Algorithms Subprojects

Faculty fellows and research assistants of the ISS’ Algorithms project have begun research to examine the design, understanding, and use of algorithmic systems and big data as it relates to inequality.

Comparing Established Compliance Procedures and Recent Research Proposals for Ensuring Non-Discrimination in Statistical Models

Solon Barocas, Information Science

This project aims to learn what companies using statistical models for employment and credit decisions already do to address concerns with bias and discrimination, how these established procedures compare to the more recent proposals from law and computer science focused on fairness, accountability, and transparency in machine learning, and where practice and research could better inform each other.


Computational Due Process: The Agency of Data Subjects between Compliance and Resistance

Malte Ziewitz, Science & Technology Studies

Understanding algorithmic systems has become a key concern for policy-makers, engineers, and academics. But how do ordinary people make sense of something that is said to be inscrutable? What kind of recourse do they have if they feel misrepresented or mistreated? This project maps, examines, and evaluates the different kinds of recourse that are available to so-called ‘data subjects’ in credit scoring and web search.


The Design and Development of Hiring and Productivity Tools

Ifeoma Ajunwa, Organizational Behavior, Law

Although hiring algorithms and productivity tools have permeated many sectors of the workforce, little is known about the design and development behind such algorithmic tools. I have gained access to two research sites: 1) a developer of hiring algorithms, and 2) a developer of productivity and work surveillance applications. The proposed project involves both ethnographic research, as well as, in-depth structured interviews of the workers at those two sites. This project will thus inform a deeper understanding of ethical issues associated with the development of algorithmic hiring and work productivity tools.


Estimation of Joint Causal Effects from Observational Data for Counterfactual Fairness

Martin Wells, Statistical Science; Ben Baer, Statistical Science; Daniel Gilbert, Statistical Science; David Kent, Statistical Science

Recent work on fairness in machine learning has framed the problem of discrimination that is based on protected attributes in terms of Pearl’s causal reasoning. Causal inference formulates the distinction between associational and causal concepts by noting an associational concept is any relationship that can be defined in terms of a joint distribution of observed variables, and a causal concept is any relationship that cannot be defined from the distribution alone.  Consequently, in order to obtain causal claims from observational data, one needs to make some, strong and possibly untestable, assumptions about the underlying dependence structure defined by an underlying directed acyclic graph (DAG).  In practical applications these DAGS may be high-dimensional. This subproject explores methods for estimation of causal effects from observational data and applies these methods to assessing fairness in machine learning.


Organizing Transparency: Tracing the Regulation of Algorithmic Accountability in NYC

Malte Ziewitz, Science & Technology Studies; Maximilian Heimstädt, Organization Studies, Witten/Herdecke University

On December 18, 2017, the New York City Council unanimously passed a bill that established a task force to examine the city’s ‘automated decision systems’ – systems that significantly impact New Yorkers’ lives by matching students with schools, assessing teacher performance, or detecting Medicaid fraud. In this project, we accompany the legislative process and trace the considerations involved in passing regulation for algorithmic accountability through a mix of interviews and documentary analysis. How do different actors and stakeholders think about the promises and challenges of such regulation? What options are considered, justified, and undermined? What can this process teach us about attempts to ensure accountability in computational systems? We document and analyze this process with the help of recent work in organization studies and science & technology studies.


The Quantified Worker: Law and Technology in the Modern Workplace

Ifeoma Ajunwa, Organizational Behavior, Law

This forthcoming book from Cambridge University Press will examine the role of technology in the workplace and its effects on management practices as moderated by employment, anti-discrimination, and privacy laws.


Strategies of Algorithmic Management among Cultural Workers

Brooke Erin Duffy, Communication; Ifeoma Ajunwa, Organizational Behavior, Law

While algorithmic systems are radically reshaping the production and distribution of media and cultural content across industrial contexts, the impact on independent cultural workers is less understood. This project will draw upon in-depth interviews with self-employed cultural workers to better understand how their “algorithmic imaginaries” (Bucher, 2017), particularly those involving social media, shape their work processes and products.