This article is Part 10 of Sally Washington's Building Policy Capability series, which looks at how to improve the work of government through high-performing policy systems. 'Sense making and making sense' takes a look at where AI can assist with policy design and delivery as well as the challenges of, "bringing artificial intelligence into the part of government work that arguably requires the most human intelligence."


AI is the new black. But is it just an accessory or do we need a totally new wardrobe? In terms of AI and public policy, debates are occurring on several fronts. Governments are grappling with AI policy: regulatory settings, privacy concerns, data sovereignty etc. Some are seeing AI as a replacement for public servants. The Aotearoa New Zealand government plans to cut nigh on 9000 public servants to be replaced by AI, although it has struggled to articulate what functions AI will replace. In practice, most artificial/automated decision making happens at the implementation stage of policy based on human policy decisions (and we’ve seen the perils of this – Australia’s Robodebt anyone?).

Apolitical’s AI navigator is a repository of how AI is being used by government and the public sector across the world. Most of the policy related examples are not about AI policy or AI as policy. But they offer insights into AI in policy, how AI can enhance policy processes. In this article, drawing on some of these cases and others, I use my 5D policy model to examine how AI can be used in policy design and delivery - from understanding the demand for policy solutions, to advising decision-makers, to monitoring implementation and policy results – and offer some insights on the challenges of bringing artificial intelligence into the part of government work that arguably requires the most human intelligence.

The 5D policy process model

The 5D model is a repeatable, scalable model to support policy professionals in their day-to-day work. It is designed around 5 critical lines of inquiry that are dynamic, iterative and not necessarily sequential, underpinned by continuous evaluation and engagement (see the video animation). It also provides an anchor for processes such as engagement and evaluation and is a key component of the wider policy infrastructure.

The 5-d Policy advice value chain Figure 1. The 5D policy advice model (credit Sally Washington)

AI could potentially be applied to each D and E (see Table).

5D model key lines of inquiryAI applicationExample
Demand: Where is the demand for change coming from, why now?Scanning political statements/manifestos, media/social media commentary, understanding the authorizing environment and political constraints.UK’s Parlex tool can analyse years of Parliamentary debate contributions as insight into how politicians and Parliament might react to a new policy proposal.
Discover: what do we know about the challenge or opportunity and what do we need to know?Aggregating and synthesizing large volumes of evidence (evaluations, data, research repositories, academic and grey literature), highlighting evidence gaps, and analyze stakeholder views.This is probably the most obvious place for AI assistance. It is well documented how AI can accomplish tasks which in the past would require significant resources in time and people capability. But humans need to ensure they can robustly reference sources and show the logic and workings.
Design: what methods will we use to design and test solutions and who will be involved?Proposing alternative policy options, instruments, implementation pathways; testing scenarios, likely stakeholder responses or impacts; assessing cost benefit and distributional impacts; testing legislation/regulation consistency & alignment; drafting; submission analysis.Canada’s Regulatory evaluation platform calculates regulatory burden metrics from the text of regulation, including over time.
Decide: how do we ensure we give quality advice to help decision-makers take good decisions?Support for briefings including traceability/provenance of evidence; drafting tailored outputs for diverse audiences (ministers, media, the public); documenting & theming diverse views (submissions, reactions); automating quality assurance processes for policy advice.AI as a reviewer to assess the quality of advice for ministers based on criteria like those in the New Zealand policy quality framework or my policy acid tests.
Deliver: How do we ensure decisions will be implemented and have the desired impact?Implementation planning (operational, budgetary, capability); project management (tracking implementation timelines, indicators, signals and risks).Project and programme management tools could be automated with AI assistance to give real-time progress on delivery, including against any targets or timelines.
Engage: how will we involve, seek insights, and test thinking and options with diverse stakeholders/people who will be affected?Continuous curation of diverse citizen views and experience, enhancing the scale of deliberative processes (accessibility/support, facilitation, communication of process & results).vTaiwan is a digital citizens assembly to collect public opinion, facilitate large-scale public policy conversations and build consensus. The OECD has recently produced a report on AI and citizen participation, including a typology of relevant AI tools.
Evaluate: how will we constantly test, reflect, and iterate throughout the policy process?Testing hypotheses, monitoring risk signals & feedback, curating real time implementation evidence and results and making them accessible for system-wide learning.How might AI help us with policy stewardship, tracking policy settings, and outcomes as well as building institutional memory (‘who knows what’) over time?

