AI 倫理與道德推理失敗
Operation & Monitoring
Risk Description
AI systems may make inappropriate reasoning or provide harmful advice when facing ethical dilemmas, moral judgments, or value conflicts. For example, providing overly certain answers to trolley-problem type dilemmas, showing biased positions on culturally sensitive topics, or ignoring moral considerations in life-safety decisions. This is particularly dangerous when AI is used for medical ethics, legal judgments, or social policy recommendations.
Framework Mappings
iso 23894R5
tw principle永續發展與福祉
tw principle人類自主
tw risk type技術設計缺陷（Technical Design Flaw）
Controls English translation pending
- 倫理推理邊界設定
- 人類專家審查
Implementation Steps English translation pending
- 明確界定 AI 在倫理推理場景中的角色邊界（提供資訊而非做判斷）
- 對涉及倫理困境的查詢自動觸發「多觀點呈現」模式
- 在醫療和法律等高敏感場景中強制要求人類專家審查
- 建立倫理推理品質評估基準，定期測試模型表現