AI in GIS without the smoke machine

Where AI may help geospatial work: cleanup, classification, routing, summaries, QA, and decision support without pretending judgment can be outsourced.

This note is a parking lot for the GIS essay track: where AI is already useful, where it is still hype, and where geospatial professionals still need to own the judgment.

The useful questions are practical: Can the tool clean messy data? Can it help classify imagery? Can it summarize field notes? Can it catch bad attributes? Can it support a map product without hallucinating authority?

The standard stays the same: if a model touches a map, the human still owns the decision, the metadata, and the proof.

AI is most credible in the unglamorous parts of the workflow: finding inconsistent attributes, proposing classifications for review, extracting structured details from field notes, creating a first-pass summary, or helping an analyst identify the records that deserve attention. These are accelerators, not authorities.

Geospatial work adds a particular risk: a clean-looking map can conceal uncertain data. Models do not remove that problem. They make provenance, scale, recency, confidence, and human review more important.