How to Govern AI Ethically: Frameworks, Risks, and Real‑World Practices

AI Governance and Ethics Overview

The presentation focuses on AI governance and ethical issues that arise as AI technologies are widely adopted. Building on data governance foundations, new topics in AI governance are examined.

1. AI Business Application Scenarios

AI systems are defined as autonomous entities that process data, learn, and adapt, featuring three core characteristics: data processing, autonomous decision‑making, and continuous self‑improvement. Eight major application directions are identified, including customer management, human‑resource analytics, fraud detection, data privacy protection, credit risk assessment, content moderation, process automation, and personalized marketing.

2. AI and Data Risks

Three primary risk categories are highlighted:

Data quality risk – poor data leads to inaccurate models ("garbage in, garbage out").

Data security risk – automated decisions must be protected against leakage, tampering, or misuse.

Data compliance risk – strict adherence to regulations such as China’s Personal Information Protection Law and Data Security Law is required.

Additional AI‑specific risks include model bias, lack of explainability, and broader social‑ethical concerns such as job displacement.

3. AI Governance Framework

International research identifies core principles: transparency & explainability, fairness & non‑discrimination, accountability, privacy & data security, human oversight, remediation & competitiveness, data quality, and societal well‑being. Thirteen essential elements are proposed, covering responsibility assignment, compliance assessment, scenario cataloguing, data value enhancement, fairness safeguards, reliability verification, transparency standards, human‑in‑the‑loop mechanisms, privacy protection, security layers, full‑cycle management, risk management, and value realization.

4. AI Governance Practice

Practical guidance emphasizes integrating AI governance with traditional data governance, avoiding data silos, balancing automation with human oversight, and strengthening privacy safeguards through encryption, anonymisation, and privacy‑preserving computation. Open‑source model licensing and compliance considerations are also discussed.

5. Q&A – Measuring AI Governance Benefits

Key benefit metrics include risk reduction, trust enhancement, innovation enablement, fairness promotion, and contribution to sustainable development goals. Examples such as AI‑driven intelligent Q&A and AI2SQL illustrate efficiency gains and accuracy improvements after data quality enhancements.

Overall, the session provides a comprehensive roadmap for establishing, operating, and continuously improving AI governance within enterprises.