Commentary: How China Can Stay Ahead of The Game With AI Integration

Research comparing AI strategies across countries shows a consensus on AI's vast potential and the associated risks. A unique feature of China's strategy is its focus on integrating AI with the real economy, particularly in traditional manufacturing, where companies like Huawei and SAIC are merging into the AI network [para. 1][para. 4]. Since 2017, China has tracked over 4,000 AI companies to map a comprehensive value network of its AI sector, revealing strong connectivity among AI companies, universities, government bodies, and industrial parks. These connections are more concentrated and closer compared to other industries [para. 2][para. 3]. This connectivity underscores China's significant integration of AI within its economic infrastructure [para. 3][para. 4].

From 2015 to 2020, AI applications predominantly utilized judgment-based structures, which relied on tens of thousands of parameters. Despite some tech companies exploring larger models, the extensive data and associated costs deterred widespread adoption [para. 5]. With the emergence of models like GPT-3, equipped with hundreds of billions of parameters, a new era of generative applications began. By 2021, China had developed several domestic models with similarly large-scale parameters [para. 6]. The role of AI expanded further with OpenAI's ChatGPT gaining mainstream traction by late 2022 [para. 7]. China's large AI models are achieving international prominence and advancements in energy-efficient training methodologies. However, the scalability of AI models remains challenging, as improvements in performance become less dependent on model size [para. 8][para. 9].

China must nurture original and innovative talent, shifting focus from mere computing power to algorithm creativity and data applications. Significant reforms in the educational system and innovation environment are critical for achieving this [para. 10]. Historically, China's tech strategy has centered on catching up with global advancements. For AI, however, the country needs to transition from following existing methodologies to pioneering new frontiers, particularly as it seeks to address challenges like hardware deficits exacerbated by U.S. restrictions. Despite technological clarity, catching up in semiconductor manufacturing remains complex [para. 12][para. 13]. Moreover, as Moore's Law nears its limits, China faces challenges and opportunities in leading the post-Moore era through original innovations, like 3D integration and new substrate materials [para. 13].

A more urgent issue is the absence of foundational algorithms developed within China, given that many mainstream AI frameworks originated elsewhere. While China has made improvements and gained AI patents, there still lies a need to develop foundational inventions [para. 14]. Future AI development should blend historical scientific knowledge with real-world data to enhance model efficiency and support groundbreaking advancements [para. 15].

In terms of energy efficiency, AI development faces challenges, with large models consuming significant power compared to the human brain's efficiency. Training and inference processes of substantial AI models demand vast energy resources, posing a challenge for sustainability [para. 16][para. 17][para. 18]. AI's path towards personalization and trustworthiness in applications, such as Qualcomm's development of a new chipset, indicates progress toward running complex models on mobile devices [para. 19]. As AI leads to more natural human-computer interfaces, personalized intelligent assistants are poised to boost quality of life and efficiency, paving the way for expansive applications [para. 20].

AI's most commercial applications currently lie within the automotive industry and embodied intelligence, with autonomous driving significantly altering industry landscapes. However, competition remains fierce in the evolving electric vehicle market [para. 21][para. 22]. The integration of intelligent technology into the automotive sector could radically transform industry chains, impacting auxiliary services, and leading to potential market consolidation [para. 23]. As for embodied intelligence, commercial applications such as caregiver robots must prioritize actual needs over humanoid designs [para. 24][para. 25].

In manufacturing, AI general models can enhance processes by increasing quality, reducing costs, and improving energy efficiencies, but their integration is not yet complete. Overcoming challenges like merging general models with specialized technologies and attaining model accuracy are crucial for full incorporation [para. 26][para. 27][para. 28]. Successful examples, like Jiangnan Shipyard, illustrate potential gains from intelligent deployments, emphasizing the importance of comprehensive, real-time industrial data to facilitate AI's role in manufacturing [para. 29][para. 30].