Defining the Rules of the Game: Towards a Legal Framework for Public Sector AI
3598_On the Way to Legal Framework for AI in Public Sector.
The paper "On the Way to Legal Framework for AI in Public Sector" outlines the urgent legal challenges and regulatory requirements for integrating AI into governance. It proposes a foundational roadmap for legal personality, liability, and transparency in "Smart Government" ecosystems.
TL;DR
As AI transitions from a tool to a decision-maker in the public sector, the legal landscape remains dangerously human-centric. This paper identifies the core legal paradoxes—from liability in autonomous systems to the "right to human interference"—and proposes a structured roadmap for creating a "Smart Government" that is both innovative and legally accountable.
Problem & Motivation: The Accountability Gap
The rapid adoption of AI in infrastructure, medicine, and legal drafting has outpaced the evolution of law. Traditional legal systems assume a sentient, human agent at the center of every decision. When an AI system manages a city's transport or healthcare data, who is liable when the algorithm fails?
The authors highlight a critical friction point: The "Black Box" Problem. If public decisions are made by opaque algorithms, they violate the fundamental democratic principle of transparency. Without a legal framework, the use of AI risks eroding public trust and infringing on civil rights.
Methodology: The Five Pillars of AI Law
The paper deconstructs the legal challenge into five thematic areas:
- Legal Personality: Should AI be treated as a "person," a "tool," or a new category of "electronic agent"?
- Human Rights: Protecting individuals from automated bias and establishing the right to demand human review.
- Liability: Moving from "operator-fault" to a system that accounts for the unpredictability of Big Data analysis.
- Transparency: Treating public sector algorithms as "Open Data" that must be auditable by the public.
- Public Trust: Ensuring that AI involvement in critical social spheres (unemployment, crime) is socially accepted and ethically governed.
Strategic Recommendations: A Roadmap for Regulation
The authors propose five concrete steps to bridge the gap:
- Standardized Definitions: Legally defining what constitutes "AI" versus simple automation.
- Risk-Based Certification: Not all AI is equal; medical AI requires more rigorous certification than a simple administrative chatbot.
- The Three-Tier Admission Model:
- Prohibited Areas: Where human judgment is non-negotiable (e.g., certain judicial verdicts).
- Admissible Areas: Where AI assists but humans decide.
- Recommended Areas: Where AI efficiency is necessary (e.g., complex data processing).
- Algorithmic Disclosure: Ensuring that the "code" behind public policy is accessible, similar to the precedents set in French administrative law.
Critical Analysis & Conclusion
The value of this work lies in its early recognition (2018) of issues that are now at the forefront of global policy (such as the EU AI Act). While the paper provides a strong philosophical and categorical foundation, it acknowledges its own limitations: specifically, the difficulty of enforcing technical transparency without compromising security or intellectual property.
Takeaway: The "Smart City" of the future cannot exist without a "Smart Legal System." The transition is not just technological—it is a fundamental restructuring of how we define responsibility and rights in a machine-augmented society.