A museum’s AI search starts with the places people know

Volunteers trying an early Australian Museum collection explorer often began with places they knew. Familiar geography gave them a way into unfamiliar specimens. A local connection could come before a scientific name.[1]

A March 2026 preprint describes the resulting Australian Museum Collection Explorer: a conversational search interface and a map covering 1,685,922 specimen records with valid locality data as of 20 February 2026. The records were accessed through the Atlas of Living Australia.[1]

The appeal is easy to understand. A collection can be available online and still feel remote: a visitor must know what to ask for, which name to use and what the result represents. Starting with a place supplies a question people can bring from their own lives. What has been collected near a town they know? What could a dated specimen reveal about that place’s past?

The museum’s Collection Enhancement Project shows the work underneath that encounter. Its digitisation programme includes transferring specimen information into a database, making images and assigning barcodes to specimens and storage locations. Records can include species, collection date and place, publication references and storage details. These are handles for finding evidence again.[2]

A question has to become the right search

The Explorer translates requests into defined lookups for specimen searches, statistics or individual records, returning links to the underlying data. Species, place and time become search conditions.[1]

A useful response therefore has two jobs. It needs to make the result understandable, and it needs to preserve the meaning of the question. A precise number is helpful only when the right organisms, geographic boundary and period have been counted. Following the record link should make those choices visible.

One published example exposes that distinction: the paper’s Christmas-beetle query returns a count for the broader Scarabaeidae family. The museum’s own identification guide places Christmas beetles in the genus Anoplognathus within that family. A family-wide total cannot simply serve as a count of that genus.[1][3]

This is a concrete reason to keep the scientific name alongside the familiar one. The correction changes the population being searched, rather than merely improving the answer’s wording. It also shows the usefulness of an interface that leaves a path back to the catalogue.

What a point on the map represents

A separate Atlas of Living Australia programme illustrates another obstacle. Its Downing Herbarium project includes specimens with detailed written locality descriptions but missing geographic coordinates. Turning those notes into coordinates is part of making the records usable in maps. The broader programme also supports label transcription, imaging and taxonomic checks.[4]

The implication is that an empty patch on a collection map can have several explanations. A specimen may never have been collected there, may await digitisation, or may lack a location that the software can plot. Map density describes the available collection record; interpreting it as the density of living animals would answer a different question.[2][4]

The Explorer’s design study involved 12 DigiVol volunteers using an earlier 59,295-record bird prototype. Ten valued the map and nine wanted specimen images. These were design findings, with no demonstrated public-learning gain for the final application.[1]

The promising educational move is still substantial: let a visitor begin with a place that matters to them, then make it possible to follow the curiosity into a named, dated, retrievable specimen. A successful encounter would leave them with something more specific to investigate than the question they brought in.

Report an error or suggest a correction