AI 傷害的可感知性與可量化性挑戰

Risk Description

Many AI harms are gradual, indirect, or systemic, making them difficult to perceive timely and quantify accurately. For example, algorithmic discriminations long-term impact on individual careers, recommendation systems slow erosion of information diversity, or AI automations economic impact on specific communities. This harm invisibility makes it difficult for victims to present evidence, regulators to intervene, and organizations to find motivation for correction.