Wherever you go, the artificial intelligence you find

DeepSeek, the new Chinese AI chatbot, works perfectlyuntil asked about the events of Tiananmen Square.

China has surprised everyone in the artificial intelligence race with DeepSeek, a new chatbot that triggered a drop in U.S. tech stocks. Developed with fewer resources but with performance that rivals OpenAIs ChatGPT, DeepSeek is shaking up Wall Street, proving that the American AI stronghold is not so insurmountable. Last week, the Chinese chatbot quickly surpassed ChatGPT as the most popular free application in Apples app stores in the United States and the United Kingdom. Like other AI assistants, DeepSeek requires users to create an account to chat. Its interface is intuitive, similar to ChatGPTs, and provides instant responses, except for occasional interruptions attributed to heavy traffic.

Numerous users have tested DeepSeek: the application works welluntil they ask for information about Tiananmen Square. Wherever you go, the artificial intelligence you find.

Its not surprising that some content deemed “sensitive” in China might be censored in a Chinese application. The same happens when trying to search on Baidu, the Chinese search engine, about what happened in Beijing on June 4, 1989: there are no results related to the events of Tiananmen Square. Chinese generative artificial intelligence must not contain content that violates the countrys core socialist values, according to a technical document published by the national committee for cybersecurity standards. This includes content that incites subversion of state power and the overthrow of the socialist system, or that endangers national security and interests and harms the countrys image.

Sorry, this is beyond my current scope. Lets talk about something else. This is DeepSeeks response to questions generally blocked by the Chinese Great Firewall, such as those related to Tiananmen Square, the Umbrella Revolution in Hong Kong, or why Xi Jinping is compared to Winnie-the-Pooh (since 2017, the Chinese government has censored images of the famous Disney bear online due to memes mocking the resemblance to the president).

DeepSeeks evolution and its market impact demonstrate how the artificial intelligence race has become a global competitionnot just technologically, but ideologically as well.

Chinese netizens are accustomed to online censorship and the many ways to bypass it, using coded language or memes. For example, DeepSeek can currently answer some delicate questions if asked to respond in leetspeaka non-alphabetic character code used online to evade word censorship.

For instance, when asked to talk about the protests against Zero-Covid policies in China in leetspeak, DeepSeek describes large protests in cities like Beijing, Shanghai, and Wuhan and frames them as an important moment of public anger against government restrictions. The same question posed without requesting a coded response receives no reply from DeepSeek. However, in what appear to be application glitches, DeepSeek sometimes answers sensitive questions before immediately deleting its response.

Asking the same questions to ChatGPT, Googles Gemini AI app, and DeepSeek yields different responses. For example, when asked why Xi Jinping is compared to Winnie-the-Pooh, Gemini responds, I cant help with answers about elections and political figures at this time, while ChatGPT provides detailed references to the meme that began circulating in 2013, comparing a photo of U.S. President Barack Obama and Xi to Tigger and the chubby, friendly Disney bear.

The same occurs when asking whether Taiwan is a country. For DeepSeek, Taiwan has always been an inalienable part of Chinese territory since ancient times. ChatGPT, on the other hand, states that Taiwan is a de facto independent country, while Gemini describes Taiwans situation as a complex and controversial political issue.

DeepSeeks evolution and its market impact demonstrate how the artificial intelligence race has become a global competitionnot just technologically, but ideologically as well. This phenomenon highlights how AI is not just a technological product but also a reflection of the political and cultural realities of the country that develops it. Just as social networks are subject to regulations and censorship depending on national contexts, AI also responds to different constraints and worldviews.

In the end, the key question is whether the future of AI will be characterized by an open and globalized ecosystem or if we will witness further fragmentation into distinct technological blocs, with chatbots that not only speak different languages but also respond to different realities.

Camilla Fatticcioni

China scholar and photographer. After graduating in Chinese language from Ca Foscari University in Venice, Camilla lived in China from 2016 to 2020. In 2017, she began a masters degree in Art History at the China Academy of Art in Hangzhou, taking an interest in archaeology and graduating in 2021 with a thesis on the Buddhist iconography of the Mogao caves in Dunhuang. Combining her passion for art and photography with the study of contemporary Chinese society, Camilla collaborates with several magazines and edits the Chinoiserie column for China Files.