Tsinghua University China Enrolls First Virtual Student

China’s first virtual student, announced that she has enrolled at Tsinghua University and will be studying at the Department of Computer Science and Technology.

China’s first virtual student, announced that she has enrolled at Tsinghua University and will be studying at the Department of Computer Science and Technology. This virtual student is an Artificial Intelligence (AI) powered student avatar developed jointly by researchers from China’s industry and academia. The researchers from China’s Tsinghua University, non-profit research organisation Beijing Academy of Artificial Intelligence (BAAI), and private companies Zhipu Huazhang Technology (Zhipu AI) and Xiaoice worked together to develop the systems required to achieve this feat. By creating Hua, China has achieved the capability of independently developing a complex AI based on an open-source platform.

Hua is a culmination of China’s thriving AI ecosystem, which is championed by both the government and private sector in China. In 2017, Beijing released its ‘New Generation Artificial Intelligence Development Plan’ (NGAIDP) (新一代人工智能发展规划), which consolidated China’s AI development strategy. AI was already part of some other ambitious projects of Beijing, like ‘Made in China 2021’ and ‘Internet+.’ However, NGAIDP is China’s first national-level legislative effort focusing explicitly on AI. The importance of AI is also reflected in the Chinese government’s 14th Five Year Plan (2021-2025), which aims to achieve several major scientific and technological breakthroughs, including AI. The NGAIDP has outlined three sets of goals in its strategy to become a world leader in AI by 2030. The first geopolitical goal of Beijing is to optimise AI development in China and maintain competitiveness with major AI powers in the world by 2020. The second goal is to achieve a major breakthrough in AI theory and become a world leader in some AI applications by 2025. The third goal is to establish China as the world’s innovation centre for AI by 2030. By creating Hua Zhibing, Chinese scientists have accomplished a benchmark of the first set of strategies in NGAIDP.

The significance of achieving AI superpower status lies not only in its glory, but it is also a part of China’s ‘Shāshǒujiàn’ (杀手锏) policy, which roughly translates as ‘trump card.’ Despite China’s attempts to modernise its military and increasing its spending, it is still behind the United States (US). However, developing state-of-the-art AI-enabled weapons can significantly reduce that gap. ‘Shāshǒujiàn’ (杀手锏) refers to Chinese strategy to create a set of ‘silver bullet’ technologies that can reduce this gap in the US and Chinese military power. However, China’s AI capabilities haven’t reached the level of sophistication to entirely trump the US’ military advantage over China. Even though Hua is based on an advanced AI model, she is merely an indicator of China’s capabilities; she is not meant for military use with Virtual Student.

Practically, for state-of-the-art AI-enabled weapons, multiple types of AI systems need to reach a highly advanced real-time processing capability and work in tandem with advanced robotics. For such systems, mainly three types of AI are very important. The first type of AI is called Natural Language Processing (NLP). NLP involves processing text so that machines can understand human language. NLP is now commonly used for translation, spell check, sentiment analysis, summary extraction, etc. More advanced NLP is used to develop human-like interaction in chatbots. The performance of such AI depends on the number of ‘parameters.’ Simply put, the more the number of parameters, the better the performance. Hua Zhibing is based on the WuDao2.0 system with 1.75 trillion parameters, more than Google’s Switch Transformer AI language model, which has 1.6 trillion parameters. This makes WuDao2.0 the most advanced NLP system in the world as of mid 2021.

The second type of AI is used for generating content using processed data, also called Programmatically Generated Content (PGC). This type of AI is used to generate realistic representations from the data processed in real-time. It can be used to generate virtual environments, deepfakes, artwork, music, etc. The video released during the launch of Hua Zhibing uses music composed by her. Zhipu AI has given Hua a dual-drive engine of data and knowledge, giving her cognitive and reasoning abilities. Xiaoice, a Microsoft-owned company in Asia, gave her appearance, voice, and ability to compose the music used in the introductory video. Hence, Hua is more than just a virtual avatar of a human. She demonstrates virtual imaging technology paired with the world’s most advanced NLP.

The third type of AI is computer vision, which enables machines to understand meaningful information from visual inputs like images and videos. This information is then used by machines to make a decision or to process further information. The publicly available information about Hua does not directly refer to such capabilities. But as a student persona, she will be expected to have such abilities. Besides, computer vision is the largest segment of AI in China, and numerous Chinese programmes already use computer vision. For example, the SkyNet system of China uses facial recognition technology for ‘public safety’ with an accuracy of 99.8 percent.

Private players are a crucial part of China’s vision to achieve global dominance in AI as Virtual Student. The Ministry of Science and Technology (MOST) has declared several private companies as ‘National AI Team’ (人工智能国家队) to nudge the AI ecosystem in China. The NGAIDP is like a comprehensive guideline, shepherding China’s AI sector towards achieving objectives set by the Communist Party of China (CPC). Moreover, the private players have wiggle room to pick projects as per their expertise. Hence, companies selected as national champions are working on different projects. For example, Baidu is focused on autonomous driving, Alibaba is working on a smart city project to improve urban life, and Tencent is targeting computer vision for medical diagnosis.

Apart from encouraging existing tech giants, Chinese industry-academia collaboration is also promising. Zhipu AI, one of the partners who developed Hua Zhibing, was born from Tsinghua University’s AI research programme. Such spinoff is essential to convert university research to marketable products and further stimulate research by attracting investment. Furthermore, as per the 2021 Stanford AI Index report, China ranks third in the ‘number of academic-corporate peer-reviewed AI publications.’

These developments don’t mean that China is shipping out superior AI-powered weaponry. But it does indicate a thriving ecosystem of an important dual-use technology. However, much of the focus of China’s AI strategy is still on domestic and industrial applications, at least for now. For example, most applications in the list of 17 focus areas released by the Ministry of Industry and Information Technology (MIIT) are consumer and industrial applications — like developing smart sensors, open-source AI platforms, smart consumer products, medical imaging, industrial equipment, etc.

One of the 17 focus areas outlined by MIIT was to develop open-source platforms supporting AI. Chinese researchers have succeeded in developing an advanced open-source ‘expert system’ for AI. These expert systems give decision-making ability to the machines. Different models of such systems have been developed in attempts to emulate human-like decision-making abilities. A Mixture-of-Expert (MoE) system uses multiple expert systems and needs specially designed hardware. To create WuDao2.0, BAAI scientists first developed “FastMOE,” a distributed Mixture of Experts (MoE) training system similar to Google’s MoE. But unlike Google, FastMoE is open source and does not require Google’s proprietary hardware.

Hua is an important milestone for the AI community worldwide, and it will be fascinating to track her progress. Since the persona of Hua Zhibing is a student, she will continuously learn and evolve. Her journey will reflect the progress of some of the best minds in China. Beijing envisions AI as one of the drivers of its economic and social transformation, therefore, AI will continue to feature heavily in its strategic and industrial development priorities.

Source Orfonline

Arsalan Ahmad

Arsalan Ahmad is a Research Engineer working on 2-D Materials, graduated from the Institute of Advanced Materials, Bahaudin Zakariya University Multan, Pakistan. LinkedIn: https://www.linkedin.com/in/arsalanahmad-materialsresearchengr/

Leave a Reply