人工智能歧视的法律治理

Legal Governance of Artificial Intelligence Discrimination

人工智能在我国的大规模部署带来了歧视数字化的法律风险。个体层面由内隐偏见驱动的无意识歧视和社会层面由结构性不平等驱动的系统性歧视,或影响人工智能的开发设计,或干预数据生成、采集和利用,在问题建构、数据理解、特征选择等环节将偏见和结构性不平等嵌入人工智能,诱导其输出歧视性结果。面对数字时代更加隐蔽且广泛的歧视,我国基于责任规则、信息约束和平权行动构建的既有禁止歧视法律体系陷入归责不能、约束失效等困境。治理人工智能的歧视需要实现反歧视法律数字化转型,以算法解释化解信息单向透明趋势,以算法审核抑制代码歧视风险,以非歧视准则规范人工智能开发、利用行为。此外,还需借助数字平权行动,推动社会权力结构变迁,消除滋长歧视的结构性不平等。

The large-scale deployment of artificial intelligence in China has brought legal risks of digitalized discrimination.Unconscious discrimination driven by implicit bias at the individual level and systemic discrimination driven by structural inequality at the social level may affect the design of artificial intelligence or intervene in relevant data generation,collection and utilization.By grafting bias and structural inequality into artificial intelligence through problem construction,data understanding,feature selection,etc.,they induce it to output unexpecting discriminatory results.In the face of more insidious and widespread discrimination in the digital age,traditional legal measures for prohibiting discrimination based on liability rules,information constraints and affirmative actions have all suffered the failure of liability fixation or constraint failure.Governing artificial intelligence discrimination requires the digital transformation of anti-discrimination laws.It is necessary to rely on algorithmic interpretation to rebalance increasing information asymmetry,algorithmic audit to prevent discrimination from being embedded into code,and non-discriminatory rules to standardize artificial intelligence development and utilization conduct.In addition,digital affirmative action is also needed to accelerate the transition of the social power structure to eliminate structural inequality that breeds discrimination.