Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Collective Recourse for Generative Urban Visualizations

Rashid Mushkani

TL;DR

Diffusion-based urban visualization can amplify group-level harms; the paper proposes collective recourse, a structured pipeline where organized community reports trigger fixes in the model stack and planning artifacts. It formalizes four recourse primitives—counter-prompts, negative prompts, dataset edits, and reward-model tweaks—within a diffusion-stack framework and introduces a mandate score $M(b) = s \cdot (1 - \exp(-n/\tau_n)) \cdot r \cdot q$, with mitigation when $M(b) \ge \tau_M(h)$. A synthetic evaluation on 240 reports shows prompt-level fixes are fastest but least durable, while dataset edits and reward-model tweaks are slower but more durable, achieving 93% precision and 75% recall at a threshold near 0.12, with higher representativeness (rotating juries) boosting recall to ~81% with minimal precision loss. The approach integrates with participatory governance, dashboards, and rotating citizen juries to mitigate risks such as overfitting to vocal groups, while acknowledging limitations of simulation-based evaluation and the need for field deployment.

Abstract

Text-to-image diffusion models help visualize urban futures but can amplify group-level harms. We propose collective recourse: structured community "visual bug reports" that trigger fixes to models and planning workflows. We (1) formalize collective recourse and a practical pipeline (report, triage, fix, verify, closure); (2) situate four recourse primitives within the diffusion stack: counter-prompts, negative prompts, dataset edits, and reward-model tweaks; (3) define mandate thresholds via a mandate score combining severity, volume saturation, representativeness, and evidence; and (4) evaluate a synthetic program of 240 reports. Prompt-level fixes were fastest (median 2.1-3.4 days) but less durable (21-38% recurrence); dataset edits and reward tweaks were slower (13.5 and 21.9 days) yet more durable (12-18% recurrence) with higher planner uptake (30-36%). A threshold of 0.12 yielded 93% precision and 75% recall; increasing representativeness raised recall to 81% with little precision loss. We discuss integration with participatory governance, risks (e.g., overfitting to vocal groups), and safeguards (dashboards, rotating juries).

Collective Recourse for Generative Urban Visualizations

TL;DR

Diffusion-based urban visualization can amplify group-level harms; the paper proposes collective recourse, a structured pipeline where organized community reports trigger fixes in the model stack and planning artifacts. It formalizes four recourse primitives—counter-prompts, negative prompts, dataset edits, and reward-model tweaks—within a diffusion-stack framework and introduces a mandate score , with mitigation when . A synthetic evaluation on 240 reports shows prompt-level fixes are fastest but least durable, while dataset edits and reward-model tweaks are slower but more durable, achieving 93% precision and 75% recall at a threshold near 0.12, with higher representativeness (rotating juries) boosting recall to ~81% with minimal precision loss. The approach integrates with participatory governance, dashboards, and rotating citizen juries to mitigate risks such as overfitting to vocal groups, while acknowledging limitations of simulation-based evaluation and the need for field deployment.

Abstract

Text-to-image diffusion models help visualize urban futures but can amplify group-level harms. We propose collective recourse: structured community "visual bug reports" that trigger fixes to models and planning workflows. We (1) formalize collective recourse and a practical pipeline (report, triage, fix, verify, closure); (2) situate four recourse primitives within the diffusion stack: counter-prompts, negative prompts, dataset edits, and reward-model tweaks; (3) define mandate thresholds via a mandate score combining severity, volume saturation, representativeness, and evidence; and (4) evaluate a synthetic program of 240 reports. Prompt-level fixes were fastest (median 2.1-3.4 days) but less durable (21-38% recurrence); dataset edits and reward tweaks were slower (13.5 and 21.9 days) yet more durable (12-18% recurrence) with higher planner uptake (30-36%). A threshold of 0.12 yielded 93% precision and 75% recall; increasing representativeness raised recall to 81% with little precision loss. We discuss integration with participatory governance, risks (e.g., overfitting to vocal groups), and safeguards (dashboards, rotating juries).

Paper Structure

This paper contains 28 sections, 2 equations, 1 figure, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Research questions.
  3. Contributions.
  4. Background and related work
  5. Generative models in planning and their biases
  6. From individual recourse to collective recourse
  7. Documentation, governance, and the limits of transparency
  8. Red-teaming and third-party oversight
  9. Problem formulation
  10. Recourse primitives in the diffusion stack
  11. Verification on subgroup test sets
  12. Pipeline design
  13. Coordinated advocacy channel with enforcement.
  14. Citizen review and division of labor.
  15. Evaluation: simulation study
  16. ...and 13 more sections

Figures (1)

  • Figure 1: Collective recourse pipeline tying community reports to model fixes and planning workflows. Citizen review and public dashboards operationalize accountability Arnstein1969LadderRaji2020ClosingGap.