Democratic governments are open and allow the public to voice their opinions on government activities and the use of public authority. We believe this should apply not only to human actions but also to the ways we automate those actions with the help of algorithms. Every citizen should have access to understandable and up-to-date information about how algorithms affect their lives and the grounds on which algorithmic decisions are made. It is no wonder transparency is referred to as the most cited principle for trustworthy AI.
We think it is time to align the actions of our governments to meet this expectation. For this, we introduce the concept of a public AI register as a means for transparency and civic participation in government use of AI. In this paper, we share our work on AI registers and suggest the concept for broader adaptation.
About the register
The AI register is a standardised, searchable and archivable way to document the decisions and assumptions that were made in the process of developing, implementing, managing and ultimately dismantling an algorithm. With this, transparency, and when applicable, explainability, can be given for public debate, independent auditors, and individuals citizens. For civil society, it is a window into the artificial intelligence systems used by a government organisation. Ultimately, we hope it will become a catalyst for meaningful democratic participation and a platform for fostering mutual trust.
Building on the suggestions raised in various AI ethics and human-rights related research, policy papers and existing legal frameworks, we suggest a documentation framework for AI registers that is ready for any government organisation to take into use. With this, we hope not only to bring down the barriers of moving into action but also, to welcome feedback for making it even better.
We suggest government organisations to ensure meaningful transparency via AI registers into the following aspects of all their algorithmic systems: purpose and impacts, accountability, datasets, data processing, non-discrimination, human oversight, risks and their mitigation measures, and explainability. Producing this information, in collaboration with AI technology suppliers and other partners, and publishing it for civil participation, should become a standard procedure across government AI projects.
We welcome our readers to explore how it looks in practice by visiting our public AI registers: City of Amsterdam Algorithm Register and City of Helsinki AI Register. Both registers are enabled by Saidot’s AI transparency platform. We hope they will inspire active civic participation through our AI registers, helping us further align our AI with the values of the people we serve.
Source: Public AI Registers, 2020.