Documentation Practices of Artificial Intelligence

TL;DR

This paper surveys the landscape of AI documentation through a quantitative literature review, addressing how data, model, system, and usage cards support transparency and accountability in AI. It finds that data and model cards dominate current practice, while system and usage cards are less common, and that audiences remain heavily skewed toward developers with limited automation in documentation production. The study documents correlations between card type, audience, and multimodality, and highlights opportunities to broaden stakeholder engagement, increase interactivity, and accelerate automation. Overall, the work provides a roadmap for more holistic, accessible, and scalable AI documentation to improve governance, trust, and responsible deployment across diverse contexts.

Abstract

Artificial Intelligence (AI) faces persistent challenges in terms of transparency and accountability, which requires rigorous documentation. Through a literature review on documentation practices, we provide an overview of prevailing trends, persistent issues, and the multifaceted interplay of factors influencing the documentation. Our examination of key characteristics such as scope, target audiences, support for multimodality, and level of automation, highlights a dynamic evolution in documentation practices, underscored by a shift towards a more holistic, engaging, and automated documentation.

Figures (4)

• Figure 1: Temporal dynamics and distribution of scope.
• Figure 2: Distributions of the dimensionality and target audience.
• Figure 3: Distributions of the multimodality and level of automation.
• Figure 4: Correlations between characteristics of documentation checklists.