Artifacts required or recommended for explainability and oversight.
16 resources
The foundational paper introducing model cards as a framework for documenting machine learning models. Model cards provide standardized documentation covering intended uses, performance metrics across groups, ethical considerations, and limitations.
Hugging Face's guide and templates for creating model cards. It provides practical guidance on documenting model details, intended uses, bias and limitations, training data, and evaluation results in a standardized format.
A framework for documenting datasets used in machine learning. Datasheets answer questions about motivation, composition, collection process, preprocessing, uses, distribution, and maintenance to facilitate responsible data use.
OpenAI's System Card for GPT-4 documents the model's capabilities, limitations, and safety evaluations. It serves as an example of comprehensive AI system documentation covering risk assessment, safety mitigations, and deployment considerations.
Model cards are structured templates that provide standardized overviews of how AI models are designed and evaluated. They serve as key documentation artifacts supporting responsible AI practices by promoting transparency and accountability in model development.
A standardized template for creating model cards that document the performance characteristics and intended use context of machine learning and AI models. The template addresses the lack of standardized documentation procedures for communicating model performance and other relevant information to stakeholders.
A template for creating datasheets for datasets, designed to improve transparency and accountability in AI systems by providing structured documentation of dataset characteristics. The template helps practitioners systematically document dataset creation, composition, collection processes, and recommended uses.
A LaTeX template for creating datasheets for datasets, based on the academic paper 'Datasheets for Datasets'. The template helps document dataset motivation, composition, collection process, and recommended uses to improve transparency and accountability in AI systems.
Red Hat introduces AI system cards as a framework for transparently documenting AI deployments beyond just the models themselves. The approach covers architecture diagrams, constituent models, training data sources, evaluation benchmarks, and security fixes to enable community inspection and improve AI system governance.
An educational article explaining key aspects of AI model cards, which are documents that provide transparency about AI models' creation, deployment, and characteristics. The resource discusses different approaches taken by companies like Google and IBM to implement model documentation practices.
Guidance document for UK public sector organizations on implementing the Algorithmic Transparency Recording Standard. Provides a standardized template for documenting key information about algorithmic tools used by government bodies to enhance transparency and accountability.
A practical guide to implementing transparency requirements under the EU's Digital Services Act, focusing on algorithmic transparency and content moderation processes. The resource includes free templates to help digital service providers create required transparency reports and statements of reasons.
Wolt's algorithmic transparency report documenting the algorithms used in their delivery platform operations. The report follows government transparency templates and provides structured disclosure of algorithmic systems in use, similar to the City of Amsterdam's algorithm register approach.
This resource outlines frameworks for AI system disclosures, including both confidential reporting to government authorities and public-facing transparency mechanisms. It discusses the concept of AI 'nutritional labels' as standardized, accessible disclosure formats that present key information about AI models in a comparable form.
A template provided by Princeton University for disclosing the use of AI tools in document creation. The template includes standard language indicating AI assistance was used and that content has been human-reviewed and edited.
This report examines state-level AI disclosure requirements for political advertisements and communications across multiple US states. It covers laws passed in California, Florida, Hawaii, Idaho, Indiana, Michigan, New York, Nevada, North Dakota, Oregon, Utah, Washington and Wisconsin that mandate disclosure of AI-generated content including deepfakes in political contexts.