JADR news post general visual

Call for Papers for the Gold-level Open Access Book: Geography According to ChatGPT

Mutual Insights: Extracting Geographic Insights From Foundation Models While Refining Their Representations of the World

Call for Book Chapters

In 2023, we invited the community to a kick-off session exploring how foundation models such as the large language model GPT or the text-to-image model Midjourney describe and depict geographic space, categorize geographic features, perform in spatial reasoning, and (implicitly) apply principles of spatial data analysis. Our starting assumption was that these foundation models have been mischaracterized as stochastic parrots, thereby implying that they can only repeat (parrot) and thus we cannot learn anything truly new from them about the world. Instead, we argued that foundation models are better understood as distorted mirrors enabling us learn from both their depiction of geographic space and their distortions, e.g., biases. After all, spatial data scientists have plenty of experience with understanding and mitigating distortions, e.g., of cartographic scale. However, one major downside of foundation models is their tendency to hallucinate, i.e., to dream up facts; limiting their usefulness as an instrument (in the tradition of Geographic Information Observatories) for studying geography and grand societal challenges.  With the introduction of of retrieval augmented generation (RAG) and related techniques, this may be changing. In fact, In a 2019 article, we asked provocatively whether our community will be able to develop a chatbot-like autonomous GeoAI system (GeoMachina) that can replace a junior GIS Analyst by 2030. From today’s perspective we may already get there soon with first systems and Geo-foundation models on the horizon.

Scope: For this peer-reviewed chapter book, to be published by IOS PRESS as gold-level open access book within their "Frontiers in Artificial Intelligence and Applications" series, we solicit teams to submit chapters that align with the idea of “mutual Insights”, i.e., how we can learn from foundation models and what we can contribute in return to further improve these models. For instance, how does the world look according to foundation models and how will using them change our own perception? If foundation models encode cultural trends, can we use them for prediction and simulations? Given that biases in foundation models have a clear geo-spatial footprint (e.g., merely 3% of training data for images come from China and India combined), how does this reflect on their ability to depict the full breadth of geographic form? How does the world look like according to foundation models (not limited to ChatGPT)? Should there be global foundation models, or is there a need for more local models?  The rise of Generative AI (GenAI) foreshadows a crisis in trust. Do we want to serve raw foundation models and visualize their (potential) biases or create pre-debiased models, risking that the definition of ‘bias’ itself will become a political tool? What are geographic and spatio(temporal) blindspots in the abilities of foundation models, e.g., with regard to topological reasoning? How can spatially explicit AI/ML research contribute to improving foundation models? Put differently, the book will not be about mere applications of foundation models and GenAI nor will its key focus be on developing entirely new models.

Topics may include (but are not limited to):

1.    Understanding foundation models through a geographic lens

  • Evolution of large language models in understanding and representing geographic concepts
  • Evaluation of the geographic information generated by foundation models
  • Uniqueness/special aspects of geo-foundation models 
  • Foundation model assisted knowledge discovery 

2.    Responsible use of foundation models for geographic applications 

  • Geographic knowledge extraction and spatial reasoning using foundation models
  • Foundation models for cartography and maps
  • Foundation models for earth sciences
  • Foundation models for human mobility and trajectory generation
  • Automating GIS with foundation models 
  • GeoMachina-style autonomous GIS bots

3.    Theory and mathematical foundations underlying geo-foundation models 

  • Diffusion probabilistic models for geospatial data 
  • Mathematical foundations of geographical representations in geo-foundation models

4.    Developing geo-foundation models 

  • Development of spatially explicit foundation models or geo-foundation models
  • Development of multimodal geo-foundation models
  • Development of geo-foundation models with geospatial knowledge graphs

5.    Ethical concerns of using geographic foundation models 

  • Biases and ethical considerations in geo-foundation models
  • Trust in geo-foundation models 
  • Philosophical foundations of geo-foundation models 
  • Benchmarking geo-foundation models 

Submitting Chapter (Abstracts)

Geography according to ChatGPT will consist of both, community solicited, i.e., open call, chapters, and those authored by the volume’s editors, e.g., an introduction, technical background for a general audience, concluding remarks, and so forth. Overall we aim for 10-15 chapters. Each chapter should be between 12-16 pages long, including references, figures, and other materials. Please read the scope statement in the call above carefully. Interested teams are requested to send an abstract of 250 words together with a title and list of authors to krzysztof.janowicz@univie.ac.at beforehand. Please keep in mind that the selection process will not only look at the (quality of the) proposed work, but also try to balance topics, author teams, geographic regions, and so forth. The books sub(title) may still change slightly. 

Expected Timeline

(Please note that in some cases we may require multiple round of reviews. While we aim at a constructive back and forth between reviewers and authors, we may decide not to include your chapter if the reviewer’s feedback is not addressed in time.)

  • Mandatory submission of chapter abstracts: June 14, 2024
  • Submission of selected chapters: September 9, 2024
  • Reviews back to authors: October 31, 2024
  • Final accepted chapters: November 30, 2024 
  • Camera-ready copies: December 20, 2024
  • Open access publication: Early 2025

The Editors

Krzysztof Janowicz, University of Vienna, AT
Ling Cai, ByteDance, US
Gengchen Mai, University of Texas at Austin, US
Lauren Bennett, Esri, US
Rui Zhu, University of Bristol, UK
Song Gao, University of Wisconsin, Madison, US
Yingjie Hu, University at Buffalo, US
Zhangyu Wang, University of California, Santa Barbara, US