Towards AI Accountability Infrastructure: Gaps and Opportunities in AI Audit Tooling

Explainable AI (XAI)AI Ethics, Fairness & AccountabilityAlgorithmic Transparency & AuditabilityCybersecurity EngineersAI/ML Researchers & EngineersHCI Researchers

Research Background and Issues

  • Issues and Challenges: Although AI auditing is increasingly recognized as a critical means to uncover risks and limitations in deployed AI systems, conducting effective AI audits remains highly challenging. Key obstacles include the lack of standardized methodologies, unreliable results, and auditors' limited access to necessary information.
  • Significance: AI auditing is essential for achieving accountability and transparency in artificial intelligence. However, compared to fields like finance and healthcare, the maturity of the auditing ecosystem in the technology industry is significantly lower, insufficient to support comprehensive governance needs.
  • Research Motivation: Existing tools aim to support AI standards setting and system evaluation, but they often fall short in driving accountability in practice. This study seeks to understand the current tool ecosystem, analyze practitioners' needs, and identify opportunities to expand the scope of tool development.

Solution

  • Methods/Solutions:

    1. Data Collection: Conduct interviews with 35 AI audit practitioners to analyze their tool requirements and challenges in auditing processes.
    2. Tool Analysis: Classify and analyze 435 existing audit tools, constructing a functional stage model for these tools.
    3. Findings and Improvement Directions: Identify limitations and shortcomings of existing tools and provide recommendations to advance tool development from single evaluations to infrastructure supporting comprehensive accountability.
  • Innovations:

    • This study offers a comprehensive landscape analysis of AI audit tools and detailed classification of tool usage stages, ranging from "harm identification" to "audit communication and accountability."
    • Through interviews and literature analysis, it defines which tools can truly meet the needs of accountability practices, rather than merely supporting technical metric calculations.
  • Implementation Steps:

    1. Compile a tool inventory and create preliminary classifications.
    2. Iteratively optimize the classification method based on practitioners' needs.
    3. Evaluate tool coverage at each stage and assess whether they meet accountability goals.

Research Outcomes

  • Specific Outcomes:

    • Proposed a staged model for the AI auditing process, covering the complete pathway from identifying system harms to fostering societal accountability.
    • Found that current tools primarily focus on assessment tools (e.g., performance analysis and standards management), while significant gaps remain in harm identification, transparency infrastructure, audit communication, and advocacy tools.
    • Highlighted key pain points, such as difficulties in data access, lack of standardization, and delays in tool maintenance and updates.
  • Advantages Compared to Existing Solutions:

    • This study emphasizes accountability itself rather than merely technical evaluations.
    • It comprehensively addresses specific needs ranging from external audits to internal audits, and from technical tools to participatory work.
  • Evaluation and Experimental Results:

    • Although 77.9% of existing tools are open-source, many tools (especially those for data transparency and performance analysis) lack practical support for accountability mechanisms.
    • Tool development is heavily concentrated on technical functionalities, with insufficient support for participatory engagement of diverse groups.
  • Limitations and Future Directions:

    • Tool data is primarily sourced from the European and American markets, which may limit the global applicability of the research conclusions.
    • Future efforts should focus on how policies and legal frameworks can support the expansion of the audit tool ecosystem, as well as the feasibility of maintaining long-term quality in open-source tools.
    • Beyond technical development, participatory accountability tools should be designed to support diverse societal groups in collectively governing AI.

Quick Actions

Share

Share this page

ios_share

https://hci.top/en/papers/chi/189574/2025

AdRecommended

Learn AI Coding at CodeNow

open_in_newOpen DOI Link
DOI: https://dl.acm.org/doi/10.1145/3706598.3713301
At a Glance

Paper Snapshot

fact_check
dataset
Source
CHI
calendar_month
Year
2025
emoji_events
Award
No award tagged
group
Authors
5 authors
sell
Subtopics
Explainable AI (XAI), AI Ethics, Fairness & Accountability, Algorithmic Transparency & Auditability
work
Professions
Cybersecurity Engineers, AI/ML Researchers & Engineers, HCI Researchers
article
Content Status
Full text indexed
hub
Related Papers
10 related papers