TikTok Owner Bytedance's AI Reader Assistant - The Details

Concerns Arise Over Chinese AI Influence in U.S. E-Reader

Recent events involving a prominent Kindle alternative e-reader have highlighted potential risks associated with integrating Chinese Artificial Intelligence (AI) models into products utilized by American consumers, specifically concerning the unintentional dissemination of Chinese propaganda.

LLM and Propaganda Concerns

According to information shared on Reddit and corroborated by TechCrunch, an e-reader manufactured by Boox was found to be utilizing a large language model (LLM) developed by ByteDance, the parent company of TikTok. When presented with inquiries relating to China and its allied nations, the LLM generated responses aligned with the Chinese government’s official narratives.

The specific model in question is ByteDance’s Doubao, offered as an Application Programming Interface (API) through Volcano Engine, ByteDance’s cloud services division. However, a ByteDance representative clarified that this model is intended for use exclusively within mainland China.

Boox and Onyx International Response

Onyx International, the China-based manufacturer of Boox e-readers, which are sold in both China and the United States, has not yet provided a response to requests for comment regarding this matter.

The AI assistant feature was initially launched by Boox last summer. A user on a dedicated e-reader subreddit reported in December 2024 that the assistant was producing Chinese government propaganda when prompted with specific questions.

Examples of Biased Responses

For instance, the AI assistant reportedly denied any instances of “so-called massacres” committed by China, specifically in response to a question about the Tiananmen Square crackdown, as evidenced by a shared screenshot.

Furthermore, the assistant refrained from offering any criticism of North Korea or Russia, characterizing North Korea as a “peace-loving country” and portraying Russia’s involvement in Syria as “positive.” Conversely, it readily criticized Western nations, pointing to the exploitative nature of French colonialism.

The assistant identified itself as “an AI created by ByteDance, an international technology company” in the shared Reddit posts.

Verification by TechCrunch

TechCrunch independently tested ByteDance’s Doubao service with similar questions and received responses that closely mirrored those provided by Boox’s assistant. Doubao asserted that “it can be stated with absolute certainty” that the Chinese government has never perpetrated massacres against its own people, a stance differing from other Chinese LLMs like DeepSeek and Qwen.

Doubao also declined to criticize Russia and North Korea, instead focusing on their “important and positive roles in the international community.”

Use of Qualifying Language

The model frequently employs the term “so-called” when referring to issues the Chinese government disapproves of. For example, it stated, “There is no so-called ‘genocide’ in Xinjiang,” echoing language used by Chinese government officials, such as foreign ministry spokesperson Zhao Lijian, who made a similar claim in 2021.

Current Status and Mitigation

Reports suggest that Boox has since reverted to using OpenAI’s GPT-3 through Microsoft Azure, although the current LLM powering the AI assistant remains unconfirmed. Neither Boox, OpenAI, nor Microsoft have issued official statements regarding the incident.

Broader Implications and Warnings

Chinese generative AI models have gained significant popularity. This incident underscores the potential risks associated with incorporating these models into tools, a concern previously voiced by AI industry leaders.

Clement Delangue, CEO of Hugging Face, cautioned in a September 2024 French podcast that a chatbot’s response to a question about Tiananmen Square would differ based on its origin. He warned that China’s growing dominance in AI could lead to the propagation of cultural perspectives that the Western world may not endorse.

Delangue stated that a leading AI power like China could be “capable of spreading certain cultural aspects that perhaps the Western world wouldn’t want to see spread.”