A controversy surrounding a popular e-reader has highlighted the risks of integrating Chinese AI technology in products distributed in international markets. Boox, a competitor to the Kindle, came under scrutiny when users discovered that its newly launched AI assistant, powered by a Chinese AI model, was delivering responses that parroted Chinese government propaganda.
The language model (LLM) in question was ByteDance's Doubao, offered through the company's Volcano Engine cloud services. While Doubao is intended solely for use within China's mainland, it somehow made its way into Onyx International's Boox e-reader, which is sold globally, including in the U.S. Screenshots from Reddit posts revealed the AI assistant’s tendency to deny or downplay sensitive topics, such as the Tiananmen Square massacre, referring to such events as "so-called massacres" and presenting China's official government stance as fact. Users also noted other problematic responses. The assistant refused to critique allies of China, such as North Korea and Russia, framing North Korea as a “peace-loving country” and highlighting Russia’s “positive role in Syria.” At the same time, it was quick to criticize Western countries, including negative remarks about French colonialism.
The Reddit post documenting these responses went viral, sparking widespread backlash. Publications like The Decoder and popular YouTube channels such as The China Show quickly picked up the story. When TechCrunch tested ByteDance’s Doubao, it replicated the patterns seen in the Boox assistant. Doubao’s responses heavily echoed language used by Chinese government spokespeople, including terms like "so-called" to dismiss topics sensitive to the Chinese government. For instance, Doubao claimed there was “no so-called genocide in Xinjiang,” aligning with statements from Chinese officials.
Following the backlash, reports suggest that Boox has switched its AI assistant back to OpenAI’s GPT-3 via Microsoft Azure. However, the company has not released an official statement, leaving uncertainty about the precise LLM currently in use. OpenAI and Microsoft also did not provide comments on the issue.
The incident serves as a broader cautionary tale for technology companies using Chinese generative AI tools. These models, while increasingly powerful and widely adopted, carry inherent risks, especially when distributed in markets with differing political and cultural values. Industry leaders, including Hugging Face CEO Clement Delangue, have raised concerns about potential cultural and political biases being embedded into AI tools and their broader implications. As Delangue noted, AI models developed within China inevitably reflect the priorities and constraints of their origin, which may conflict with the expectations of users in other regions.
While the immediate backlash over Boox's AI assistant has subsided, the story underscores the need for greater scrutiny and transparency when integrating AI technologies across global markets. The question remains: how should companies navigate the balance between leveraging advanced technologies and addressing their inherent geopolitical complexities?
