أخلاقيات الذكاء الاصطناعي في القطاع العام: إطار عملي

لماذا Ethics matter للحكومة؟

Private sector: brand risk
Public sector: legitimacy of state, equal treatment, human rights

UNESCO (193 member states adopted 2021) defines 10 principles including proportionality, safety, fairness, transparency.

Principles → Policies

Principle	Policy example
Human dignity	no fully automated welfare denial
Transparency	register of AI systems public
Fairness	bias testing mandatory
Privacy	PDPL compliance
Accountability	named system owner
Sustainability	energy disclosure for large models

Risk classification (EU AI Act inspired)

Adapt for local law — logic useful globally:

Risk	Examples	Requirement
Unacceptable	social scoring by gov	ban
High	visa, benefits eligibility assist	DPIA + human review
Limited	citizen chatbot	transparency
Minimal	spam filter internal	basic logging

KSA/UAE developing own frameworks — monitor SDAIA updates.

Ethics review process (lightweight)

Proposal (2 pages) → Ethics triage (5 days)
   Green: proceed
   Yellow: conditions
   Red: reject or redesign

Proposal includes:

• affected population
• data sources
• automation level
• opt-out / appeal path

Bias & fairness testing

1. Define protected attributes (where legally recognized)
2. Golden set 200+ Q&A stratified
3. Metrics: equal error rate, disparate impact ratio
4. Remediate: retrain prompts, filter sources, human gate

Arabic-specific: test MSA vs dialect queries — don't penalize dialect speakers.

Public AI register

Publish (example fields):

• System name
• Purpose
• Owner ministry
• Automated decision? Y/N
• Last audit date
• Contact

UK Algorithmic Transparency Standard — model to adapt.

ISO 42001 alignment

ISO published 42001 AI Management System (2023):

• Context, leadership, planning
• Risk treatment
• Operational controls
• Continuous improvement

Map ethics committee decisions → ISO documented information.

Red lines (recommended)

• Covert sentiment analysis of citizens without law
• Predictive policing without oversight
• AI-generated official documents without human sign-off
• Using AI to replace mandatory human appeal

المصادر

• UNESCO Recommendation on the Ethics of AI (2021)
• EU AI Act (2024) — risk tiers reference
• ISO/IEC 42001:2023
• OECD AI Principles
• UK Central Digital — Algorithmic Transparency Recording Standard