Developing best practices for emerging technologies, and making sure that they are put into play, is a challenging and often inaccessible goal for many organizations – especially for complex and unsolved topics such as AI ethics. The process consumes time and resources, often without a blueprint for how to proceed or clear definition of success. The task can be an unreasonable burden for any single individual or organization.
Collectively, however, organizations can pool thought leadership, insights, and resources to develop best practices and implementation tools that can be leveraged by all. Ultimately, through the collective use of these best practices, responsible industry norms are adopted and implemented.
The Partnership on AI (PAI) was founded to serve exactly this purpose: to be a platform where experts from industry, academia and NGO’s can come to contribute their time and ideas – sometimes across competitors – to create thoughtful best practices and resources to implement them in service of AI technologies that benefit people and society.
The ABOUT ML initiative is a product of this process, with a focus on best practices in the documentation for the complete Machine Learning lifecycle. This multiyear initiative brings together creators, users, and people impacted by AI/ML technologies, representing multiple viewpoints and a range of industries and fields in order to develop a resource that will serve as a practical toolkit for implementing transparency for individuals, teams, and organizations around the world. The goal is to collectively design best practices that, when implemented over time, will become new industry norms which support operationalizing the common AI principle of “transparency.”
Our approach takes inspiration from the iterative, multi-stakeholder process by which many internet standards have been developed. Like these processes, we have created a public forum where anybody can contribute to ongoing discussions of related topics, invite public comment on drafts of publications, and are working with an expert Steering Committee to provide guidance on future updates. In order to ensure that we have input from the broad range of constituents affected by AI, we have partnered with the Tech Policy Lab (University of Washington) to incorporate their Diverse Voices process into ABOUT ML on a regular basis. The Diverse Voices process was developed to solicit views and feedback from communities who are often least likely to be consulted in the formation of technology policies that most directly impact them. The process is additionally well-aligned with PAI’s foundational principles of increasing inclusion in technology policy.
With ABOUT ML, PAI is synthesizing research, feedback, and insights into industry guidelines that will better equip us all to create more transparency in how ML systems are built in order to increase algorithmic accountability and reduce unintended harm. Our mutual success depends on seeking and receiving input and contributions from as broad a set of communities as possible.
We encourage you and your organization to join the project by participating in the online discussion (always open) and sharing comments on the version 0 ABOUT ML draft (currently open for public comment until September 14, 2019, and set to reopen in early Jan 2020 as draft v1). More information about the project itself, including timeline, goals, and Steering Committee membership can be found here.