From tech giants racing to develop the next groundbreaking AI tool to government regulators issuing the world’s earliest and most detailed regulations on generative AI models, 2023 marked a pivotal year in China for AI. 

Fueled by the rise of technologies like ChatGPT, major players, including Baidu, Alibaba, and Tencent, not only unveiled new AI technologies, but also began integrating large language models into productivity services such as information summaries, text translations, and office task streamlining. 

And amid a slew of new applications, such as advanced camera apps and content creation tools, which offer unprecedented convenience but also raise privacy concerns, Chinese regulators stepped up efforts to rein in the technology. 

Here are some of the most important AI developments in 2023:

Big tech

The momentum began to build in March, just months after ChatGPT’s global surge in popularity.

Baidu, China’s leading search platform, unveiled its first language model, ERNIE Bot, that month. It boasted five key capabilities, including advanced mathematical reasoning and a better understanding of the Chinese language. Baidu’s training data included an array of web, search, image, voice, and knowledge graph data.

By the end of August, ERNIE Bot had undergone significant upgrades and was opened to the public. In October, Baidu’s CEO Robin Li introduced the latest version of ERNIE Bot, confidently stating that its general capabilities were “by no means inferior compared to GPT-4.”

In parallel, Baidu was also engaged in generating AI content, leveraging large language models combined with cross-modal learning. A prominent example of these efforts is Wenxin Yige, an AI platform that facilitates the creation of artwork.

In April, Alibaba introduced its AI chatbot, Tongyi Qianwen, which was soon upgraded in October to enhance its ability to understand complex commands and improve collaborative functions. Beyond just developing large language models, Alibaba also focused on practical applications, especially in interpreting multimedia content and streamlining e-commerce operations.

The company also developed AI-powered assistants like Tongyi Tingwu and Tongyi Wangxiang, which specialize in processing audio, video, and image content. And in September, Alibaba launched Taobao Wenwen, the first AI-powered e-commerce assistant. Earlier this month, the company unveiled two large models capable of processing languages from Southeast Asia, including Vietnamese, Indonesian, Thai, and Malay. 

Tencent, another major player in China’s tech sector, launched its large AI model, Tencent Hunyuan, in September, with an update in October. This model drew attention for its broad application in AI-generated imagery, particularly within the advertising industry.

In contrast to its peers, ByteDance, the owner of TikTok, maintained a more subdued presence in the AI arena. In April, its enterprise software, Lark Suite, introduced an AI assistant named My AI. The same month, ByteDance’s cloud service, Volcano Engine, debuted a cloud platform for training large-scale models. However, ByteDance’s own large-scale AI model remains under wraps.

In the public sector, Fudan University, located in Shanghai, launched the Computing for the Future at Fudan in June, the largest cloud-based research computing platform in Chinese universities. Developed in partnership with Alibaba Cloud, this platform provides immense computing power, rivaling that of major tech firms. 

Concerns, government regulation 

As AI applications rapidly transform daily life, they have also drawn public ire over privacy and copyright concerns. Recognizing these issues, Chinese regulators have stepped up efforts, issuing detailed regulations on AI models to rein in technology and protect individual rights.

For instance, Miaoya, an AI photo app, gained popularity for its user-friendly interface and high-quality portraits. But it soon faced backlash over its data usage policy, which allowed the use of users’ data in any form of media or technology, both now and in the future. This policy was eventually revised following significant controversy. 

Similarly, the AI tool of the Instagram-like lifestyle app Xiaohongshu, Trik AI, came under fire for allegedly using users’ artwork to train its AI engine, sparking widespread debate about the ownership rights of AI-generated artwork.

In April, China was among the first countries to issue detailed regulations on AI, with the Cyberspace Administration of China introducing draft rules on AI-generated content. By July, these rules were revised to be more accommodating for AI technology providers, emphasizing not only the responsibility of users, but also shifting the onus on providers to ensure accurate AI-generated content. 

Officially implemented on Aug. 15, these rules marked the beginning of formal registration for large-scale AI models. The same month, eight entities, including Baidu, ByteDance, SenseTime, and Baichuan Intelligence, registered their products and officially recognized AI solutions.

In December, the results of the first “Large Model Compliance Test” were revealed, with models from Baidu, Tencent, 360 Group, and Ali Cloud meeting national standards in universality and intelligence. Additionally, a special committee on governing AI security, established in October, released the first batch of Chinese basic corpus for large models in December, further propelling AI development.

Beyond national standardization, platforms such as Douyin and Bilibili mandated AI-generated content labels to combat online misinformation and inappropriate content. 

Future

Generative AI is poised to become a critical driver in 2024, with its impact on the economy and labor market anticipated to be significant. A McKinsey & Company report projects that AI could contribute up to $25.6 trillion to the global economy, with generative AI alone accounting for $7.9 trillion. 

In China, the AI-driven industrial shift will also necessitate skill upgrades or retraining for an estimated 220 million workers by 2030.

A key area of innovation lies in multimodal generative AI. By 2024, as these technologies advance, AI models are expected to encounter more complex and varied interactive scenarios, spanning from smart home applications to medical diagnostics and autonomous driving. 

Furthermore, with advancements in terminal computing power and the integration of AI into applications, AI’s presence in everyday devices is also set to increase dramatically. According to an International Data Corporation report, by 2024, AI terminals are projected to make up 55% of the Chinese market, and over 70% of terminal devices are expected to be AI-enabled.

Editor: Apurva. 

(Header image and icons: dgh5528 and Arafat Uddin/IC)