According to recent reports, China's drive to outpace the U.S. in generative AI is encountering obstacles due to government-imposed censorship. While the U.S. imposes restrictions on China's access to high-end chips and hardware, the demands of China's authoritarian regime may prove more significant in determining the global AI race's outcome.
China's chief regulator, the Cyberspace Administration of China (CAC), is implementing rigorous reviews of AI models developed by major tech firms such as ByteDance, Alibaba, Moonshot, and 01.AI. The Financial Times reported that these companies must collect thousands of sensitive keywords and questions that violate "core socialist values," including phrases like "inciting the subversion of state power" and "undermining national unity."
According to the Wall Street Journal, the CAC requires companies to prepare between 20,000 and 70,000 questions to test the models' ability to produce safe answers, with about half of these questions relating to political ideology and criticism of the Communist Party. Additionally, AI models must be programmed to refuse responses to politically controversial queries, such as those about the Tiananmen Square uprising of 1989 or President Xi Jinping's policies. The AI systems must terminate the conversation if users persist with such questions.
China's AI firms, already behind in the generative AI game, are further slowed by the need to align with government censorship. Training AI models are delayed as data must be meticulously scrubbed of politically sensitive content. Despite these efforts, the inherent unpredictability of generative AI means there's always a risk of unintended outputs, making it difficult for the government to be entirely confident in these systems.
The resources spent building these ideological guardrails divert from enhancing AI speed and utility. China's previous AI developments, like image recognition and computer vision, have been used for government surveillance, particularly in regions like Xinjiang. The advent of ChatGPT prompted Beijing to push for parity with Western AI advancements, instituting detailed regulations a year ago to govern AI deployment. These rules protect citizens' privacy and data from corporate abuse but not from governmental control.
In contrast, the U.S. Biden administration's AI executive order mandates modest transparency for cutting-edge AI capabilities. American tech firms' efforts to manage AI misinformation and hate speech have faced criticism as censorship from conservative circles. Meanwhile, the EU is developing comprehensive regulations on data use and privacy for AI, which some U.S. companies argue could stifle AI development in Europe.
China's ongoing regulatory and censorship challenges highlight the complex landscape of AI development under authoritarian regimes. The stringent controls imposed by the Chinese government may hinder the country's ability to compete effectively in the global AI market.