Participatory Approaches to Machine Learning
ICML 2020 Workshop (July 17)
ICML Virtual Site:
icml.cc/virtual/2020/workshop/5720
Discord server:
https://discord.gg/KSAwXKs
Twitter hashtag:
#PAML2020
Citing the workshop:
@misc{paml,
author={Kulynych, Bogdan and
Madras, David and
Milli, Smitha and
Raji, Inioluwa Deborah and
Zhou, Angela, and
Zemel, Richard},
title={Participatory Approaches to Machine Learning},
howpublished={International Conference on Machine Learning Workshop},
month=July,
year=2020
}
Overview
The designers of a machine learning (ML) system typically have far more power
over the system than the individuals who are ultimately impacted by the system
and its decisions. Recommender platforms shape the users’ preferences; the
individuals classified by a model often do not have means to contest a decision;
and the data required by supervised ML systems necessitates that the privacy and
labour of many yield to the design choices of a few.
The fields of algorithmic fairness and human-centered ML often focus on
centralized solutions, lending increasing power to system designers and
operators, and less to users and affected populations. In response to the
growing social-science critique of the power imbalance present in the research,
design, and deployment of ML systems, we wish to consider a new set of technical
formulations for the ML community on the subject of more democratic,
cooperative, and participatory ML systems.
Our workshop aims to explore methods that, by design, enable and encourage the
perspectives of those impacted by an ML system to shape the system and its
decisions. By involving affected populations in shaping the goals of the overall
system, we hope to move beyond just tools for enabling human participation and
progress towards a redesign of power dynamics in ML systems.
Topics
We invite work that makes progress on the workshop themes, including but not
limited to:
-
Recourse and reachability: Creating systems that ensure the user
has the potential to adapt their predictions or recommendations
(Ustun, Spangher, and Liu 2019;
Dean, Rich, and Recht 2020).
-
Richer interactive ML: Incorporating richer user feedback into
ML systems. Limitations of current methods in fully capturing a
user’s ‘preferences’
(Schnabel, Bennett, and Joachims 2018;
Yang et al. 2019)
-
Mechanism design or computational social choice and learning:
designs that incorporate preference elicitation in allocative ML
systems, such as adaptive experimentation or learning-based
algorithmic governance
(Narita 2019;
M. K. Lee et al. 2019;
Kahng et al. 2019)
-
Collective and participatory design applied to community
involvement in ML: Qualitative and quantitative methods and
frameworks aimed at giving a voice to communities affected by ML
systems. Tools for collective training processes and collective
design of ML systems
(Katell et al. 2020;
Brown et al. 2019;
Halfaker and Geiger 2019;
Patton et al. 2020).
-
Deferral and/or abstention: Classification frameworks which
incorporate deferral to humans, or at least a non-classification
outcome.
(Madras, Pitassi, and Zemel 2018;
El-Yaniv and Wiener 2010).
-
Documentation & audit methods: Work which is intended to inform
and engage users in model design, development or deployment
processes and its limitations
(Raji et al. 2020;
Mitchell et al. 2019).
-
Contestation: Technological methods and tools for
analyzing/protesting/contesting the outcomes of ML systems in the
absence of centralized cooperation
(Kulynych & Overdorf et al. 2020).
-
Analysis or audits of amplifiers of systemic injustice in decision systems
(Arnold et al. 2020;
Pierson et al. 2020;
Obermeyer et al. 2019;
Ali & Sapieżyński et al. 2019;
Raji and Buolamwini 2019;
Buolamwini and Gebru 2018).
-
Tools that support community and worker organization
(Matias and Mou 2018;
Irani and Silberman 2013;
Salehi et al. 2015).
Schedule
The central part of our workshop is the livestream on the ICML virtual
site. The livestream will broadcast invited talks and
panels, as well as the latest announcements and possible schedule changes. The interactive
discussions with other participants—poster and breakout sessions—will happen on our Discord
server.
