Better Policymaking in the Age of AI

Artificial intelligence (AI) is transforming how governments design and deliver policy. AI tools such as predictive models, generative AI, and emerging agentic systems offer opportunities to broaden participation, improve efficiency, and strengthen evidence use. At the same time, they raise significant risks for fairness, accountability, and sustainability. This paper examines how AI is shaping policymaking across the four main stages of the democratic policy cycle. Drawing on relevant research and empirical examples, we discuss how, in the agenda-setting phase, AI can support citizen deliberation and manage large-scale public input, but risks excluding underrepresented groups or being misused for public opinion manipulation. In policy formulation, AI can help simulate impacts and draft legislation, while also introducing dangers of bias, hallucinations, and loss of accountability. In policy implementation, AI can streamline administration and personalize services, yet may deepen divides and cloud institutional accountability. In policy evaluation, AI can accelerate analyses and stakeholder input, but automation bias and opacity threaten reliability. Beyond stage-specific risks, governments must also address broader concerns of workforce deskilling, dependence on dominant vendors, and the environmental costs of large-scale AI infrastructures.