📄 中文摘要
人工智能系统已广泛应用于影响人们生活的决策中，如信用审批、欺诈检测、招聘、承保和客户支持自动化。尽管人工智能的采用速度迅猛，但治理框架却未能跟上。大多数组织仍使用传统软件测试方法来测试人工智能系统，这种方法并不适用，因为人工智能系统的行为与传统软件截然不同。传统软件是确定性的：相同的输入每次产生相同的输出。而人工智能系统是概率性的，它们从数据中学习模式，适应新输入，并可能在不同交互中产生不同的输出。因此，组织需要在监管者、攻击者或用户揭露失败之前，审计人工智能系统，以确保其准确性和合规性。
📄 English Summary
Auditing AI Systems: A Practical Guide to Testing Models for Bias, Compliance, Security, and Explainability
AI systems are now integrated into decisions that impact people's lives, including credit approvals, fraud detection, hiring, underwriting, and customer support automation. Despite the rapid adoption of AI, governance frameworks have struggled to keep pace. Most organizations continue to test AI systems using traditional software testing practices, which are inadequate due to the distinct behavior of AI systems. Traditional software is deterministic, producing the same output for the same input every time, while AI systems are probabilistic, learning patterns from data and adapting to new inputs, potentially leading to varied outputs across interactions. Organizations must audit AI systems proactively to address issues of bias, compliance, security, and explainability before regulators, attackers, or users highlight failures.