Thursday, June 15, 2023

AI Map Search

I am trying to keep track of developments in the use of artificial intelligence in mapping. At the moment the most interesting experiments seem to involve using AI to help answer geospatial queries. This is leading to some interesting maps which use AI to map answers to users' queries. 

Over the last few months I've been reading with interest on Twitter how developers have been using AI models (mainly ChatGPT) to create Overpass Turbo queries, to display requested OSM features on an interactive map, and to help make ChatGPT Generated Web Maps

A few developers have already released working interactive maps which make use of AI to generate maps based on queries or prompts.


I know very little about Trident beyond that it bills itself as an "interactive Smart Maps assistant" and that it generates maps based on OpenStreetMap data. This means that Trident is quite good at finding pubs in London or cafes in New York City but can't respond to queries such as 'Show me the four largest mountains in the world". 

I have also had little success searching Trident using distance requests (for example 'show me cafes within 200 yards of subway stations in NYC'). Despite these limitations to the type of queries that Trident appears able to respond to it does seem to produce the most complete responses to point-of-interest type queries. This is because it makes use of OpenStreetMap's huge repository of geographical data.

Texttomap uses ChatGPT to generate maps from user prompts. This means that Texttomap can be very good at answering geographical questions such as 'what are the four longest rivers in America?' and showing the results on a map. 

Texttomap can also map queries such as 'show me pubs in Westminster, London'. However Texttomap doesn't appear to use OpenStreetMap data to generate results. I have therefore found Texttomap results for point-of-interest type queries less complete than those generated by Trident. However it does appear to be very good at answering and mapping geographical queries such as 'What are the ten biggest mountains in the world?"

MapsGPT uses OpenAI to answer geographical questions. It requires users to enter a two part prompt to generate a map of searched places in a defined area. For example you could use MapsGPT to find kid-friendly activities in San Francsico. MapsGPT will then e-mail you a link to a generated interactive map. For example here is the map MapsGPT generated for me when I requested restaurants near Spitalfield's London

Like Texttomap MapsGPT uses an AI to find specific points of interest rather than OpenStreetMap data. Again I suspect that if you want to find a type of point-of-interest within a defined area that you will get more detailed and more complete results using Trident instead.

No comments: