Mixed-initiative generation of virtual worlds: A comparative study on the cognitive load of WFC and HSWFC

Procedural Content Generation (PCG) has long been proposed as an answer to the increased workload imposed on designers of virtual worlds, although often at the cost of controllability and expression of designer intent. Mixed-initiative approaches promise to overcome this, and valid proposals have been made of mixed-initiative editors, e.g. driven by the Wave Function Collapse (WFC) algorithm [12]. However, stock WFC operates on a simple tileset without any hierarchy or semantic clustering, preventing designers from fluently expressing level of detail, leaving the constraint solving burden partly on them instead of on the algorithm. Hierarchical Semantic Wave Function Collapse (HSWFC) claims to solve this problem by hierarchically structuring its tileset of semantic tiles, featuring meta-tiles that can undergo intermediate collapses [2]. Employing an HSWFC interactive editor offering such features may significantly reduce cognitive load and, thus, make such an editor more appealing for widespread use. We describe a user study to test this hypothesis, by comparing and discussing the cognitive load assessed on designers using either a stock WFC-driven editor or an HSWFC-driven editor. Our findings confirm that there is a significant reduction in cognitive load when the HSWFC editor is employed in comparison to the stock WFC editor.