Contestable Camera Cars: A Speculative Design Exploration of Public AI That Is Open and Responsive to Dispute

Document Title

Contestable Camera Cars: A Speculative Design Exploration of Public AI That Is Open and Responsive to Dispute

Document Information

• Subject Area: Public Artificial Intelligence (AI), Human-Computer Interaction Design, and Ethics
• Keywords: Artificial Intelligence, Automated Decision-Making, Camera Cars, Contestability, Local Government, Public AI, Machine Learning, Urban Sensing, Design Exploration, Futures Studies

Research Background and Issues

• Identified Problems or Challenges:

  1. Local governments are increasingly adopting AI technologies for automated decision-making in public services, but concerns persist about potential social risks, including undermining democratic governance and infringing on individual dignity and autonomy.
  2. Existing research on AI contestability is largely theoretical, lacking practical guidance for specific design contexts.
  3. There is a lack of clear articulation on how to integrate AI into urban democratic governance systems.

• Significance of the Research:

  1. The uncontestability of public AI may lead to opaque decision-making processes, eroding public trust in administrative institutions.
  2. Treating contestability as a means to enhance system transparency and continuous improvement is crucial for safeguarding individual rights and increasing citizen engagement.

• Research Motivation and Related Work: This study uses camera cars (primarily employed for parking enforcement and other urban monitoring tasks) as a case study to explore how contestability mechanisms can improve the acceptability of public AI and the quality of public services. By combining theory and participatory design, the study generates future visions and evaluates municipal officials' feedback on these designs and the challenges of implementation.

Solution

• Proposed Methods or Solutions:

  1. Vision Video: Develop a concept for a future "contestable camera car" and use video to demonstrate how a contestable AI system operates, including risk points, information transparency, and citizen feedback mechanisms.
  2. Methodological Framework: Based on the "Contestable AI Design Framework," incorporating mechanisms such as user participation and data feedback loops.
  3. Interview Analysis: Conduct semi-structured interviews with 17 municipal officials to explore their perspectives on the design and implementation challenges of "contestable camera cars."

• Innovative Contributions:

  1. Integrates contestability design with a specific public AI scenario (urban parking surveillance camera cars), concretizing theoretical applications.
  2. Employs speculative design to discuss the possibilities and limitations of future AI systems, fostering resonance and policy discussions.
  3. Proposes "Five Contestability Loops in Public AI," identifying critical intervention points.

• Implementation Steps:

  1. Draft a design brief to clarify evaluation criteria for the concept video.
  2. Produce a concept video using visual and narrative formats to present the vision.
  3. Conduct interviews with policymakers and AI users (municipal officials), analyze their feedback, and identify research themes.
  4. Apply reflexive thematic analysis to code and conceptualize interview data.

Research Outcomes

• Specific Outcomes:

  1. Video Dissemination: Created a 2-minute concept video using waste detection as an example to showcase contestable design in public AI, emphasizing feedback loops between citizens and the government.
  2. Challenges Summary: Identified three major themes (13 specific challenges) for implementing contestable design in public AI, such as:
     • Limited citizen technical literacy or restricted feedback channels.
     • The need for public AI to better integrate with democratic processes and regulations.
     • Municipal departments requiring enhanced capacity for execution, such as interdisciplinary collaboration and resource optimization.

• Advantages Over Existing Solutions:

  1. Provides a contextualized case study of contestable design, surpassing the limitations of theoretical research.
  2. Constructs a "Five Contestability Loops in Public AI" model, pinpointing critical intervention points for public AI.
  3. Encourages expanding citizen participation to the early stages of system development, rather than limiting it to localized feedback.

• Experimental or Evaluation Results:

  1. Municipal officials generally agreed with the vision of contestable design but highlighted current barriers to implementation, such as organizational structure, resource constraints, and public capacity.
  2. The video was considered effective in sparking discussion but requires further integration with specific regulations and departmental practices.

• Limitations and Future Directions:

  1. Limitations:
     • The sample was limited to municipal officials, excluding ordinary citizens and other stakeholders.
     • The study focused more on AI pilot projects rather than large-scale applications.
  2. Future Directions:
     • Expand the scope of research to include citizens and social organizations, exploring contestable AI design in multi-stakeholder dialogues.
     • Develop more granular guidance strategies to empower frontline designers and policy implementers.
     • Further investigate the second layer of feedback loops (identifying and improving systemic flaws).