Opening RemarksLive talk (Iivestream)
1:00 PM - 1:15 PM UTC
· Organizing committee
AI’s Contradiction: King’s Radical Revolution in ValuesLive talk (livestream)
1:15 PM - 1:45 PM UTC
· Tawana Petty
Abstract
Video (SlidesLive)
Dr. King called for a radical revolution of values in 1967. He understood that if we did not "begin the shift from a thing-oriented society to a person-oriented society," and prioritize people over machines, computers and profit motives, we would be unable to undo the harms of racism, extreme materialism, and militarism. If we were to take Dr. King's challenge seriously today, how might we deepen our questions, intervene in harmful technologies and slow down innovation for innovation's sake?
What does it mean for ML to be trustworthy?Talk (livestream)
1:45 PM - 2:15 PM UTC
· Nicolas Papernot
Abstract
Video (Youtube)
The attack surface of machine learning is large: training data can be poisoned, predictions manipulated using adversarial examples, models exploited to reveal sensitive information contained in training data, etc. This is in large parts due to the absence of security and privacy considerations in the design of ML algorithms. Yet, adversaries have clear incentives to target these systems. Thus, there is a need to ensure that computer systems that rely on ML are trustworthy.
Fortunately, we are at a turning point where ML is still being adopted, which creates a rare opportunity to address the shortcomings of the technology before it is widely deployed. Designing secure ML requires that we have a solid understanding as to what we expect legitimate model behavior to look like. We structure our discussion around three directions, which we believe are likely to lead to significant progress.
The first encompasses a spectrum of approaches to verification and admission control, which is a prerequisite to enable fail-safe defaults in machine learning systems. The second seeks to design mechanisms for assembling reliable records of compromise that would help understand the degree to which vulnerabilities are exploited by adversaries, as well as favor psychological acceptability of machine learning applications. The third pursues formal frameworks for security and privacy in machine learning, which we argue should strive to align machine learning goals such as generalization with security and privacy desiderata like robustness or privacy. We illustrate these directions with recent work on model extraction, privacy-preserving ML and machine unlearning.
Turning the tables on Facebook: How we audit Facebook using their own marketing toolsTalk (livestream)
2:15 PM - 2:45 PM UTC
· Piotr Sapieżyński
Abstract
Video (SlidesLive)
Researchers and journalists have found many ways that advertisers can target—or exclude—particular groups of users seeing their ads on Facebook, comparatively little attention has been paid to the implications of the platform's ad delivery process, where the platform decides which users see which ads. In this talk I will show how we audit Facebook's delivery algorithms for potential gender and race discrimination using Facebook's own tools tools designed to assist advertisers. Following these methods we find that Facebook delivers different job ads to men and women as well as white and Black users, despite inclusive targeting. We also identify how Facebook contributes to creating opinion filter bubbles by steering political ads towards users who already agree with their content.
Poster Session 1Poster session (Discord)
2:45 PM - 3:30 PM UTC
Breakout Sessions 1 / BreakBreakout discussions (Discord)
3:30 PM - 4:15 PM UTC
Panel 1Panel (livestream)
4:15 PM - 5:00 PM UTC
· Tawana Petty, Nicolas Papernot, Piotr Sapieżyński, Aleksandra Korolova, Deborah Raji (moderator)
Affected Community Perspectives on Algorithmic Decision-Making in Child Welfare Services Talk (livestream)
5:00 PM - 5:30 PM UTC
· Alexandra Chouldechova
Abstract
Video (SlidesLive)
Algorithmic decision-making systems are increasingly being adopted by government public service agencies. Researchers, policy experts, and civil rights groups have all voiced concerns that such systems are being deployed without adequate consideration of potential harms, disparate impacts, and public accountability practices. Yet little is known about the concerns of those most likely to be affected by these systems. In this talk I will discuss what we learned from a series of workshops conducted to better understand the concerns of affected communities in the context of child welfare services. Through these workshops we learned about the perspectives of families involved in the child welfare system, employees of child welfare agencies, and service providers.
