I. Introduction

Our work on “Algorithms and Justice,” as part of the Ethics and Governance of Artificial Intelligence Initiative, explores ways in which government institutions are increasingly using artificial intelligence, algorithms, and machine learning technologies in their decisionmaking processes. That work has been defined (loosely) by the following mission statement:

The Algorithms and Justice track explores ways in which government institutions incorporate artificial intelligence, algorithms, and machine learning technologies into their decisionmaking. Our aim is to help the public and private entities that create such tools, state actors that procure and deploy them, and citizens they impact understand how those tools work. We seek to ensure that algorithmic applications are developed and used with an eye toward improving fairness and efficacy without sacrificing values of accountability and transparency. Our work begins with a focus on the United States, while developing more generalizable lessons and best practices.

What follows is a retrospective from this first year of work, describing research modes and outputs and identifying takeaways. Headlines include the following:

• The Ethics and Governance of Artificial Intelligence Initiative is well-positioned to be a resource for government actors, who are open to expert engagement and input on difficult technical questions. From legislators to judges to state attorneys general, we were pleased that our efforts at outreach with government officials whose work intersects with the use of autonomous technologies were well-received and that state actors are so interested and open to further guidance and input on these issues.
• Stakeholders with an interest in the use and development of algorithmic tools are demanding clear data and information to inform decisionmaking, and academic initiatives like ours are well-suited to become data clearinghouses. After months carefully designing variables and functionality alongside members of the research and advocacy communities, data collection for the risk assessment database is underway for a launch at the end of summer 2018. Our team is already finding new information about how risk assessments are developed and used and uncovering trends across tools and developers that will be both fodder for significant research and a basis for further data collection efforts.
• Procurement is one component of a larger process and must be considered in its broader context. We often speak of the importance of government procurement officers and of ensuring those responsible for purchasing decisions understand the impact of algorithmic technologies. But, procurement is part of a process that extends beyond the point of purchase or licensing and includes assessment of the technical development process, adoption of implementation guidelines, and rigorous testing and review. We have been researching this larger ecosystem over the past year to provide robust resources and guidance to government such that they can consider the context and make smart systems-level decisions.
• The operations of government actors in this space raise some issues that are unique, novel, and discrete and other issues closely connected to concerns raised by the operations of private and commercial actors. It is impossible to completely divorce consideration of government use of AI, algorithms, and machine learning technologies from a broader conversation about fairness, transparency, equality, inclusion, and justice, as they relate to development and use of these tools in the private sector.

We look ahead with an eye toward maximizing impact and engagement with key constituencies as we move our work forward…”