Geospatial Epistemology in Literary Creativity: A Comparative Study of Human and AI Textual Constructs
Chapter
Publication Date:
2027
abstract:
This study examines the epistemological differences between human and AI-generated texts, emphasizing the role of geospatial perception in literary production. Human-authored texts are grounded in spatial and experiential markers, whereas AI-generated texts lack such contextual grounding. Epistemologically, we draw on the notion of situation space from situation semantics. The logical framework, grounded on semantic system
of situation-based modal logic, supports the idea that human narratives are semantically situated, while AI systems operate through statistical associations devoid of lived spatial experience. To explore these distinctions, the study analyzes sixty-four human-authored and 192 AI-generated texts responding to identical literary prompts through NLP techniques—document embedding (BERT), topic modeling, and sen-timent analysis—combined with statistical and geospatial methods. Quantitative results show that human texts achieved higher cluster coherence (silhouette = 0.62 versus 0.41), stronger neighborhood preservation (0.85 versus 0.72), and greater intra-cluster semantic similarity (cosine = 0.78 versus 0.61). PCA indicated higher variance explained by the first components in human data (67 versus 48%). Visualization through heatmaps and semantic graphs confirmed richer spatial anchoring and thematic diversity in human narratives, while AI texts displayed tighter but less contextually grounded structures. These findings highlight measurable divergences in cognitive modeling between human and AI writing, contributing to the debate on AI’s creative and geospatial limitations.
of situation-based modal logic, supports the idea that human narratives are semantically situated, while AI systems operate through statistical associations devoid of lived spatial experience. To explore these distinctions, the study analyzes sixty-four human-authored and 192 AI-generated texts responding to identical literary prompts through NLP techniques—document embedding (BERT), topic modeling, and sen-timent analysis—combined with statistical and geospatial methods. Quantitative results show that human texts achieved higher cluster coherence (silhouette = 0.62 versus 0.41), stronger neighborhood preservation (0.85 versus 0.72), and greater intra-cluster semantic similarity (cosine = 0.78 versus 0.61). PCA indicated higher variance explained by the first components in human data (67 versus 48%). Visualization through heatmaps and semantic graphs confirmed richer spatial anchoring and thematic diversity in human narratives, while AI texts displayed tighter but less contextually grounded structures. These findings highlight measurable divergences in cognitive modeling between human and AI writing, contributing to the debate on AI’s creative and geospatial limitations.
Iris type:
2.1 Contributo in volume (Capitolo o Saggio)
Keywords:
Artificial intelligence, Geographical epistemology, Modal logic, Humanistic geography, Statistical analysis
List of contributors:
Pascale, Francesco De; Cuconato, Simone; Bona, Matteo; Ferrari, Donato
Book title:
Proceedings of International Conference on Information Technology and Intelligence. ICITI 2025.
Published in: