Whose eyes on the street? Generative AI and a discrete choice experiment on how sociodemographic presence shapes urban safety perception in Seoul

Le résumé fourni par la source

Perceived safety shapes how people walk, linger, and inhabit cities, yet decades after Jane Jacobs’ “eyes on the street” the evidence isolating human presence from confounding environmental cues remains scarce. We put that premise to a direct experimental test, combining generative AI image editing with a within-pair discrete choice experiment. Using Google Gemini to insert pedestrians rendered to appear Korean into 30 Seoul street scenes under a factorial design (0–5 people; child / adult / elderly; male / female / mixed), we administered 12 pairwise safety comparisons each to 1,174 quota-recruited residents and fit a hierarchical Bayesian discrete choice model with respondent random effects and rater-by-scene interactions. A two-person mixed-age, mixed-gender pair was judged safer than the empty street about 64% of the time (versus 50% by chance), and a five-person group about 73% ( ˆβ = 0.58 and 1.00 log-odds); this boost is robust across land-use contexts and quality-control regimes. Female presence reads as a stronger safety signal than male presence at every group size and for both rater genders, while a single male reads as no safer than an empty street. Observer heterogeneity concentrates on mixed-demographic scenes: female raters read them as somewhat safer, and elderly raters as somewhat less safe, than their counterparts (+0.23 and −0.22 log-odds; P ≥ 0.95 ), whereas rater effects in homogeneous scenes are far smaller. Our results provide quasi-experimental evidence that who is present shapes safety perception as much as how many, with implications for urban design that invites diverse co-presence rather than filtering it.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.