Guiding AI Through Government Roles
==Governments can guide AI by acting in three roles at once: as users of AI in public services, as regulators of AI risks, and as enablers of broader AI adoption in the economy.== [:cite[1]{ln=1}] [:cite[2]{ln=1}]...
==Governments can guide AI by acting in three roles at once: as users of AI in public services, as regulators of AI risks, and as enablers of broader AI adoption in the economy.== [:cite[1]{ln=1}] [:cite[2]{ln=1}] Start with problems, not hype. Governments should first ask whether AI is the right solution for a specific public problem, given available budgets, data, and institutional capacity, rather than assuming every problem needs AI. [:cite[3]{ln=5}] [:cite[4]{ln=1}] [:cite[4]{ln=2}] Match AI ambition to government readiness. The report recommends choosing among adopting, adapting, or advancing AI based on an agency’s readiness in data quality, human capital, infrastructure, and governance capacity. Early stage governments should usually start with adoption, connected governments can adapt, and only advanced governments should consider building complex models from scratch. [:cite[7]{ln=1}] [:cite[6]{ln=1}] [:cite[5]{ln=1}] [:cite[5]{ln=3}] Begin with low risk, practical uses. For less digitally mature agencies, governments can start by training staff to use general purpose AI for low risk tasks such as drafting, summarizing, or translating, or by procuring proven solutions for specific workflows. [:cite[7]{ln=2}] [:cite[7]{ln=3}] Build internal capacity, not just technical ambition. The report stresses that governments do not mainly need frontier AI researchers; they need staff who can understand workflows, evaluate vendor claims, reduce administrative friction, and ensure systems evolve as policies change. [:cite[8]{ln=3}] [:cite[8]{ln=4}] [:cite[8]{ln=5}] Give digital teams authority and expand AI literacy. Governments should empower existing digital service teams, involve them in procurement and system decisions, and build AI literacy across the broader civil service so officials know what AI can and cannot do and when to override it. [:cite[10]{ln=2}] [:cite[10]{ln=3}] [:cite[9]{ln=2}] [:cite[11]{ln=2}] [:cite[12]{ln=1}] [:cite[12]{ln=5}] Issue clear guidelines and provide approved tools. The report says AI licenses and guidelines are essential for appropriate and effective use by public officials, especially to reduce unsafe “shadow AI” use. [:cite[13]{ln=1}] [:cite[14]{ln=1}] [:cite[14]{ln=4}] Make government data AI ready. Governments should digitize records, improve routine data systems, and create structured, machine readable data pipelines, because unreliable data can scale small errors into large harms. [:cite[4]{ln=5}] [:cite[4]{ln=6}] [:cite[15]{ln=1}] [:cite[15]{ln=2}] [:cite[16]{ln=2}] [:cite[16]{ln=3}] Connect data across agencies through digital public infrastructure. The report highlights digital public infrastructure—such as digital identity, payments, registries, data exchange, and verifiable credentials—as critical for linking data across government and making AI both more useful and more accountable. [:cite[17]{ln=1}] [:cite[17]{ln=2}] [:cite[17]{ln=5}] [:cite[17]{ln=6}] [:cite[18]{ln=1}] [:cite[18]{ln=3}] Govern data carefully. Because governments collect sensitive data under conditions where citizens often have little choice, they need governance that balances beneficial reuse with protection against misuse, surveillance, and reuse without consent. [:cite[19]{ln=1}] [:cite[19]{ln=2}] [:cite[19]{ln=3}] [:cite[19]{ln=5}] Procure AI without losing control. Governments should buy only AI they can govern, preserve the ability to change course, require data portability, avoid exclusivity, and maintain internal teams with authority over vendors to reduce lock in. [:cite[20]{ln=3}] [:cite[20]{ln=4}] [:cite[20]{ln=5}] [:cite[21]{ln=1}] [:cite[21]{ln=2}] Use procurement to shape better local AI markets. Public procurement can crowd in local AI solutions, especially in high stakes sectors where local context matters, and can lower barriers for smaller innovative firms to compete. [:cite[22]{ln=2}] [:cite[22]{ln=3}] [:cite[22]{ln=5}] [:cite[23]{ln=1}] [:cite[23]{ln=4}] Evaluate and monitor continuously. Governments should set benchmarks before deployment, validate outputs in context, monitor performance over time, and document results transparently so AI improves outcomes and risks are caught early. [:cite[20]{ln=1}] [:cite[20]{ln=2}] [:cite[24]{ln=1}] [:cite[24]{ln=5}] [:cite[25]{ln=6}] Use existing laws now, and add new rules where gaps appear. The report says governments do not start from zero: data protection, cybercrime, consumer protection, constitutional law, and sectoral rules may already apply to AI, so the first step is to assess the current landscape and identify gaps. [:cite[26]{ln=1}] [:cite[26]{ln=2}] [:cite[27]{ln=1}] [:cite[28]{ln=2}] [:cite[28]{ln=3}] Combine regulation with standards. There is no single best practice AI law for every country; instead, governments should combine laws and regulation with voluntary standards and ethical frameworks, using sta...