Skip to main content

Artificial intelligence and GIS

AI changes GIS because it can extract, infer and generate spatial information quickly. It can find candidate place names in reports, classify imagery, detect landscape features, write code and combine clues that previously required a skilled analyst.

The useful questions are technical and practical: what information goes into the model, where is it processed, what new information comes out, how reliable is it, who has authority over the use, and what happens to prompts, files and derived outputs afterwards?

For hands-on desktop workflows, start with Local AI and Māori GIS. Before loading Māori or organisational information, read Before you use AI with Māori information. For the broader kaupapa discussion, see AI and Māori GIS.

Local is one control​

Running a model locally can reduce disclosure to an external AI provider. It does not by itself make the use sovereign, authorised, culturally appropriate, secure or accurate.

A sound approach may also require Māori data governance, organisational security and privacy requirements, software approval, procurement, legal review, research ethics, data-sharing agreements, or guidance from the people with authority or interests in the information.

Readers should consider the current guidance that applies to their own kaupapa and organisation rather than treating this page as a complete approval process.

Public information can still require care​

Public availability is not the same as unrestricted cultural or analytical use. Historical records, Waitangi Tribunal material, planning reports, maps and open datasets may contain whakapapa-linked information, sensitive locations, disputed evidence, approximate geography or kōrero provided for a particular purpose.

AI can make information easier to aggregate, compare and infer. GIS can then turn that result into a precise-looking location. Keep the original source and context, distinguish evidence from interpretation, and consider whether the derived dataset or map changes the sensitivity or meaning of the source.

See Open data and research, Inference risk and Provenance and source checking.

Location inference​

Removing coordinates does not necessarily remove location information. Models can infer place from terrain, vegetation, buildings, signs, text, coastlines and other visual or textual clues.

That matters for photographs, historical descriptions and environmental material as much as formal GIS layers.

See Inference risk.

Feature discovery​

Remote sensing and machine learning can identify possible archaeological forms, vegetation patterns, erosion, land-cover change and other features not supplied as an existing GIS layer.

This can be very useful for taiao and historical research. The output should still be described for what it is: a model result or candidate feature, with method, confidence and source imagery retained.

A precise AI-produced polygon is not automatically a verified feature.

Model training is a separate use​

Uploading material to an AI-enabled service can involve several different kinds of processing:

  • generating the requested output
  • temporary or longer retention
  • safety or quality review
  • product analytics
  • evaluation
  • model improvement or training

These behaviours vary by product, account type and contract, and can change. Check the current service terms and administrative settings for the tool actually being used.

For an organisation handling important Māori spatial information, distinguish using a model to perform the task from allowing the same material to become training or evaluation material for another purpose. Also check what logs, indexes, embeddings and derived copies remain after the visible chat or workspace is removed.

Derived information matters too​

An AI workflow can create a new spatial relationship from sources that were individually ordinary. For example, public imagery plus historical descriptions may produce a ranked set of likely locations.

Record important derived outputs with:

  • model/tool and version where available
  • input sources
  • prompt or method
  • date
  • confidence/uncertainty
  • human/source checking undertaken
  • status such as candidate, checked, rejected or working interpretation

This keeps the whakapapa of the result visible.

Questions for an AI-enabled GIS service​

When the tool will receive project files, imagery or substantial document collections, check:

  1. Where are files and prompts processed?
  2. How long are they retained?
  3. Are they used for shared-model training or evaluation?
  4. Can those uses be disabled for the account or contract?
  5. Are subprocessors involved?
  6. Can service staff review submitted content?
  7. Can prompts, files and generated records be exported or deleted?
  8. What logs remain when the feature is turned off?
  9. What new spatial information can the model infer?
  10. What organisational approval, privacy, security, procurement or governance requirements apply?
  11. Who has authority or interests in the Māori information being processed?
  12. Could the output create a new sensitivity by joining, ranking or locating information?

These questions are useful for any organisation, and particularly relevant where whakapapa-linked, whenua or taiao information is involved.

Local or private inference​

Local models can reduce the amount of working material sent to external AI services. They can be useful for Treaty reports, archive collections, local place-name extraction and other document-heavy mahi.

A local workflow changes where the information is processed. It does not make model output true and it does not decide what a place means.

The Local AI guides show how to keep documents on the desktop, test offline operation, extract candidate ingoa wāhi and retain source references before moving checked results into GIS.

Human and source checking​

AI is particularly good at producing plausible-looking mistakes. Keep AI findings separate from the source evidence until they have been checked.

For place-name research, a useful sequence is:

PDF/report
   ↓
AI candidate extraction
   ↓
page/source check
   ↓
place/source research
   ↓
structured working table
   ↓
GIS

See Find ingoa wāhi and Check the places.

AI for taiao and imagery​

For environmental teams, AI can help classify imagery, prioritise areas for review or find change. Retain the source raster and model/method so a later analyst can understand how the derived layer was produced.

Pair model outputs with Imagery, LiDAR and elevation, Taiao data and field observations rather than treating automated detection as the whole evidence base.

Research beyond this site​

AI governance is changing quickly. MāoriGIS.nz should be one source among several. Check current Māori-led frameworks, official guidance and your organisation's own policies before significant use, and revisit them when the model, software, data collection, purpose or audience changes.

Useful starting points include:

See Further reading and frameworks for a wider research list. New guidance may justify changing the workflow, adding controls, using another tool, reducing what is processed or deciding that some kōrero should not be analysed or mapped with AI.

Sources​

Last reviewed: 2 September 2026

Last updated on