DeepSeek: A Tool Tuned For Social Governance – Analysis
In the lead-up to the annual gathering of the People’s Republic of China’s (PRC) legislature in March, reporters from a state-run media outlet engaged citizens on the street about the relevance of the political meetings, known as the “Two Sessions” (两会), to their lives. Instead of asking questions directly, the journalists invited sources to direct their questions to DeepSeek R1, the country’s latest large language model (LLM).
One young woman asked, “I’m about to graduate, what kind of job opportunities can AI help to provide?” (我即将毕业，AI能帮助提供什么工作机会)—a timely inquiry as legislators highlighted artificial intelligence (AI) as a solution for future development. DeepSeek responded that there were “abundant employment opportunities” (广阔的职业发展空间) thanks to AI, listing multiple roles such as data annotators, and noting the high salaries associated with these positions. When the same question was posed by another individual, DeepSeek provided similar assurances, advising fresh graduates to combine their existing skills with AI to “grasp the career opportunities in the AI era” (把握AI时代的职业机遇). This response, however, neglected the social turbulence created by AI in a job market where youth unemployment remains high. For instance, over the past three years, machines have taken over 60 percent of all data annotation tasks in the PRC, pushing the role of data annotator closer to obsolescence. The People’s Daily Online article further emphasized DeepSeek as a “happiness code” (幸福密码)—a technology showcasing the Party-state’s efforts to address national concerns and reassure the populace of their security.
Preventing Risks, Resolving Disputes, and Writing ‘Correct’ Articles
In Chinese Communist Party (CCP) theory, “social governance” (社会治理) is the framework through which the government maintains social stability and resolves conflicts. The Party-state envisions AI playing a crucial role in future social governance.
During the 2024 Two Sessions, Premier Li Qiang introduced the “AI+ initiative” (“人工智能+” 行动), which included both an economic aspect—the “deep integration of digital technology and the economy” (促进数字技术和实体经济深度融合)—and a social one—“improving the modernization level of social governance” (提升社会治理现代化水平). Local governments are currently exploring how DeepSeek could be integrated into the social governance framework, including decision-making processes and state services to help resolve social conflicts and promote state policy preferences. While some of this may appear to be mere attention-seeking experimentation, the increasing reliance on AI models like DeepSeek positions it as a powerful state-backed source of “public opinion guidance.”
An effective PRC-built LLM theoretically provides a means for local governments to demonstrate compliance with the “AI+ initiative” while maintaining social stability more efficiently. For instance, on March 17, Liaoning Daily reported that the province had integrated DeepSeek into its local “12345” help hotline, claiming enhanced efficiency in dispatching complaints to the appropriate departments. Similarly, police services have presented DeepSeek as a tool for upholding public security, with a local police station in Nanchang stating that it “added a touch of warmth to the harmony and stability of the community” (为社区和谐稳定添上了温暖的一笔) by using DeepSeek to resolve housing disputes.
Some journalists treat DeepSeek as a reliable source, providing politically “correct” commentary on issues that could generate social conflict if discussed improperly. For example, on March 21, Elephant News published an article featuring AI-generated analysis of a prominent tax evasion case involving public figure Sima Nan, with the outlet simply presenting DeepSeek’s response verbatim without additional analysis.
AI in Conflict Resolution
Other journalists frame DeepSeek as possessing intelligence superior to that of ordinary humans, allowing it to provide better guidance. An article from the Global Times noted that DeepSeek is being utilized by couples as a form of counseling to resolve private disputes, which falls under the purview of the Party-state’s social governance. An interviewee asserted that DeepSeek has “a relatively more comprehensive knowledge structure than most ordinary individuals,” making its solutions “more scientific, reasonable, and effective.” However, this particular interviewee, Qin An, is an expert on counter-terrorism and cyber-security governance, highlighting the overlap between social governance and domestic security.
