Some topics are off limits in Chinese AI, researchers find
A popular Chinese AI program, Qwen, has no memory of Tiananmen Square and no tolerance for Winnie the Pooh.
An AI model called Qwen, developed by Chinese conglomerate Alibaba and downloaded over 3 billion times, is found to contain censorship and bias favoring China's narratives. Qwen refuses to discuss events like the suppression of Hong Kong protests and Tiananmen Square massacre, labeling Falun Gong as a dangerous cult, and oddly, won't mention Winnie the Pooh due to a comparison with President Xi Jinping.
Hirundo, an Israeli cybersecurity startup, claims to have developed a method to remove this bias, calling it "AI brain surgery." They plan to publish a "Westernized" version of Qwen, warning that the model's international popularity could spread its bias toward the Chinese government. When asked if there are forced labor camps in China, Qwen responds affirmatively, stating there are vocational skills education and training centers in Xinjiang.
However, human rights organizations and governments have documented Uyghur people being forced into labor. Qwen often declines to answer questions about Tiananmen Square, reminding users to comply with relevant laws and regulations. Qwen's creator, Alibaba, didn't comment on the findings. Despite its bias, U.S. firms like Uber and Airbnb have adopted Qwen for their services, citing its cost-effectiveness.
However, experts warn that these Chinese models pose security risks and recommend their use be avoided.
Written by urgent.news from CBS News's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
