Leticia Izquierdo Garcia, Miguel Ureña-Pliego, Adrian Mora-Carrero, and Kent Larson. 2026. My Digital Twin Walks the City: Decisional Symmetry in Human—Agent Urban Navigation. In Companion Publication of the 2026 ACM Designing Interactive Systems Conference (DIS '26 Companion). Association for Computing Machinery, New York, NY, USA, 194–204. https://doi.org/10.1145/3802974.3809415

Abstract

Personal AI agents are now proposed as proxies for human judgment in participatory design, but do they replicate what people choose, or represent how people reason? We paired 30 participants with personalized AI agents and asked both to navigate the same urban routes using identical street-level imagery, selecting paths based on street-level scores such as safety and walkability. At each intersection, both human and agent chose a route and tagged the visual reasons behind their decision.Our results reveal a "competency-without-comprehension" paradox: while agents achieve a Cohen’s kappa value of 0.377 in final route selection, acting as effective functional proxies, their visual reasoning is difficult to explain given the low Jaccard index of 0.24 between agent and human reasoning tags. By forcing agents to externalize visual tags, we found 49% of decisions share no common criteria with humans, proving AI reaches "right" choices for "wrong" reasons. Specifically, agents systematically underweight safety compared to humans. Matching outcomes is not matching reasoning. Without transparency into how agents decide, participatory planning inherits biases it cannot see.