Cover Story Part Three | China's Path to Artificial Intelligence Development (AI Translation)

封面报道之三｜人工智能发展的中国路径

The "New Generation Artificial Intelligence Development Plan" issued by China's State Council in 2017 set milestones for 2020, 2025, and 2030 to synchronize AI development with economic and social advancement. China's strategic plan distinguishes itself by promoting AI integration with the real economy, a direction emphasized by the Communist Party's National Congress reports. Research efforts focus on fieldwork in enterprises and big data analysis to support this agenda [para. 1][para. 3].

Observations from 2017 onwards show substantial growth in China's AI industry, with sample companies increasing from over 400 to more than 4,000. This network reveals that Chinese AI enterprises, universities, government bodies, investors, and industrial parks are intricately linked, creating a highly interconnected value network that exceeds the connectivity seen in other sectors [para. 3][para. 5]. AI's integration into the manufacturing sector is marked by companies like Huawei, SAIC, Changan, and Geely becoming key entities within the AI value network [para. 5][para. 7].

By 2020, despite pandemic-related challenges, assessments showed China's AI development was on track, with significant progress in research projects and the establishment of fifteen open innovation platforms [para. 9][para. 11]. AI applications transitioned from the "judgmental" phase to "large models" typified by GPT-3, marking a shift towards models with tens of billions of parameters and generative capabilities [para. 11][para. 13].

China has accelerated its development of large-scale models, exemplified by Tsinghua University's GLM model reaching parameter scales akin to international standards. Cross-modal applications, like digital humans at the 2022 Winter Olympics, showcase these advancements. The launch of OpenAI's ChatGPT, a general-purpose dialogue model, further popularized AI, drawing hundreds of millions users due to its language mimicry capabilities [para. 13][para. 15].

Large-scale AI models present untapped potential, facing challenges like the growing energy consumption during scaling and training. Optimizations in energy efficiency and training are ongoing within China, which is making strides toward international performance benchmarks [para. 17][para. 19]. Despite the prospects of scaling AI models, natural limitations akin to human brain development suggest that advances in intelligence won't solely depend on increasing model size [para. 21][para. 23].

China's longstanding industrial strategy has been to close technology gaps. In AI, it seeks to pioneer new paths in computing, algorithms, and data. Challenges remain in resolving hardware gaps, like chips, and fostering innovation beyond current mainstream AI frameworks. Explorations into new algorithms and technologies, such as pulse neural networks and privacy-enhancing technologies, are underway. These explore how statistical characteristics from data can generate new datasets, facilitating AI training while maintaining privacy [para. 23][para. 27][para. 29].

Zhang Bo from Tsinghua University highlighted AI's future direction as blending knowledge and data to enhance AI model efficiency. This involves integrating centuries of scientific understanding with real-world data [para. 31]. Reducing AI's energy consumption is another critical focus, with research into spiking neural networks showing promise for energy efficiency [para. 33][para. 35].

Regarding AI applications, there's potential in developing personalized, trusted AI assistants through terminal-side models, a challenge Qualcomm is addressing with its latest chipsets. Autonomous vehicles present the most immediate AI application prospects, driven by the need for safer, more efficient transportation. Embodied intelligence, primarily in mature settings like car factories, and humanoid robots in elder care, represent further avenues for AI product growth [para. 37][para. 41][para. 43].

In manufacturing, AI can generate value when it directly enhances quality, reduces costs, and improves efficiency. The deep integration of AI in manufacturing is crucial, exemplified by companies like Foxconn optimizing tool usage through intelligent detection, reducing defects and costs. However, challenges in integrating general models with specialized production technologies remain, requiring innovation beyond traditional AI model accuracy, moving towards greater interpretability [para. 47][para. 49][para. 51].

Overall, China's AI strategy is poised to harness scientific and technological breakthroughs to progress beyond existing frameworks, addressing energy efficiency, developing trusted AI applications, and achieving practical integration in manufacturing for sustained advancement [para. 53][para. 55].