رؤى / تقرير بحثي
من التجارب إلى نموذج تشغيلي
توزيع صلاحيات القرار قبل تفويض العمل إلى الذكاء الاصطناعي
توليف للمصادر الأولية وتحليل mayden. المنهج والحدود موضحان.
الملخص التنفيذي
- قدرة النموذج لا تمنحه صلاحية التصرف.
- حدد من يملك مرجعية الأدلة والتصرف والاستثناءات والتغيير.
- اجعل المراجعة المستقلة قادرة على تغيير قرار الإطلاق.
اقرأ التقرير الكامل بالإنجليزية
يتوفر الملخص التنفيذي بالعربية. يتضمن التقرير الإنجليزي التحليل الكامل والرسوم التوضيحية والأسئلة القيادية والمراجع.
قراءة التقرير بالإنجليزيةتنزيل التقريرمنهج البحث وحدوده
كيفية قراءة هذا البحث
تفاصيل المنهج والمراجع أدناه باللغة الإنجليزية.
- Approach
- Structured synthesis of eight governance, engineering and human–AI research sources with an original decision-rights architecture.
- Source selection
- Selected sources clarify authority, oversight, validation, documentation and maintenance. Accessed 10 October 2026, including the April 2026 banking guidance rather than presenting superseded guidance as current.
- Limitations
- Normative frameworks do not demonstrate causal performance improvements. The human–AI meta-analysis covers heterogeneous pre-July-2023 experiments. Banking and government material is used by analogy, not as universal legal advice.
المراجع / 8 مصادر
الأدلة التي يستند إليها التحليل
- [1]Artificial Intelligence Risk Management Framework (AI RMF 1.0) ↗
National Institute of Standards and Technology · January 2023
Government framework · Part 2; GOVERN, MAP, MEASURE and MANAGE tables
Guidance is not experimental proof that any specific governance structure works.
- [2]Artificial Intelligence: An Accountability Framework for Federal Agencies and Other Entities ↗
US Government Accountability Office · 30 June 2021
Government oversight framework · Highlights and framework principles
Normative framework, not a measured causal estimate.
- [3]Supervisory Guidance on Model Risk Management ↗
Federal Reserve; interagency SR-26-2 · 17 April 2026
Supervisory guidance · Footnote 3; governance and validation sections
Not applicable to generative or agentic AI models. Any extension in this report is mayden analysis, not regulator endorsement.
- [4]Model Cards for Model Reporting ↗
Mitchell et al.; Google Research · 2019
Research framework · Abstract
Documentation can expose limits; it does not independently validate a deployment.
- [5]When Are Combinations of Humans and AI Useful? ↗
Vaccaro, Almaatouq and Malone; author manuscript · 9 May 2024; arXiv version 1
Preregistered systematic review and meta-analysis · Sections 2.1 and 4; 106 experiments and 370 effect sizes, publications January 2020–June 2023
Studies through June 2023; task heterogeneity and publication bias; not current-model ROI.
- [6]Artificial Intelligence Playbook for the UK Government ↗
Government Digital Service · 10 February 2025
Government guidance · Principles; human oversight; life cycle and commercial guidance
UK public-sector context; not a substitute for jurisdiction-specific advice.
- [7]Hidden Technical Debt in Machine Learning Systems ↗
Sculley et al.; NeurIPS · 2015
Engineering research · Abstract
Predates generative AI; used for structural insight, not current failure-rate estimates.
- [8]The Cybernetic Teammate: A Field Experiment on Generative AI and Teamwork ↗
Dell’Acqua et al.; Organization Science 37(4), 1217–1242 · 12 June 2026
Preregistered randomized field experiment · Abstract; Section 3 experimental design; Section 5.4.1 and Figure 8(b,d) (idea selection versus selected-idea quality); footnote 8 (participant and complete-survey counts)
One-day workshop, specified tools, one company; not a staffing or autonomy experiment. Published n differs from older working paper.
