NIST发布最新版人工智能风险管理框架草案（英）

Remarkable surges in artificial intelligence (AI) capabilities have led to a wide range of innovations with the potential to benefit nearly all aspects of our society and economy – from commerce and healthcare to transportation and cybersecurity. AI technologies are often used to achieve a beneficial impact by informing, advising, or simplifying tasks.

Managing AI risk is not unlike managing risk for other types of technology. Risks to any software or information-based system apply to AI, including concerns related to cybersecurity, privacy, safety, and infrastructure. Like those areas, effects from AI systems can be characterized as long- or short-term, high- or low-probability, systemic or localized, and high- or low-impact. However, AI systems bring a set of risks that require specific consideration and approaches. AI systems can amplify, perpetuate, or exacerbate inequitable outcomes. AI systems may exhibit emergent properties or lead to unintended consequences for individuals and communities. A useful mathematical representation of the data interactions that drive the AI system’s behavior is not fully known, which makes current methods for measuring risks and navigating the risk-benefits tradeoff inadequate. AI risks may arise from the data used to train the AI system, the AI system itself, the use of the AI system, or interaction of people with the AI system.