Model Openness Framework: Aiming for Transparency in AI, Curbing ‘Openwashing’

The Model Openness Framework (MOF) is a proposed system that rates machine learning models on their openness and completeness. Developed by researchers from LF AI Data, Generative AI Commons, Linux Foundation, University of Oxford, Columbia University, and IBM, the MOF aims to prevent misrepresentation of models claiming to be open, guide researchers in providing all model components under permissive licenses, and help identify models that can be safely adopted.

The MOF is particularly important for Generative AI (GAI), addressing transparency, reproducibility, bias, and safety concerns. The framework also tackles the issue of openwashing, where models are promoted as open-source but lack an open-source license.

What is the Model Openness Framework (MOF), and Why is it Important?

The Model Openness Framework (MOF) is a proposed system that rates machine learning models based on their completeness and openness. This system was proposed by a team of researchers from LF AI Data, Generative AI Commons, Linux Foundation, University of Oxford, Columbia University, and IBM. The MOF follows principles of open science, open source, open data, and open access. It requires specific components of the model development lifecycle to be included and released under appropriate open licenses.

The MOF aims to prevent misrepresentation of models claiming to be open, guide researchers and developers in providing all model components under permissive licenses, and help companies, academia, and hobbyists identify models that can be safely adopted without restrictions. The researchers believe that wide adoption of the MOF will foster a more open AI ecosystem, accelerating research, innovation, and adoption.

The MOF is particularly important in the context of Generative AI (GAI), which offers unprecedented possibilities but has raised concerns about transparency, reproducibility, bias, and safety. Many open-source GAI models lack the necessary components for full understanding and reproduction, and some use restrictive licenses, a practice known as openwashing.

How Does the MOF Address Concerns About AI?

Artificial intelligence (AI) has seen remarkable advances in recent years, driven by growth in computational capabilities, increased volumes of available training data, and improved deep learning algorithms. However, with this growth in capabilities, so have grown concerns regarding the transparency, reproducibility, ethics, and safety of AI systems.

The MOF addresses these concerns by requiring specific components of the model development lifecycle to be included and released under appropriate open licenses. This ensures that models claiming to be open are indeed open, and that they can be fully understood and reproduced. The MOF also helps to guide researchers and developers in providing all model components under permissive licenses, which can help to alleviate concerns about bias and safety.

The MOF also addresses the issue of openwashing, where models are promoted as open-source but do not utilize an actual open-source license. This can lead to confusion and potential legal consequences for those altering the license and those using the model. The MOF helps to prevent this by providing a clear and consistent framework for rating the openness of models.

What are the Benefits and Risks of Open Models?

Open models offer several benefits over closed source models, including security and performance advantages through distributed development and auditing, adaptability and customization for diverse domains and languages, and advances in the fields of science. However, the openness of models also introduces risks, such as enabling the generation of disinformation or illegal content.

According to one study, open foundational models have five distinctive properties that present both benefits and risks: broader access, greater customizability, local adaptation and inference ability, the inability to rescind model access, and the inability to monitor or moderate model usage. Striking a balance between harnessing the innovation of open models and addressing associated risks remains a critical challenge in navigating the evolving landscape of AI, particularly GAI.

The MOF can help to address these challenges by providing a clear and consistent framework for rating the openness of models. This can help to ensure that the benefits of open models are realized, while the risks are appropriately managed.

How Does the MOF Promote Transparency and Reproducibility in AI?

Many state-of-the-art foundation models are black boxes, making it hard to explain their internal logic or ensure they behave fairly. Models are often released with technical reports and model cards that provide little to no details on the source and treatment of their models’ training data and subsequent fine-tuning, or the methods used for model alignment. Model evaluation results often cannot be reproduced independently due to lack of disclosure of evaluation data and methods, leaving the public to largely rely on claims reported by model producers.

The MOF promotes transparency and reproducibility in AI by requiring specific components of the model development lifecycle to be included and released under appropriate open licenses. This ensures that all necessary information for understanding and reproducing a model is available, and that the model can be independently evaluated.

What is the Future of Open Models and the MOF?

In recent years, countless organizations and individuals have released thousands of open models on online platforms, creating a rich variety of foundation and fine-tuned models available for use. However, while the number of publicly available models has been growing, a concerning trend has emerged: many of these models are being promoted as open-source but do not utilize an actual open-source license.

The MOF aims to address this issue by providing a clear and consistent framework for rating the openness of models. The researchers believe that wide adoption of the MOF will foster a more open AI ecosystem, accelerating research, innovation, and adoption.

However, the future of open models and the MOF will depend on the continued commitment of researchers, developers, companies, and academia to the principles of open science, open source, open data, and open access. It will also depend on the ability of the AI community to navigate the challenges and risks associated with open models, and to strike a balance between harnessing the innovation of open models and addressing associated risks.