Graphonauts: Introducing a Hybrid Human-AI Pipeline for Text-to-Game Transformation via Knowledge Graphs

"Graphonauts" presents a hybrid human-AI pipeline for transforming dense scholarly texts into playable games. A domain expert and Claude Sonnet 4.5 co-construct a compact knowledge graph from Malinowski's Argonauts of the Western Pacific, and a deterministic tool then renders that graph as a 2D platformer in which nodes become platforms, relation types map to movement mechanics, and a "dig" tool reveals deeper source material. The authors contend that expert oversight is necessary because ethnographic prose resists full automation, and they frame the resulting experience as "experiential hermeneutics," combining epistemic, kinaesthetic, and interpretive pleasures.

The prototype's evaluation is autoethnographic and limited in scope. A 20-node graph was tested in two-participant sessions of roughly five minutes each, supplemented by reflective notes. Players reported that traversal supported rapid conceptual orientation, that the dig mechanic encouraged deliberate contemplation, and that spatial distance between concepts prompted new questions about how ideas relate. The authors also identify risks of oversimplification, ontological flattening, and accessibility barriers, and they acknowledge the absence of formal user or community validation.

The paper offers no quantitative results, no benchmark data on extraction accuracy, and no comparison against alternative methods. Its findings are qualitative, self-reported, and explicitly exploratory, and they are not generalizable beyond this single text and this pair of evaluators. The primary contribution is therefore methodological: a proposed framework for knowledge-graph-driven game design in the digital humanities, supported by an appendix detailing the extraction prompt and a seven-type entity taxonomy with six relation types, rather than an empirically established claim about the approach's effectiveness.