Virtual Imaginative Geographies: Generative AI and the Representation of Landscape Imagery
Geographies · 28.02.2025 · 5 (1) · 9
DOI: 10.3390/geographies5010009 ↗Kurz gefasst
Fachartikel in der Zeitschrift Geographies, online erschienen am 28.02.2025. Verfasst von Ghieth Alkhateeb und 3 weiteren Autor:innen. Die Arbeit wurde bisher 11-mal zitiert (Crossref) und nennt 38 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Estonian Government Scholarship for Doctoral Students—G.A..
Abstract (Originalsprache)
Generative AI (GenAI), particularly text-to-image (TTI) models, is reshaping landscape representation by transforming textual descriptions into visual outputs. However, these models often reinforce biases embedded in their training datasets, shaping how landscapes are perceived and represented. This research examines the biases in GenAI-generated landscape imagery through the lens of Edward Said’s “imaginative geographies”, focusing on how geographic references, cultural archetypes, and methodological factors influence AI outputs. We employed a structured approach to create prompts based on the Global Ecosystem Typology (GET), using Midjourney V6.1 as the primary tool for image generation. We extracted landscape descriptors from GET classifications and structured them into both simple and detailed prompts. The analysis involved comparing AI-generated images to ecological classifications and reference images to assess biases. The findings reveal the following three key types of biases: (1) geographic biases, where certain locations act as semantic triggers, leading to culturally stereotypical portrayals; (2) representation biases, where landscapes frequently depicted in AI training datasets appear with greater accuracy, while underrepresented landscapes are simplified into essentialized visual tropes; and (3) methodological biases, where prompt complexity influences representational accuracy but does not eliminate pre-existing cultural hierarchies. To address the bias challenges, the research presents the following four key recommendations for future research and practice: (1) incorporating finer-scale ecosystem classifications; (2) diversifying training datasets; (3) engaging local communities in participatory data collection; and (4) refining prompt-writing methodologies. These insights contribute to broader discussions on AI bias, emphasizing the necessity of critically evaluating the role of generative models in shaping landscape imaginaries.
Förderung
- Estonian Government Scholarship for Doctoral Students—G.A.
Zitieren
APA
Alkhateeb, G., Storie, J., Bell, S., & Suškevičs, M. (2025). Virtual Imaginative Geographies: Generative AI and the Representation of Landscape Imagery. Geographies, 5(1), 9. https://doi.org/10.3390/geographies5010009
Vancouver
Alkhateeb G, Storie J, Bell S, Suškevičs M. Virtual Imaginative Geographies: Generative AI and the Representation of Landscape Imagery. Geographies. 2025;5(1):9. doi:10.3390/geographies5010009
BibTeX
@article{Alkhateeb2025,
author = {Alkhateeb, Ghieth and Storie, Joanna and Bell, Simon and Suškevičs, Monika},
title = {Virtual Imaginative Geographies: Generative AI and the Representation of Landscape Imagery},
journal = {Geographies},
year = {2025},
volume = {5},
number = {1},
pages = {9},
doi = {10.3390/geographies5010009}
}
EUVIMED