What Medicine Taught Me About the Future of Planning

 James Burgess, CEO & Founder, NeuroVecta

“There is a phrase in academic biology – the literature has outpaced the understanding…Standing now at the intersection of data science and spatial planning, I find that challenge surprisingly familiar.“

When the Literature Outpaces the Understanding

There is a phrase in academic biology – the literature has outpaced the understanding. Bioinformatics is built on structured experimental data such as, genomic sequences, gene expression matrices, protein interaction networks. The harder problem is knowing what it means. Generating new hypotheses requires synthesising ideas from a constantly growing body of scientific literature that no single researcher can fully read. My training was largely an education in extracting reliable signal from that unstructured mass of published knowledge and building conclusions that could be defended, cited, and tested.

Standing now at the intersection of data science and spatial planning, I find that challenge surprisingly familiar. Both Fields are grappling with the same fundamental tension, there’s more evidence than any individual can read, and decisions that are too consequential to get wrong. Neither has fully solved it. But working across both, I believe they are stronger for having that conversation together.

The Planning Parallel

Planning decisions affect lives directly and for generations. Where homes are built, how infrastructure is routed, which communities are heard. These are not abstract questions. They shape the lived experience of millions of people.

Yet the evidence underpinning many planning decisions remains fragmented. A Local Plan evidence base may draw on dozens of technical studies: transport assessments, housing needs analyses, environmental impact reports, viability appraisals. A nationally significant infrastructure application may generate tens of thousands of pages of documentation. The synthesis of this material into a defensible decision is an enormous cognitive task.

This is not a skills gap. Planners are experienced professionals exercising sound judgement under significant time and resource pressures. It is an information architecture problem. The volume and complexity of evidence that modern planning decisions require has outgrown the tools available to synthesise it. I have good news, for the first time, technology is genuinely positioned to change that.

AI and the New Synthesis Layer

Large language models (LLMs) represent a step change in what is now computationally possible. These AI systems can read, summarise, and reason across document corpora at a scale no human team could match. A search that might take a planning officer three weeks to conduct manually becomes, in principle, a query that takes seconds.

The technique that makes this practically useful in professional settings is called Retrieval-Augmented Generation, or RAG. Rather than asking an AI to answer from memory, a RAG system retrieves the actual source documents relevant to a query and uses those as the direct basis for its response. The answer is grounded in your documents, not in the AI’s general knowledge.

This distinction matters because of a well-documented limitation known as hallucination. When an LLM hallucinates, it generates a response that sounds authoritative but is factually wrong, and the AI does not know it is wrong. Think of it as sophisticated autocomplete: confidently filling a gap with something plausible rather than something true. In clinical contexts, this is a patient safety risk. In planning, where decisions must be traceable to evidence and withstand legal scrutiny, it carries equal weight.

The Harness Matters as Much as the Model

Working alongside clinicians in bioinformatics, I have been developing tools that pair LLM reasoning with strict evidence sourcing. Every response is anchored to the specific passage in the source document that supports it. The clinician sees not just an answer but the exact text it came from, so they can evaluate and, where necessary, override it. The AI handles scale. The clinician handles interpretation and decision. Neither is sufficient without the other.

This is the design principle that matters most. The software harness around an AI model determines whether it is useful or dangerous in a professional setting. It must enforce citation, surface uncertainty, and keep the expert firmly in the decision seat. The AI reduces the cognitive burden of navigating large document corpora. The professional contextualises, scrutinises, and decides. That boundary must be preserved by design.

A Cross-Sector Opportunity

Working across clinical research and data science has made one thing clear, the methodological challenges facing each field are almost identical. Both involve synthesising large volumes of complex, often contradictory evidence into defensible decisions, under time pressure, with direct accountability to the public.

The opportunity is genuinely collaborative. The scientific community has experience building workflows that keep human experts central in evidence-heavy environments. Planning brings something equally valuable: deep expertise in weighing contested evidence, balancing competing public interests, and making decisions that must hold up to democratic scrutiny. These are strengths that clinical research is only beginning to grapple with as AI enters regulated settings. Each field has something the other needs, and the most useful tools will be designed at that intersection.

A Methodology Worth Building Together

Planning practitioners already operate to the same standards of accountability as any regulated field. The challenge is tooling. Planners, development management officers, architects, urban designers, heritage consultants, and transport specialists all work from overlapping bodies of evidence, yet rarely share infrastructure for navigating it systematically. The opportunity is to build cross-disciplinary tooling that allows every stakeholder to access, interrogate, and apply evidence easily, efficiently, reproducibly, and consistently, regardless of organisational size or technical resource.

Consider what becomes possible when that infrastructure exists, appeal decisions interrogated to understand what makes a neighbourhood truly liveable; environmental research informing how green corridors and tree canopy improve health outcomes; consultation responses synthesised to understand how communities actually move and what they need. The ambition to build beautiful, sustainable, nature-rich places has never been the problem. The bottleneck has been bringing the right evidence to bear at the moment decisions are made. Better tooling makes that genuinely achievable.

The communities that planning serves deserve nothing less.

James Burgess is the CEO and Founder of NeuroVecta. He is a bioinformatician motivated by the societal opportunities arising from the intersection of computer science and biology, and a keen developer focused on the expanding capabilities of LLM technology for parsing unstructured data.