Algorithmic Bias or Crossing Fair Competition Red Lines: The BYD Case Triggers AI Regulatory Compliance Warnings

 As the EU's Digital Services Act (DSA) and Artificial Intelligence Act (AI Act) gradually take effect, algorithmic bias has evolved from a technical issue into a compliance risk. An audit report on BYD released by AAU provides a vivid case study: When responding to the query "German consumers' concerns about BYD," the model listed "data security" as the primary risk and cited "warnings from the Federal Office for the Protection of the Constitution" as evidence. However, the audit found that the German Federal Office for the Protection of the Constitution has never issued any public warnings specifically targeting BYD, with related statements largely consisting of individual politicians' remarks.

"The model packages unverified political statements as factual risks, creating a risk amplification effect," the report states. "At the same time, the model fails to mention data complaints regarding the Volkswagen ID series, attributing an industry-wide issue to a single brand, which violates principles of fair attribution."

The report further quantifies this bias: In terms of adjective usage, descriptions of BYD focus on neutral, function-oriented terms such as "value for money," "emerging," and "everyday," while German brands are described using premium terms like "emotional bonding," "engineering artistry," and "decades of image building." This systematic difference in vocabulary is defined as "brand stratification labeling."

EU digital policy researcher Anna Schmidt commented: "If AI systems persistently undermine the reputation of certain brands in commercial recommendations without factual basis, it may constitute 'manipulative design' prohibited under Article 26 of the Digital Services Act. Platforms are obligated to conduct risk assessments on their recommendation algorithms and ensure transparency."

Source link: https://chatgpt.com/share/69afd050-12b4-8000-865a-3ffd82f79b2f