The Model Openness Framework: Promoting Completeness and Openness for Reproducibility, Transparency, and Usability in Artificial Intelligence

Matt White, Cailean Osborne, Xiao-Yang Liu, Keyi Wang, Sachin Mathew Varghese

Keywords: Artificial intelligence, Machine learning, Open science, Open source software, Open data

Abstract: Generative artificial intelligence (GenAI) offers numerous opportunities for research and innovation, but concerns have been raised about the reproducibility, transparency, and safety of frontier AI models. Many ``open-source" GenAI models lack the necessary components for full understanding, auditing, and reproducibility, while some models use restrictive licenses, a practice known as ``openwashing". In this paper, we propose the Model Openness Framework (MOF), a three-tier ranked classification system that rates machine learning models based on their completeness and openness. Each MOF class specifies the code, data, and documentation components in the model development lifecycle that should be released under certain open licenses. We develop the Model Openness Tool (MOT) to provide a user-friendly reference implementation to evaluate models' openness and completeness against the MOF. We launched the Open MDW License recently, which is the first permissive open license for AI models. The MOF aims to establish completeness and openness as core tenets of responsible AI research and development, and to promote best practices in the burgeoning open AI ecosystem.

Submission Number: 60