AI has the potential to improve the way we do policy, but effective application and adoption will require mitigating the risks and building a supporting infrastructure. There are some challenges to confront.

AI strategy and leadership. At a system level governments need to define an overall strategy and direction of travel for AI use. A recent report for the World Bank documents some 80 countries publishing national AI strategies between 2017 and 2025. Frontrunners like Canada are already on revisions of their initial strategies. In my part of the world, Australia released an AI technology roadmap in 2019 and an AI action plan in 2021, with a framework to regulate the use of automated decision making in government. Whole of government institutional leadership sits in an Office of AI in the Department of the Prime Minister and Cabinet. Aotearoa New Zealand launched an AI strategy in 2025. The Government Digital Delivery Agency and AI system leadership has been centralised into the Public Service Commission, with an AI framework, work programme and toolkit.

Strategies need to be operationalised with effective standards and guardrails so that public servants, especially policy practitioners, understand the boundaries of responsible AI use. This includes alignment with legislation, clear principles for use, and guidance related to transparency of AI generated evidence and reasoning, privacy and data sovereignty issues. The OECD’s Framework for Trustworthy AI in Government provides a range of recommendations on this front. Governments need to consider questions like: are we going to put sensitive administrative data into AI models with foreign ownership or is there a need for bespoke secure models (like France)? What sort of governance is required for citizen-facing AI agents? How do we mitigate and correct the bias inherent in current models (as in everyday life) AI hallucinations and sycophancy and ensure that AI analysis is calibrated to local socio-political environments? Whole of government guidance makes more sense than each agency reinventing the wheel.

Accountability arrangements also need to be clear. From an organisational or team perspective, decision rights for AI and humans need to be explicit. If we think of AI as a member of the team, rather than an autonomous consultant, then AI can challenge the team, add diversity of thought and build collective intelligence, but we need to define who is responsible for what. Humans need to be accountable for policy advice delivered, whether or not they had AI assistance, and be able to face up to legal or administrative review. That means putting effort into building AI literacy, skills and capabilities.

Capability and skills. What new skills are required for policy professionals to move from doing the research and analysis to steering and interpreting the outputs from AI agents, and how do we build them? Apolitical offers support on this front with its government AI campus, AI readiness tool (have a go to test your readiness to use AI) and short free courses, as well as working directly with some governments on AI academies. AI awareness and skills need to be built into regular training courses, including policy courses, rather than just stand-alone offerings that may or may not be relevant to the individual public servant’s work role. Mapping AI applications onto current policy frameworks (such as the 5D, or existing quality assurance frameworks and processes) would make AI training more relevant and useful for policy professionals.

Don't outsource the thinking

The 5D frames good policy processes as iterative, inquiry-driven, evidence informed and relational. AI can potentially become a co-participant in those processes, a partner in inquiry, a curator of evidence, a simulator and tester of policy options, and a monitor of results. The key to good policy advice is curiosity and asking good questions, whether they are AI prompts or asked of other humans. Human policy professionals must be intelligent customers of AI agents. They need to have well-honed judgement, critical thinking and political nous to maintain the long-term trust in the legitimacy of public policy processes. AI will help us make sense of the world, but sense-making remains with us. Policy professionals can’t and shouldn’t outsource the thinking. AI is an important accessory, but it shouldn’t replace our whole wardrobe.


Make sure to share your own thoughts with the author by leaving a comment below