On the morning of May 22nd, at the Map and Location Services Forum of the 2025 Global Spatio Temporal Intelligence Conference (WGDC25), Professor Ai Tinghua from the School of Resources and Environmental Sciences at Wuhan University and an expert in the field of maps+AI delivered a special speech entitled "The Great Transformation of Map Mapping under Generative AI Technology". He pointed out that intelligent map generation under AIGC can be achieved through two routes. The first task is to enhance map mapping technology (upper level integrated development) by using LLM's existing mature system tools, that is, to expand and develop DeepSeek, ChatGPT, Sora, etc. at the application level; The second task is to develop a specialized map language LLM tool using AI basic model algorithms (bottom level development), which is based on map language at the bottom level and develops a large map language model through the development of models such as transformer and diffusion.
In terms of research and development progress, based on task one, an intelligent map generator DoMapAI has been developed, mainly based on the spatialization of large models and the training and reinforcement of mapping models in the fields of place, location, and mapping. Based on Task 2, develop a basic model for map language, and construct a specialized pre training model for maps based on their language features. This can be roughly divided into three processes: Process 1 is tokenization of map language; Process two is corpus pre training and word embedding quantification; Process three is to fine tune the basic model and output intelligent effects. ”
Ai Tinghua said that the "great changes" faced by cartography in the AI era are mainly reflected in two aspects, namely, the "Tao" of generation (Tokenization of map language and MapGPT research) and the "Instrument" of generation (vertical development of LLM integrated agent architecture); At the same time, the professions of map makers are also changing, such as mapping system architects supported by AIGC, generative developers under LLM technology, mapping operation prompt engineers, map sample annotators, and map corpus feeders.