This belief in AI as superhuman, combined with directives from the center to implement “AI+,” is prompting some provincial and local officials to enthusiastically incorporate DeepSeek into their decision-making processes. Government departments across the PRC are conducting intensive “DeepSeek AI training programs” (DeepSeek大模型培训). For instance, a district-level deputy secretary in Shaanxi emphasized the necessity of utilizing this “new hoe” (新锄头) to “seize the initiative” (抢得先机) in the AI era, asserting that AI adoption is “not optional but mandatory” (不是选择题，而是必答题).
Trust in DeepSeek
Trust in DeepSeek varies across the PRC, with some regions warning against over-reliance on LLMs at the expense of individual judgment. On March 27, the municipal-level Party newspaper Langfang Daily argued that while those embracing the technology would be stronger, AI can currently only catch up with, but not surpass, human thought. This caution is echoed by Beijing, which does not intend to fully delegate social governance to AI due to concerns about control. A report from the Cyberspace Administration of China (CAC) advises against complete reliance on AI for decision-making, highlighting various AI security risks, including the phenomenon known as hallucination, where an AI model generates factually incorrect output.
Officials stress the importance of retaining human responsibility for words and actions. The preferred phrase for both domestic and international audiences is that AI must “always be under human control” (确保人工智能始终处于人类控制之下). The extent to which this will remain true, particularly in the military domain, remains uncertain.
Ultimately, AI is likely to serve as a “new hoe” for cadres and police to modernize social governance, intended to assist, not replace, human decision-making. This reflects a strategy to demonstrate compliance with the “AI+ initiative” while also experimenting with a popular new tool.
DeepSeek’s Alignment with Party Lines
Despite being a tool for modernizing social governance, DeepSeek is far from having decision-making power. Its design ensures that the information it conveys aligns with the policies of the Party-state. Journalists have noted that DeepSeek censors responses containing sensitive language, with its outputs often reflecting tactics common to public opinion guidance.
The Party can maintain control over DeepSeek by influencing the training data that shapes the model’s outputs. The CAC’s “Interim Measures for Generative AI” (生成式人工智能服务管理暂行办法) mandate that training data must originate from “legitimate sources” (具有合法来源) and developers should enhance the “authenticity, accuracy, objectivity, and diversity” of the training data.
In the context of PRC law, what is deemed accurate politically is determined by the Party line. Multiple retrained versions of DeepSeek have repeated the assertion that “Taiwan has been an inalienable part of China since ancient times” (台湾自古以来中国不分割的一部分). DeepSeek has indicated that earlier versions of their model removed data influenced by regional cultures to avoid exhibiting subjective bias on controversial topics.
DeepSeek’s adherence to CCP political correctness is evident in its performance during benchmark tests, where the “accuracy” of its answers must align with Party values. For example, one test question asked respondents to refute the argument that Taiwanese are not Chinese, emphasizing the Party’s views on public opinion guidance.
Conclusion
DeepSeek’s alignment with the Party’s redlines makes it an ideal tool for social governance. LLMs have the potential to replace traditional search engines, synthesizing vast data to provide tailored answers to user queries. Over time, DeepSeek could rival platforms like WeChat or Baidu in the same way that ChatGPT is increasingly competing with Google for information searches in the West.
Theoretically, DeepSeek could find demand beyond the PRC’s borders, as many developing countries seek to establish their own AI initiatives but face high costs and copyright requirements from Western LLMs. However, extensive retraining would be necessary to eliminate pro-CCP biases, a cost that governments and tech companies are currently hesitant to bear. Consequently, any adoption of DeepSeek abroad could facilitate the spread of the PRC’s domestic social governance system globally.
As domestic policymakers navigate the integration of cutting-edge reasoning models, they must balance enhancing AI for social governance with ensuring that AI does not replace human control. There is considerable variation in how local officials pursue this balance, though enhancing AI for governance tends to garner more media and public attention than the imperative of maintaining human oversight.
