Are We Asking the Right Questions?: Designing for Community Stakeholders’ Interactions with AI in Policing

Title of the Paper

Are We Asking the Right Questions?: Designing for Community Stakeholders’ Interactions with AI in Policing

Paper Information

• Research Domain: Human-Computer Interaction and the application of algorithms in public sectors
• Keywords: Algorithmic crime mapping, human-AI decision support systems, problem modeling, public sector algorithms, community engagement

Research Background and Issues

• What issues or challenges did the authors identify?

  1. Existing algorithmic crime prediction systems may exhibit biases, including racial bias embedded in historical data and biases introduced during manual operations.
  2. Experimental deployments of human-AI decision support systems (ADS) in the public sector lack designs centered around communities and practitioners.
  3. In police departments’ operational practices, the goals of algorithmic crime maps (e.g., hotspot prediction) may not align with law enforcement officers’ focus areas (e.g., regional boundaries and crime relationships).
  4. Both the public and practitioners lack trust in the transparency of algorithmic systems and their data.

• Why is this issue important? Algorithmic crime prediction influences the allocation of public resources and community safety decisions. Its biases may further exacerbate social inequalities. Meanwhile, concerns about the fairness and transparency of algorithmic systems in the public sector are growing, necessitating improvements through human-computer interaction design.

• Research Motivation and Related Work

  1. The authors aim to explore how socio-technical interactions can reduce algorithmic bias and enhance the reliability of human-AI collaborative decision-making in public services.
  2. The study builds on prior literature regarding AI transparency, community-centered design, and worker-centric algorithm design, particularly focusing on the use of artificial intelligence in public domains (e.g., child welfare, crime prediction, public education).

Solutions

• What methods or solutions did the authors propose? The authors developed an interactive crime analysis algorithm tool that uses Kernel Density Estimation (KDE) heatmaps to provide crime hotspot data. They conducted experiments and interviews to explore different groups’ usage behaviors and feedback on the tool.

• What are the innovative aspects of this solution?

  1. Using a pre-configured interactive AI tool as a “boundary object” to integrate diverse perspectives from community stakeholders (community members, technical experts, law enforcement officers).
  2. Providing open and adjustable parameters for users to operate, aiming to educate and assist decision-making.
  3. Investigating the implications of multi-stakeholder participatory design for public service AI, offering design guidance for problem modeling and system transparency.

• What are the implementation steps and key technologies used?

  1. Application Design: Developing an interactive tool simulating real-world crime mapping systems, generating hotspot analyses using the KDE algorithm.
  2. Experiment Process:
     • Community participants set parameters, generate hotspot maps, and interpret the data.
     • NASA-TLX model was used to evaluate users’ cognitive load.
     • Community members were grouped by professional background, and their interaction habits and decision-making behaviors were observed separately.
  3. Interview Analysis: Thematic analysis was conducted to extract users’ ethical and design feedback on the tool.

Research Findings

• What specific findings were achieved?

  1. Interaction Patterns: Community members and technical experts tended to adjust parameters and generate different maps to validate results, while law enforcement officers preferred relying on the default initial map and validating it based on experience.
  2. Key Feedback: Community members questioned whether the tool’s purpose aligned with public interests; technical experts emphasized the flexibility and subjectivity of data science methods; law enforcement officers suggested redesigning the tool to better fit their regional boundaries and patrol needs.
  3. Cognitive Expansion: Community participants gained an understanding of the algorithm through tool usage and identified limitations in its data estimation, reflecting both practical applications and potential risks.
  4. Anchoring Bias: Law enforcement officers focused heavily on the primary information provided by the tool, leading to cognitive anchoring phenomena.

• What advantages does it have compared to existing solutions?

  1. Enhances transparency of human-computer interaction tools, educating community members about the strengths and weaknesses of algorithms.
  2. Identifies issues from the perspectives of communities and practitioners, enabling early ethical evaluations to prevent potential societal harm.
  3. Combines law enforcement experience with AI technology, optimizing comprehensive feedback to support decision-making in public sector AI design.

• What are the experimental or evaluation results?

  1. Community members and technical experts demonstrated higher tool exploration enthusiasm in cognitive load tests (NASA-TLX scores), while law enforcement officers showed cognitive depth but lacked initiative in adjusting parameters.
  2. The experiments revealed widespread community demands for transparency and accountability in crime hotspot analysis tools, along with skepticism about fairness in police department practices.

• Limitations and Future Directions

  1. Limitations: The study was constrained by the distribution of urban population samples, with a limited sample size of law enforcement officers, and did not cover a broader range of city types.
  2. Future Directions:
     • Expand the regional and personnel sample size, adjusting the research model for regional differences.
     • Increase comparative studies across multiple systems, such as interaction designs for cross-departmental AI collaborations.
     • Investigate ways to directly involve community members in the early AI design process to promote social equity and ethical transparency.