdsourcing.
iFlytek is a partially state-owned technology firm that makes voice recognition and educational software (and surveillance tech that may have been used in Xinjiang). Its president, Liu Qingfeng, is an NPC delegate, and he wants China to build up its computing power:
Although many domestic companies have released related achievements, there is still a certain gap compared to ChatGPT in terms of intelligence. Liu Qingfeng suggested that the country should attach importance to the research and development of cognitive intelligence large models, form innovative systems around leading companies for industry-university-research cooperation, accelerate to catch up with international cutting-edge technologies, support examples of industry applications of cognitive intelligence large model technology, and promote the value of cognitive large models in education, medical care, office, human-machine interaction, and AI-generated content fields.
“China’s cognitive intelligence large models can direct its own future only if it is developed from domestic technology platforms.”
Liu Qingfeng believes that the country should further increase support for domestic AI software and hardware technology platforms, allowing large models to be built and operated on domestically produced storage, computing power, operating systems, and other basic platforms.
Specifically, Liu Qingfeng suggested that the country should establish a public computing power platform for cognitive intelligence large models, set up a mechanism for using the platform to lead the way, and allow more scientific research institutions and technology start-ups to have the opportunity to access the national public computing power platform for model training and algorithm innovation.
He proposed a national data resource platform which gathers basic data required for cognitive intelligence large models and builds a data co-construction and sharing mechanisms on the basis of law and compliance. The platform would support strategic scientific and technological forces to accelerate the research and industrialization of cognitive large models from the basis of a national data resource platform.
In addition, Liu Qingfeng believes that the country should also encourage industrial funds to actively explore equity investment agreement models that are more conducive to entrepreneurial teams and core technical talent’s long-term aims, based on investment agreement models such as that between OpenAI and Microsoft. This would build a better scientific and technological venture capital ecosystem and innovation and entrepreneurship environment.
Professor Kou Gang, a CPPCC member, teaches at the Big Data Research Institute of Southwest University of Finance and Economics. He believes China must overcome challenges in computing, data management, and market revitalization to close the AI gap:
“In the field of large-scale artificial intelligence models, the underlying architecture software and hardware are almost monopolized by foreign companies. At the same time, high-quality datasets have become one of the bottlenecks restricting the development of China’s artificial intelligence industry, while many foreign datasets restrict users with Chinese IPs from full access or do not provide services to Chinese users at all.”
In Kou Gang’s view, in the race for artificial intelligence, what matters most is technology, and China’s independent research and development capabilities urgently need to be improved.
Artificial intelligence technology poses new demands and challenges for computing power. Kou believes that China’s ability to allocate and coordinate computing resources needs to be strengthened. According to his analysis, the development of universal cornerstone models for artificial intelligence requires strong computing power support. Although China’s computing infrastructure is large in scale, there are still peaks and valleys in its use in different regions, and dynamic allocation across provinces cannot be achieved. AI computing centers in different regions are relatively independent and have not formed joint scientific research blocs, which inhibit application innovation and resource complementarity across regions.
Another difficult predicament facing computing power is that the front-end systems used by various supercomputing centers in China are different, and the rules are also different. Specialized and high-quality services are still lacking. At the user end, many small- and medium-sized enterprises and research units are restricted by high costs and face insufficient computing power.
In addition to technology and computing power, a more complete industrial ecology is needed. Kou points out some existing problems. Currently, the development of large-scale artificial intelligence models in China has become a competition among a select few institutions. In this process, there is more imitation and competition among them than there is cooperation, “which is a waste of resources for the exploration of basic research.”
Potential legal risks should not be ignored either. “As the development of large-scale models relies on massive database information and cannot verify facts or data sources, there may be two major hidden dangers: personal data and commercial secrets leakage, and providing false information. For some text, video, code, etc. protected by copyright, if they are obtained, modified, and patched together without the authorization of the rights holder, it may involve new copyright infringement,” Kou said.
