Foreword

The promise and value of Machine Learning is great, but it has been hastily operationalised over the past decade with often little regard for its wider societal impact, sometimes resulting in harmful and unfair consequences.

We, at The Foundation for Best Practices in Machine Learning, want to help data scientists, governance and management experts, and other machine learning professionals champion ethical and responsible machine learning. We do this through championing our technical and organisational best practices for machine learning, through the free, open-source guidelines you are currently reading.

The aim of these Best Practices is to be easily accessible to anyone working or interested in Machine Learning. This means that they are designed for a large audience who come from a variety of backgrounds and organisations.

At the same time these Best Practices also aim to be complete. Although this means that they can be long at times, please do not be intimidated - read as much or as little as you feel comfortable with and come back later for more. The Best Practices are designed to be adaptable to different organisation sizes, needs, risks, resources, and expected societal impact and so the implementation can be flexible.


Creative Commons Licence

Because we want to lower the barriers to ethical and responsible Machine Learning, our Best Practices have been licensed under the Creative Commons Attribution license. This means they are freely available for commercial and/or private use and/or adaption, subject to attributing (i.e. referencing) The Foundation of Best Practices of Machine Learning of course.


Who are we?

We are a team of seasoned data scientists, machine learning engineers, AI ethicists and governance experts, who are enthusiastic about lowering the barriers for pragmatic ethical and responsible machine learning.


Best regard, The Board of FBPML

May, 2021

Welcome to The Foundation for Best Practices in Machine Learning Wiki!

If you are not familiar yet with the Foundation for Best Practices in Machine Learning, and you want to know more about who we are, what we do, and what the philosophy and vision behind the Best Practices are, please visit our Homepage.

This Wiki portal contains the Best Practices of FBPML. The ones you find here are the same as in our PDF releases, except that community contributions and updates will be available here on the Wiki sooner. This Wiki is also the portal for anyone to contribute to the Best Practice, for providing (and contributing) supporting material, and serves as our community hub.

How can you contribute?

These Best Practices are open-source and rely on community contributions for continuous improvements. To find out how to contribute please have a look at our Contribution Guide.

How to get started with the Best Practices?

Overview and Explanation

On this wiki you will find both the Technical Best Practices and the Organisational Best Practices. The core content of both Best Practices are the subjects you see below.

The Technical Best Practices are scoped for a single product (which includes the ML models) and are aimed at helping your team best develop and maintain this product in an ethical and responsible way. The subjects within the Best Practices are approached through Product Lifecycle phases:

Each subject’s Best Practices are grouped by phase, so that the Risks and Controls are in the same order as you would typically encounter them during the first iteration of your product. Of course, during the lifecycle of your product, you will revisit each phase very often. Therefore, you will revisit the associated Best Practices too.

The Organisation Best Practices are scoped for the entire organisation. It advises how to effectively support product teams within an organisation. This support is clustered around the core subjects mentioned above. These are approached through Policies. Management and governance aspects that are overarching receive attention as well.
Some important definitions
“Controls” and “Aims”

The Best Practices Guidelines are written in a certain format where each “rule” consists of a number, name, Control and Aim.

  • The Control is a actions and can be understood as the instruction. The “what to do”;
  • The Aim is why you should do it, sometimes phrased as a goal, but more often as a risk. The “this is what can happen if you don’t do it” and/or “this is why it is important”.


Product
The Product is our word for the technical system around which the Best Practices revolve. It is used to refer to not only the data, the machine learning model and code, but also every component and process from start to finish that is required to produce the desired effect in practice - from UI to the protocols and processes that embed models in the organization and everything in between.
Where to get Started

User Guides: For tips and advise on how to decide where to start with the implementation of the Best Practices. Check back often, we will be continuously adding new advise.

Organisation Best Practices.

Technical Best Practices.

Other important places

Definitions: For standard definitions used throughout the document.

Special:WikiForum: If you want to discuss the Best Practices in general, we have a forum for that.

Are you wondering what the future of the Best Practices and the Foundation holds? Please have a look at the roadmap.

For frequently asked questions, please visit our website.

Follow us on LinkedIn!