From Friction to Dropout: Visual Multi-Agent Simulation of Adaptive Socratic Friction in Smart Urban Furniture

Condori-Bustincio K.F.; Lizeth Paredes-Cayllahua E.; Bustincio R.W.C.; Viana Lobato Junior W.; Bittencourt L.F. · Proceedings - 2026 22nd International Conference on Distributed Computing in Smart Systems and the Internet of Things, DCOSS-IoT 2026 (2026) · DOI: 10.1109/dcoss-iot69657.2026.00191

Resumen

Generative artificial intelligence is increasingly embedded in smart urban environments through interfaces designed to minimize interaction costs. While this zero-friction paradigm improves operational efficiency, it may also reinforce automation bias and shallow civic deliberation. This paper evaluates an alternative strategy: Socratic cognitive friction in smart urban furniture. We analyze the trade-off between productive reflection and user dropout using a visual multi-agent simulation framework. In this framework, DeepSeek-based generative agents interact with AI-enhanced public furniture under contextual constraints, including time pressure, fatigue, perceived safety, civic motivation, AI trust, and friction tolerance. We executed 480 simulation episodes across four urban scenarios, comparing three policies: zero-friction assistance, fixed Socratic questioning, and adaptive Socratic friction. Results indicate that while fixed Socratic questioning maximizes argument generation (3.81x ratio compared to baseline), it incurs a 45.0% dropout rate. The proposed adaptive policy reduces dropout to 11.25% - a 75% relative reduction compared to the fixed policy - while maintaining robust deliberation by generating 3.24 times more arguments than zero-friction assistance. This study provides a reproducible in silico protocol for stress-testing AI-mediated public interfaces prior to physical deployment.

Instituciones: Universidad Nacional de San Agustín de Arequipa