Overview
New research suggests that algorithmic decision-making using ?big data? may increase unjust discrimination, unfairness, and structural inequality in society. ?Black-boxed? automated systems in finance, media, information, transportation, and/or any application of computing can easily create outcomes that are unforeseeable by their designers; indeed, well-known algorithmic platforms operated by Google, Facebook, airbnb, and Uber have recently encountered well-publicized problems. Experts writing in Science, Nature, and elsewhere have called for a research response to this situation.
This is a proposal for a workshop at the University of Michigan that will focus on algorithm auditing, a new research design that has shown promise in diagnosing these unwanted consequences of algorithmic systems. Auditing in this sense takes its name from the social scientific ?audit study? where one feature is manipulated in a field experiment. Researchers are now applying this methodology to cyber-human systems, often in teams containing social scientists and computer scientists working together.
In a situation where a secret computational process can produce undesirable societal consequences, whether these are intended or not, scholars imagine a field of research that would allow a third party to examine and diagnose bias in an algorithmic system?s outputs. This workshop proposes to coalesce this new area of inquiry and to produce a report characterizing the state of the art and potential future directions.
The proposed workshop spans academia and industry and will include a diverse group of both junior and senior researchers. Participants will have opportunities to articulate challenges that they face, to present existing methods for auditing, and to propose research agendas that can provide new insights that advance science and benefit society.

Intellectual Merit
The purpose of the workshop is to coordinate a new research community that crosses a variety of scholarly populations that do not regularly interact but who have important insights that they can offer to one another. These are both social and computational researchers, and more specifically they will include experts in human-computer interaction, interface design, machine learning, data mining, social stratification, racial/gender/and other inequality, organizational behavior, technology law and policy, audit study methodology, research ethics, science and technology studies, information theory, and game theory.
This topic addresses basic scientific questions. Auditors require proofs that can ensure that the system being studied will not be perturbed, assurances that third-party audits impose negligible costs, techniques to certify a system as free from bias, and rigorous quantification of bias. Auditors are also playing a cat-and-mouse game with platforms who may be intentionally hiding a discriminatory decision-making process, implicating game theory and information theory.
Unwanted algorithmic discrimination is part of the study of social structure itself. Algorithmic systems are a new dimension of the maintenance and transmission of social stratification and this permit new insights into cumulative disadvantage. Methodologically, audits must confront difficult basic questions of fairness, private property, and research ethics. This also centrally engages the concerns of social researchers who have long been investigating with the introduction, adoption, and development of technological systems. How humans think about complex algorithms is a fundamental problem in human-computer interaction and cognitive science.

Broader Impacts
The kind of algorithmic systems to be examined at this workshop involve data about almost every person in the US. The social problems addressed by this workshop are key economic and social impediments to the adoption and development of fundamental advances in information and communication technologies (ICTs). Algorithm auditi