Regarding the shortcomings in China’s technological research and development, Kou suggests increasing support for independent research and development in the field of artificial intelligence. Specifically, the Ministry of Industry and Information Technology and the Ministry of Science and Technology would jointly introduce policies to incentivize and guide the key technology breakthroughs and application-ecosystem development in large-scale artificial intelligence models. They would implement emergency major science and technology special projects to accelerate technological breakthroughs, system development, and application promotion. They would also release special funds through the Ministry of Science and Technology and the National Natural Science Foundation to support the construction of high-quality artificial intelligence datasets; strictly control labeling standards, quality, and update frequency; and build platforms and ecosystems to establish mechanisms for joint innovation between universities, research institutions, and enterprises — thus promoting the transformation of technological achievements.
Regarding the problem of computing power, Kou suggests achieving national integration of computing power network scheduling. For example, on the usage end, exploring the establishment of reasonable pay-as-you-go computing power sharing mechanisms for small- and medium-sized enterprises and research institutions to jointly promote the development of technology and save resources.
Currently, China has laid out many computing power platforms. Kou suggests connecting the supercomputers, intelligent computing, and artificial intelligence computing center nodes distributed across the country through dedicated lines to form a unified national perceptual, distributed, and scheduled artificial intelligence computing power network. On this basis, he proposes the implementation of elastic resource allocation. He also suggests unifying the front-end systems of supercomputers, breaking down barriers between different systems and applications, and realizing one-click login and data exchange between new and old systems to enhance customer experience.
China’s artificial intelligence industry ecosystem is forming. However, how should we promote innovation in basic models and construct an ecosystem for artificial intelligence? This is the problem of going from “1 to 100,” and cultivating market players will be the first step.
Kou suggests incubating a batch of truly market-oriented entrepreneurial companies to break industry monopolies. Additionally, under the premise of national institutional regulation and macro control, government data can be appropriately opened to whitelisted companies, institutions, and universities, exploring the application and service of large-scale models in urban governance and empowering industries through data banks, data trusts, and other models.
How to address potential legal and regulatory risks brought about by artificial intelligence? In Kou’s view, it is necessary to accelerate the revision of relevant Internet laws and regulations and the development of ethical norms, clarify legal bottom lines and red lines in the universal basic model and generative AI technology and applications, accelerate research on intellectual property protection brought by universal basic models of artificial intelligence, and promote the construction of AI content monitoring platforms.
Qi Xiangdong is the president of Qi An Xin, a cybersecurity firm that spun out of 360. He foresees ChatGPT-powered hackers:
“ChatGPT has a very strong learning ability and can quickly learn professional knowledge in various fields, even reaching the level of human experts,” said Qi Xiangdong.
“This also means that AI can elevate the skill levels of hackers, and even ordinary people who don't understand code can become hackers.”
How should we deal with this huge challenge? Qi Xiangdong believes that the combination of “human + machine” is better than relying on the idea of “defeating AI with AI”: “The essence of network security is the competition between people on both sides of the attack and defense. In every competition, people play a leading and driving role.” He said that through training ChatGPT-related technology on massive amounts of security knowledge and data, we can train security models, develop security products, and apply them to threat detection, vulnerability mining, and other aspects. At the same time, we should strengthen the cultivation of network security talents, and relevant companies should make proactive layouts.
Lian Yuming is the founding president of the Beijing International Institute for Urban Development, a legal expert, and a CPPCC member. He’s not that worried about AI taking over:
CPPCC member Lian Yuming proposed during this year’s Two Sessions to strengthen judicial protection of data rights. Lian Yuming believes that behind phenomena like ChatGPT and others is a transformation in data rights, and the transformation signifies that people will face a series of legal and ethical issues that urgently require more specific and actionable legal provisions or guiding cases to strengthen the judicial protection of data rights. Lian Yuming stated that individuals should actively embrace the transformation and need not worry about artificial intelligence replacing humans. He believes that humans will have greater inventions to balance with ChatGPT. The phenomenon of legislation lagging behind technological progress is universal, and the urgent task is to accelerate research on data rights and the legislative process.
ChinaTalk is a reader-supported publication. To receive new posts and support our work, consider becoming a free or paid subscriber.