Actionable Recourse in Machine Learning Talk (livestream)
5:30 PM - 6:00 PM UTC
· Berk Ustun
Abstract
Video (SlidesLive)
Machine learning models are often used to automate decisions that affect consumers: whether to approve a loan, a credit card application or provide insurance. In such tasks, consumers should have the ability to change the decision of the model. When a consumer is denied a loan by a credit score, for example, they should be able to alter its input variables in a way that guarantees approval. Otherwise, they will be denied the loan so long as the model is deployed, and – more importantly – lack control over a decision that affects their livelihood. In this talk, I will formally discuss these issues in terms of a notion called recourse -- i.e., the ability of a person to change the decision of a model by altering actionable input variables. I will describe how machine learning models may fail to provide recourse due to standard practices in model development. I will then describe integer programming tools to verify recourse in linear classification models. I will end with a brief discussion on how recourse can facilitate meaningful consumer protection in modern applications of machine learning. This is joint work with Alexander Spangher and Yang Liu.
Beyond Fairness and Ethics: Towards Agency and Shifting PowerTalk (livestream)
6:00 PM - 6:30 PM UTC
· Jamelle Watson-Daniels
Abstract
Video (SlidesLive)
When we consider power imbalances between those who craft ML systems and those most vulnerable to the impacts of those systems, what is often enabling that power is the localization of control in the hands of tech companies and technical experts who consolidate power using claims to perceived scientific objectivity and legal protections of intellectual property. At the same time, there is a legacy in the scientific community of data being wielded as an instrument of oppression, often reinforcing inequality and perpetuating injustice. At Data for Black Lives, we bring together scientists and community-based activists to take collective action using data for fighting bias, building progressive movements, and promoting civic engagement. In the ML community, people often take for granted the initial steps in the process of crafting ML systems that involve data collection, storage and access. Researchers often engage with datasets as if they appeared spontaneously with no social context. One method of moving beyond fairness metrics and generic discussions of ethics to meaningfully shifting agency to the people most marginalized is to stop ignoring the context, construction and implications of the datasets we use for research. I offer two considerations for shifting power in this way: Intentional data narratives and Data trusts - an alternative to current strategies of data governance.
Panel 2Panel (livestream)
6:30 PM - 7:15 PM UTC
· Berk Ustun, Alexandra Chouldechova, Jamelle Watson-Daniels, Deborah Raji (moderator)
Poster Session 2Poster session (Discord)
7:15 PM - 8:00 PM UTC
Breakout Sessions 2Breakout discussions (Discord)
8:00 PM - 8:45 PM UTC
Poster Sessions
Each contributed paper has its own audio/video channel on our Discord server.
During the “poster” session, participants can discuss the paper with its authors in the respective channel.
Before joining a channel as a participant, please make sure to familiarize yourself with the paper
and watch the paper’s short introduction video.
Poster Session 1 (2:45 PM - 3:30 PM UTC)
Poster Session 2 (7:15 PM - 8:00 PM UTC)
Breakout Sessions
On our Discord server, you can chat with other workshop participants at any time in thematic text or
audio/video channels; we will add channels during the day if topics come up. In addition to this, as an
experiment, we will have thematic discussion sessions guided by facilitators:
Breakout Session 1 (3:30 PM - 4:15 PM UTC)
- Community Organization and ML Systems. Facilitated by Seda Gürses (TU Delft)
- Stakeholder Engagement in System Design. Facilitated by Haiyi Zhu (CMU)
- Recourse. Facilitated by Kristen Vaccaro (UIUC)
Breakout Session 2 (8:00 PM - 8:45 PM UTC)
- Community Organization and ML Systems. Facilitated by David Robinson (Cornell)
- Richer Interactive ML. Facilitated by Sarah Dean (Berkeley)
Contributed Papers
The contributed papers include both novel research papers as well as papers that have already been
published in other venues (marked as “Encore”).
