Transforming Policy Creation in the AI Era

In a bold reimagining of governance, the British civil service is facing a seismic shift in its operations due to rapid technological advances, particularly the rise of generative AI and large language models (LLMs). This comes as part of a broader mandate for a complete overhaul of state mechanisms championed by the Prime Minister.

Rewiring Policy Making

The central challenge is not whether AI will reconfigure policy making, but rather how policy makers can leverage these tools to deliver better outcomes for citizens. Although traditional methods still hold value—for instance, officials relying on their refined judgement amidst the complexities of Whitehall—there is no denying that tools such as Redbox are streamlining critical tasks. Such instruments can cut down the time it takes for ministers to understand new topics by synthesizing governmental documents at unprecedented speeds.

Key AI Tools in Use

Policy work across government departments is increasingly augmented by generative AI. Among the notable tools are:

• Redbox: A system that summarizes policy recommendations from submissions and documents, already embraced by over 1,000 users from the Cabinet Office to the Department for Science, Innovation and Technology.
• Consult: A tool that groups and summarizes public consultation responses up to a thousand times faster than its human counterparts, mirroring innovations seen in international counterparts like Singapore.

A live demonstration at the 2024 civil service Policy Festival showcased Redbox analyzing the National Grid’s operational challenges and distilling a complex Ofgem report into succinct policy advice.

The Limitations of LLMs

Despite their impressive capabilities, LLMs have boundaries that policy makers must navigate carefully:

• Accuracy Concerns: Although adept at synthesizing vast amounts of data, LLM outputs can sometimes be inaccurate or even misleading—a phenomenon often referred to as "hallucination."
• Bias and Nuance: These systems may inadvertently mirror biases embedded in their training data and can lack the contextual depth inherent in human experience. For complex sectors like healthcare, insider knowledge and a nuanced understanding of operational realities remain indispensable.
• Standardization vs. Innovation: While LLMs traditionally offer safe, normative responses, they struggle to venture into truly innovative or radical ideas unless guided meticulously by experienced policy makers.

The risk of over-reliance on these tools is significant. Policy makers must be vigilant to ensure that AI-assisted outputs do not inadvertently embed political biases or skew the evidence base available for decision-making.

Redefining the Role of Policy Makers

AI is not set to replace human expertise but instead redefines the role of policy makers. In an AI-augmented environment, their primary responsibilities will be twofold:

1. Editing and Enhancing AI Outputs: Experts will need to critically assess and refine first drafts produced by LLMs—addressing errors, correcting biases, and infusing the nuanced insights drawn from years of experience.

2. Integrative Innovation: Policy makers will combine AI-generated insights with firsthand knowledge and innovative thinking, pushing policy in bold, sometimes radical, new directions. Interactive processes where AI provides feedback on human-crafted ideas will become increasingly commonplace.

By freeing up time previously devoted to routine tasks, these tools allow officials to immerse themselves in real-time, hyper-specific, and frontline data gathering—a dimension that remains out of reach for current technologies.

Upskilling for the Future

A pressing concern in this new landscape is the potential erosion of domain expertise. As LLMs take over the aggregation and synthesis of data, traditional methods of learning through hands-on problem-solving may falter. This paradox underscores the need for the civil service to:

• Preserve Junior Tasks: Maintaining specific tasks for newer officials ensures that hands-on experience and critical domain expertise continue to develop.

• Innovate Training Programs: Exploring new models of gathering insider insights, perhaps through direct frontline experience, can empower the next generation of policy makers to be more resilient and adaptable.

Experimentation through pilot projects, such as 'test and learn' initiatives, offers a pragmatic path forward. Additionally, addressing high turnover and fostering long-term continuity across policy areas will be vital in nurturing the next wave of seasoned policy makers.

Charting the Path Forward

Policy making remains one of the most challenging yet crucial roles in governance. By blending human acumen with the power of LLMs, the civil service can not only improve efficiency but also deliver richer, more context-aware insights that resonate with citizen concerns.

Proactively shaping the integration of AI into policy making is essential. It is not enough to let technology drive change; deliberate and careful management must ensure that the benefits of these innovations are maximized while their risks are mitigated.

Jordan Urban, a senior researcher at the Institute for Government, has outlined these transformative challenges and opportunities in a future-forward vision for the British civil service.