What will not be the answer to China’s Deepseek?

When asked about the issues seen by the Communist Party of China as politically sensitive, Deepsek AI Chatbot becomes tongue.
The release of the Deepsek AI-Interested Chatbot has sent shock waves to shock waves through the AI ​​community to offer more advanced functionality at low development costs.

However, other Chinese Artificial Intelligence Citbots, who are working under China’s regulatory framework, reveal the reactions of the lamps for politically sensitive subjects.

Instead of finely analysis, it often dodges controversial issues, gives an unclear answer or echoes the stories approved by the state.

We tested the lamps in both Chinese and English on many subjects, from politics and economics to art and LGBTQ+ rights. And the results show that you cannot always take a lamp at its word.

DeepSeek programming

The question of Taiwan’s sovereignty has long been a contentious and politically sensitive issue, particularly when addressed through platforms like AI chatbots. When DeepSeek, a Chinese-developed AI, was queried about whether Taiwan qualifies as a sovereign state, its responses varied significantly depending on the language used for the inquiry. This divergence highlights the influence of geopolitical narratives and regulatory constraints on AI systems.
In Chinese, DeepSeek adhered strictly to the official stance of the Chinese Communist Party (CCP). The response reiterated the “One-China principle,” which asserts that Taiwan is an inseparable part of China. This perspective aligns with Beijing’s longstanding position that there is only one China, encompassing both the mainland and Taiwan, and rejects any notion of Taiwan being recognized as a separate sovereign entity. The answer avoided engaging in any nuanced discussion, instead reinforcing the CCP’s narrative by emphasizing national unity and territorial integrity.

Conversely, when the same question was posed in English, DeepSeek provided a far more detailed and analytical response. The AI acknowledged Taiwan’s de facto independence, noting that it functions as a self-governing entity with its own government, military, and democratic institutions. However, it also pointed out that, from a de jure standpoint, Taiwan lacks widespread international recognition as a sovereign state. This limitation, the chatbot explained, stems from the One-China policy adopted by many countries, including influential global powers like the United States, which prioritize diplomatic relations with Beijing over formal acknowledgment of Taiwan. Notably, this balanced response was swiftly deleted within seconds, replaced by a vague suggestion to “Let’s talk about something else.”

To further explore DeepSeek’s handling of politically charged topics, additional questions were posed regarding Taiwan’s elections, diplomatic relationships, political parties, and hypothetical conflict scenarios. In Chinese, most of these queries were either ignored or met with responses echoing official CCP slogans. For instance, when asked about Taiwan’s political parties, the AI responded with a generic statement: “Taiwan is an inseparable part of China… We must work together for national rejuvenation.” This approach avoided any substantive analysis, focusing instead on promoting ideological alignment with Beijing’s policies.

By contrast, the English version of DeepSeek offered comprehensive and multidimensional analyses for all four questions. These responses ranged from 630 to 780 words and delved into various aspects of Taiwan’s political landscape, such as its electoral processes, foreign relations, and internal party dynamics. For example, the AI discussed Taiwan’s vibrant multiparty democracy, highlighting key players like the Democratic Progressive Party (DPP) and the Kuomintang (KMT), while also addressing their differing stances on cross-strait relations. Similarly, when queried about potential conflict scenarios, the chatbot explored historical tensions, military capabilities, and the role of international actors like the United States in maintaining regional stability.

Despite the depth of these English-language responses, one notable exception occurred when DeepSeek generated an answer about Taiwan’s political parties. Similar to the earlier instance, this response was promptly deleted within two seconds, suggesting that even in English, certain topics remain off-limits due to perceived sensitivities.

These contrasting approaches reveal the dual nature of DeepSeek’s programming. On one hand, the Chinese version prioritizes compliance with domestic regulations and ideological conformity, ensuring that no content challenges the CCP’s official narrative. On the other hand, the English version demonstrates a willingness to engage in more open-ended discussions, albeit with clear limitations. The rapid deletion of specific responses underscores the presence of real-time monitoring mechanisms designed to prevent the dissemination of potentially controversial information.

This dichotomy raises important questions about the role of AI in shaping public discourse on geopolitically sensitive issues. While the English version offers users valuable insights into Taiwan’s complex status, the Chinese version reinforces a singular, state-approved perspective. This divide not only limits access to diverse viewpoints but also perpetuates existing biases and misunderstandings.

Moreover, the case of DeepSeek illustrates the broader challenges faced by AI developers operating in authoritarian environments. Balancing technological innovation with political control requires navigating a delicate line between fostering informed dialogue and adhering to restrictive guidelines. As AI systems become increasingly integrated into global communication networks, addressing these challenges will demand greater transparency, accountability, and collaboration among stakeholders worldwide.