LalGeo Maps is an interactive map which allows users to create map visualizations using natural language queries. For example if you type
"Show US states by population'"
into LalGeo Maps then it will produce a choropleth map of the United States coloring and numbering the states by population size.
This map comes with a legend that says
"Data provided by the U.S. Census Bureau. Accuracy based on official government statistics."
However other visualizations created by LalGeo Maps in my testing come with the warning
"Data generated by a large language model (LLM). It may be outdated, incomplete, or inaccurate. Use with caution."This is true even for other visualizations that could be created from Census Data, for example
"Show me a map of U.S. states by median household income."
Even though census data could have been used by LalGeo Maps to create this visualization the actual map created warns me that the data could be 'outdated, incomplete, or inaccurate'.
Even when the data comes with the 'Data provided by the U.S. Census Bureau', I don't trust the results. For example the result of the query
"Visualize U.S. states by percentage of population over age 65"
is a map supposedly based on U.S. Census data. This map, however, shows California with the highest proportion of its population over 65. A simple check of the Census Bureau Tables reveals this is not remotely true - California has only 15.2% of its population over 65, compared to 21.2% in Florida.
This is obviously the major flaw in using AI to create visualizations. Every use case for a mapped choropleth visualization that I can think requires a verifiable data source, allowing users to assess and evaluate the data collection methodology. Put simply I don't trust the visualizations created by LalGeo Maps - which for me makes its output virtually worthless.
The only possible remaining use of LalGeo Maps is perhaps for joke visualizations, using prompts like 'Create a map of US states by nose size', or 'Show US States by ugliest children'. Even then I would be worried that some users might mistakenly read these maps as factual.
Комментариев нет:
Отправить комментарий