Introduction
AI doesn’t make decisions in a vacuum. It learns from existing data and historical records, many of which reflect the biases and exclusions from the past.
The purpose of datasets is to make decisions: what to measure, how to categorise it, and which perspectives to prioritise. If that data contains gaps or distortions, AI doesn’t inherently correct those gaps. In some cases it can reproduce them, particularly when systems rely heavily on historical or structured datasets.
Modern AI systems – particularly large foundation models – in general are more robust to many forms of bias than earlier machine learning approaches, which were often trained on narrower or more limited datasets. However, where systems rely on structured, historical, or domain-specific data, underlying assumptions can still influence outcomes.
With the increasing ease of integrating AI into systems, there is a growing risk that we accelerate patterns that haven’t been fully examined. Decisions may appear more consistent or efficient, but they may be reinforcing inequalities embedded in the data we trained our AI systems on. This can have the adverse effect of limiting the full value that AI can deliver, particularly in data-driven, decision-making environments.
How mapping historically followed power and investment
Gaps in data and mapping are rarely random. Historically, large-scale mapping was guided by practical priorities: military planning, land ownership, taxation, and infrastructure development. In the UK for example, national mapping efforts were designed to support defence, governance, and economic activity. This meant that areas of strategic or financial importance were mapped first, and often in greater detail.
Over time, this created an uneven foundation.
Regions with higher levels of investment, denser populations, commercial importance, or critical infrastructure – which now come with the extra regulatory pressures – were more likely to be surveyed and updated. Mapping in these areas became more in-depth and reliable, while areas with less economic activity, lower strategic priority or areas that hold temporary infrastructure or developments were updated less frequently and with less precision.
As technology improved over the years, this pattern was amplified as modern GIS systems continued to build on their existing datasets rather than start from scratch. As a result, historical priorities can persist in present-day data, influencing today’s actions.
By understanding the history and the risk of building upon unbalanced data sets, it holds us more accountable to ensure ongoing data is evaluated to accurately shape present-day and future decisions.
Designing for diversity
Diversity in data is a matter of accuracy and fairness. Who is represented and measured influences who is prioritised and receives attention. An inclusive system relies on data being constantly reviewed. Data ages, environments change, populations shift, technology evolves. The reality today in one location could easily not be the same five years on in the same place.
Every person doesn’t exist in the same controlled environment, working to the same routines and ideal outcomes, and only building systems around one scenario can underrepresent the rest and put them at a disadvantage.
An example scenario could be an asset management system for field teams that assumes connectivity is always stable, workflows follow a standard practice, and environments are predictable. In reality, field conditions can throw in unexpected variables, from poor signal to time pressures and unpredictable on-site challenges.
While ideally there would always be strong connectivity and precise GPS positioning, in practice, signal strength can vary significantly, especially when field teams are working in rural areas, dense urban environments, or around underground infrastructure. If systems are designed only for ideal conditions, location accuracy can drop or data capture may fail at critical moments. Field teams work best when systems anticipate this variability, enabling offline workflows and flexible data capture when connectivity is limited.
Variability also exists in how work is carried out. Standardised workflows may not reflect the realities of working in changing weather conditions, complex environments, or time-sensitive situations. What looks like a clear, linear process in a system may not match what’s happening on the ground.
Designing for diversity means looking at the bigger picture and making variation a core requirement. Systems that are designed only for ideal use can slow teams down or fail in urgent situations, while more resilient systems are built to adapt to the less-than-ideal conditions.
Why this matters in geospatial mapping
Geospatial systems carry particular weight because they are often treated as authoritative representations of reality. But maps are not neutral reflections of the world, they have all been curated by information available, and often driven by a particular goal.
If certain communities, assets, or environments are under-mapped, decisions made from that spatial data will reflect those gaps. If historical infrastructure patterns are prominent in datasets, future planning tools may reinforce them. If automation or AI-driven analysis is layered onto incomplete mapping, particularly where systems rely on structured or historical datasets, bias can become embedded in spatial analysis.
As geospatial platforms increasingly incorporate analytics and AI-driven insights, the quality and representation embedded in the spatial data becomes critical.
Therefore, adding the human and revision element is vital in responsible geospatial practice. Regularly validating datasets, being aware of underrepresented environments, reviewing automated outputs, and updating assumptions when conditions change, all work towards a more inclusive system that not only reflects the real world more accurately, but also enables AI to deliver more reliable, effective, and trustworthy outcomes over time.