AI is changing the policy cycle. Are policymakers ready?

Artificial intelligence has become part of daily life, and increasingly, part of daily governance. From the way governments identify problems to the way they evaluate whether a policy has worked, AI is beginning to reshape every stage of the policy cycle. The question is no longer whether AI belongs in public policy; it is whether policymakers are equipped to use it well.

Where AI fits into the policy cycle

Different stages of policymaking call for different uses of AI and understanding that distinction is where good practice starts.
Problem identification and agenda setting. Before a policy can be designed, it has to be understood, and that means making sense of an overwhelming amount of unstructured information. AI tools can help here directly: processing large volumes of text to gauge public sentiment, synthesising academic literature, policy reports, and legal documents, surfacing historical trends, and condensing lengthy material into something a policymaker can actually use.

Policy identification and adoption. Once a problem is understood, the next challenge is choosing between options. This is where predictive analysis, forecasting, and simulation can help. By modelling the likely outcomes of different policy choices, AI can ground decision-making in evidence rather than intuition, and in doing so, help take some of the politics out of policy choice. Geospatial analysis is a good example in practice: mapping real data on transportation use, school locations, or hospital access can reveal concrete ways to improve public services.

Policy implementation. Once a policy is underway, the challenge shifts to targeting ; making sure resources reach the people and places that need them most. Geospatial analysis again proves useful here, for instance, mapping poverty distributions to sharpen how a programme is implemented on the ground.

Policy evaluation. The final stage of the cycle is where AI's potential is perhaps most visible. As data collection becomes cheaper and more abundant, impact analysis can draw on much larger datasets than before. Geospatial analysis and live dashboards techniques allow for real-time evaluation, giving policymakers the ability to adjust course quickly rather than waiting for a policy cycle to conclude before learning what worked.

The tool is not the decision-maker. For all its capability, AI remains exactly as a tool. It does not set policy strategy, that requires human leadership. It does not build itself, that requires engineers to design and train the underlying models. And it certainly does not interpret its own outputs in a way that accounts for fairness, context, or ethics, that responsibility sits with the people using it. This last point matters most. AI can surface patterns, generate forecasts, and process more information than any human team could manage alone, but it cannot decide what is fair and equitable, and it also cannot be held accountable. Human policymakers can. That is precisely why the people at the table need to understand not just what AI can do, but where its boundaries are, and where their own judgment must take over. This does not mean every policymaker needs to become an engineer. It does mean that understanding the value, and the limits, of AI within public systems is quickly becoming a core competency, not an optional one. Policymakers are the safeguard for ensuring AI is used ethically and fairly within government, and that role cannot be outsourced to the technology itself.

Building that understanding 

This is exactly the gap our course, Using AI for Policy, is designed to close. It offers a grounded introduction to the theory behind AI use in public policy settings, paired with real examples and cases of practice from across the policy cycle. Rather than treating AI as a black box or a buzzword, the course is built to help participants understand where it genuinely adds value, where caution is warranted, and how to use it responsibly in their own work.

I have seen this need first-hand. As I write this, I am in Sri Lanka, training government officials on AI use, and the same questions come up again and again: not "can AI do this?", but instead we discussed questions like "Should we let AI do it?”, “How do we navigate the AI platforms safely and ethically?" and “How can the AI platforms become a valuable assistant for us”. Those are exactly the questions this course is designed to help policymakers answer.

Interested in learning more? Apply for the "The use of AI to inform Policy" course by 1 September, 2026.