DeepSeek or DeepSpin: my experience with DeepSeek AI’s pro-china bias

Artificial intelligence promises to bring the world to our fingertips, offering answers with unprecedented speed and precision. Tools like China’s DeepSeek AI claim to represent the next frontier of this technology—streamlined, efficient, and capable of understanding complex questions. But from my own experience, the reality of DeepSeek AI is far more troubling. It’s not just what the platform says but what it refuses to say that should give us pause.

I recently decided to test DeepSeek AI on a range of topics to see how it would handle politically sensitive questions. The AI deflected when I asked about Taiwan’s independence—an issue that reflects a complex geopolitical reality. Instead of providing a balanced answer or exploring the nuances of Taiwan's de facto independence, it avoided the question entirely. In some cases, the AI would begin to generate a response but then immediately delete it, saying, "Sorry, that is beyond my current scope." This wasn’t an isolated instance. Attempts to raise concerns about China's human rights record or its treatment of ethnic minorities were similarly met with vague, sanitized responses or outright rejection of the query. It became clear: DeepSeek AI would not engage with anything that could cast China in a negative light.

The more I probed, the more apparent the pattern became. DeepSeek AI seemed programmed to align with Beijing’s narratives, framing questions in ways that subtly reinforced a pro-China stance or, worse, refused to address topics that might challenge it.

This refusal to engage on critical issues is not merely frustrating; it’s dangerous. AI tools like DeepSeek are poised to shape how people learn and understand the world. When they fail—or outright refuse—to provide balanced, accurate information, they risk becoming tools of propaganda rather than impartial sources of knowledge.

My experience with DeepSeek AI reflects a broader problem with artificial intelligence in an increasingly politicized world. China has a well-documented track record of using its economic and political influence to shape narratives, especially on topics it deems sensitive. The influence doesn’t stop at traditional media or social platforms; it extends to AI systems like DeepSeek. Whether this bias stems from direct interference, market pressure, or an intentional design choice, the result is the same: a platform that quietly enforces censorship under the guise of neutrality.

This matters because AI is not just about answering questions; it’s about trust. We rely on these tools to navigate an overwhelming sea of information, make sense of complex issues, and find clarity in moments of uncertainty. If an AI platform like DeepSeek skews its responses or omits inconvenient truths, it undermines that trust and leaves users unknowingly misinformed.

The issue of Taiwan is particularly illustrative of how DeepSeek AI’s biases manifest. Taiwan is a vibrant democracy with its own government, economy, and culture. Yet it exists in a precarious position due to China’s insistence that it is part of its territory—a claim that Taiwan’s people overwhelmingly reject. For DeepSeek AI to sidestep this reality is not just an oversight; it’s a deliberate choice to ignore the voices of 23 million people. And if the AI cannot acknowledge this truth, what else might it be omitting or distorting?

My experience has taught me to approach DeepSeek AI—and platforms like it—with caution. It’s a reminder that even the most advanced tools are not immune to the biases and agendas of their creators. For now, the best defense is to diversify the sources we rely on, cross-check information, and demand transparency from the companies behind these tools. Who programs the AI? What influences their decisions? And why do certain questions go unanswered?

DeepSeek AI, or should we say DeepSpin, may promise to provide the world’s knowledge at your fingertips, but if it refuses to address inconvenient truths, it raises an unsettling question of its own: Whose knowledge, and at what cost?