In April, China’s Internet regulator released a draft to govern “Generative Artificial Intelligence” (AI) (Chinese: Generative Artificial Intelligence). The regulation aims to ensure “sound development and standardized application” of AI services capable of producing text, video, voice and images.
OpenAI’s ChatGPT is just one of many technologies that are rapidly emerging around the world. The US leads in this area, but the Chinese government has overtaken the US in creating the regulatory framework. Baidu’s ERNIE Bot, China’s answer to ChatGPT, just debuted and isn’t overwhelming. However, a closer look at the development and commercialization of these technologies in China shows that the government is actively trying to get ahead of this growth trend.
This is largely driven by the Chinese Communist Party (CCP)’s desire to maintain social and political stability by maintaining an internet censorship mechanism. But the ability to deploy broad restrictions without challenge also allows Beijing to act quickly. Aside from information management, China’s advanced approach to regulating the inputs and outputs of Large Language Models (LLMs) has given legislators elsewhere, including Europe, to consider when developing their own regulatory frameworks. It could provide an interesting angle that should.
Widely adopted for commercial use
Generative AI has experienced explosive growth over the last few years, demonstrating its potential to transform economies and societies, but it also poses significant governance challenges. ChatGPT and friends have demonstrated an amazing ability to comprehend and compose text. Image generators like Midjourney can generate original artwork based on your text prompts. Advances in virtual and augmented reality technologies to create immersive simulations are transforming the field from gaming to healthcare. In addition to these breakthroughs, generative AI is permeating social media in the form of filters that enhance photos and videos. These technologies raise various privacy, security, ethical, and socioeconomic concerns.
In China, ChatGPT was quickly banned by government censorship, and companies are scrambling to develop domestic alternatives. However, generative AI is already widely adopted for commercial use. A report from the Institute of Artificial Intelligence at Tsinghua University lists several of these, including a TV show showcasing its ability to generate speech and a program to write poetry. In one case, generative AI was used to store footage from previously shot films and replace the faces of actors who violated state moral guidelines. Xinhua, China’s state-owned news agency, regularly uses simulated “AI anchors” in some of its online news segments. In a white paper released in 2020, Tencent highlighted the beneficial potential of AI-generated content in fields ranging from e-commerce to healthcare.
Once widely adopted, controversy soon ensued. In 2019, his mobile app called ZAO, which allows users to swap their faces into short movie clips of his girlfriend, became an overnight sensation. It was ordered off the shelves almost as fast when regulators found it violated users’ privacy and data rights.
Generative AI, especially those involving machine learning-altered videos known as deepfakes, are capable of much more nefarious acts. In 2021, Chinese authorities discovered that criminals created AI-generated facial videos of her from scraped internet photos to illegally sign up for online payment accounts. China is not the only country facing such threats. In 2022, the mayors of several European cities were spoofed in video calls by someone pretending to be the mayor of Kiev using altered video footage.
Regulators everywhere struggle to stay ahead of technology
Stories of misuse of generative AI are similar around the world, but government reactions are different. ChatGPT caught the drafters of the European Union’s AI law off guard, showing that legislators are struggling with general-purpose AI systems that simultaneously enable harmless and highly harmful applications.
Beijing, on the other hand, took action early on. Following the ZAO incident, China’s Cyberspace Administration (CAC) has ordered online information service providers to review and clearly label AI-generated content. The rule also prohibits the use of generative AI to generate and spread fake news. Building on these previous provisions, in January of this year, China began enforcing a law regulating the use of “deep synthesis” (poso-synthesis).
China’s governance efforts are turning attention to the broader societal implications of generative models. In addition to labeling requirements, regulations on deep synthesis include disarming rumors and disinformation to service providers, submitting algorithms for review via a filing system operated by the CAC, The data used for training is legally obtained. The April Draft rules go even further, requiring all new products to undergo a security assessment by his CAC before launch.
Besides creating regulations, the government also backs startups and projects dedicated to deepfake detection, such as RealAI, a spin-off from Tsinghua University (the EU is backing similar efforts). Considering the security risks associated with the proliferation of AI-generated content, the Ministry of Industry and Information Technology’s top think tank released China’s first industry standard for evaluating generative AI products in March.
Beijing’s approach reflects a positive stance on the penetration of AI into society, but also its ability to respond quickly to regulations unthinkable in a country governed by the rule of law. Chinese authorities are likely to be open to large-scale deployment of language models in useful areas such as self-driving cars and healthcare, but they are also concerned about the technology’s potential to mobilize public opinion and endanger national security. Underpinning China’s regulatory efforts is also a desire to maintain social and political control. For example, the draft emphasizes that AI-generated content should “reflect core socialist values and not include content that subverts state power.”
CCP’s obsession with information control could lead to tougher rules
It remains to be seen how prosperous China’s AI will be, given the Chinese Communist Party’s obsession with information control. When it comes to the internet and digital technology, China’s success can often be attributed to regulators’ initial openness to experimentation. With LLM, the Chinese government may have to put up with the risk of chatbots returning politically sensitive information to users. Democracies may be slower to implement regulation, but they can ultimately create an environment that is more enabling the transformative potential of AI.
Nevertheless, lawmakers in other countries need to reflect on China’s regulatory efforts beyond the overly broad powers the CCP grants itself to restrict freedom of expression. The draft regulation places significant responsibilities on providers of generative AI services, including labeling generated content and protecting personal data. When collecting data to train AI models at scale, providers must ensure accuracy, avoid bias, and respect intellectual property rights. It is also responsible for the output of the model. You may not create content that discriminates based on a user’s race, gender, religion, or other characteristics.
Although the feasibility of enforcement is unknown, some of these requirements could be considered by policy makers as they consider how this powerful new technology should be regulated in a liberal democratic society. there is.