Qualitative frameworks, methods, and analyses
More than a label: machine-assisted data interpretation
Maja Trębacz (University of Cambridge); Luke Church (University of Cambridge)
What are you optimizing for? Aligning Recommender Systems with Human Values
Jonathan Stray (Partnership on AI); Steven Adler (Partnership on AI); Dylan Hadfield-Menell (UC Berkeley)
Participatory Problem Formulation for Fairer Machine Learning Through Community Based System Dynamics
Donald Martin, Jr. (Google); Vinodkumar Prabhakaran (Google); Jill Kuhlberg (System Stars); Andrew Smart (Google); William Isaac (DeepMind)
Keeping Designers in the Loop: Communicating Inherent Algorithmic Trade-offs Across Multiple Objectives
Bowen Yu (University of Minnesota); Ye Yuan (University of Minnesota); Loren Terveen (University of Minnesota); Steven Wu (University of Minnesota); Jodi Forlizzi (CMU); Haiyi Zhu (Carnegie Mellon University)
Measuring Non-Expert Comprehension of Machine Learning Fairness Metrics
Debjani Saha (University of Maryland); Candice Schumann (University of Maryland); Duncan C McElfresh (University of Maryland); John P Dickerson (University of Maryland); Michelle Mazurek (University of Maryland); Michael Tschantz (International Computer Science Institute)
Algorithmic approaches and quantitative methods
Interpretable Privacy for Deep Learning Inference
Fatemehsadat Mireshghallah (UC San Diego); Mohammadkazem Taram (UC San Diego); Ali Jalali (Amazon.com); Ahmed Taha Elthakeb (UC San Diego); Dean Tullsen (UC San Diego); Hadi Esmaeilzadeh (UC San Diego)
Metric-Free Individual Fairness in Online Learning
Yahav Bechavod (Hebrew University of Jerusalem); Christopher Jung (University of Pennsylvania); Steven Wu (University of Minnesota)
Designing Recommender Systems with Reachability in Mind
Sarah Dean (UC Berkeley); Mihaela Curmei (UC Berkeley); Benjamin Recht (UC Berkeley)
Heuristic-Based Weak Learning for Automated Decision-Making
Ryan Steed (Carnegie Mellon University); Benjamin Williams (George Washington University)
Deconstructing the Filter Bubble: User Decision-Making and Recommender Systems
Guy Aridor (Columbia University); Duarte Goncalves (Columbia University); Shan Sikdar (Everquote)
Fairness, Equality, and Power in Algorithmic Decision-Making
Maximilian Kasy (University of Oxford); Rediet Abebe (Harvard University)
Iterative Interactive Reward Learning
Pallavi Koppol (Carnegie Mellon University); Henny Admoni (Carnegie Mellon University); Reid Simmons (Carnegie Mellon University)
Recourse for Humans
Kaivalya Rawal (Harvard University); Himabindu Lakkaraju (Harvard University)
Designing Fairly Fair Classifiers Via Economic Fairness Notions
Safwan Hossain (University of Toronto); Nisarg Shah (University of Toronto); Andjela Mladenovic (Independent Researcher)
Applications
Contestable City Algorithms
Kars Alfrink (Delft University of Technology); Thijs Turel (Amsterdam Institute for Advanced Metropolitan Solutions); Ianus Keller (Delft University of Technology); Neelke Doorn (Delft University of Technology); Gerd Kortuem (Delft University of Technology)
Preference Elicitation and Aggregation to Aid with Patient Triage during the COVID-19 Pandemic
Caroline M Johnston (University of Southern California); Simon Blessenohl (University of Southern California); Phebe Vayanos (University of Southern California)
Soliciting Stakeholders’ Fairness Notions in Child Maltreatment Predictive Systems
Hao-Fei Cheng (University of Minnesota); Paige Bullock (Kenyon College); Alexandra Chouldechova (CMU); Steven Wu (University of Minnesota); Haiyi Zhu (Carnegie Mellon University)
Can Algorithms Help Support Participatory Housing?
Rediet Abebe (Harvard University); Daniel Wu (Immuta)
Social scientists argue that the US may be in the worst housing crisis in decades, with millions of families experiencing homelessness and housing instability each year. Amid this housing crisis, participatory housing systems – such as Community Land Trusts (CLTs) – have emerged as alternative models to expand housing opportunities for families with fewer socioeconomic means.
In this position paper, we examine what role algorithms and platforms can and have played to support efforts for participatory housing. We highlight examples ranging from supporting diverse coalition building to providing a framework for implementing participatory budgeting to identifying opportunities to drive down prohibitive costs. We highlight existing and new opportunities from a perspective that acknowledges and asserts the secondary role that algorithms can and should play in this domain. We close with a discussion on the benefits and harms of algorithmic tools for housing.
Adapting a kidney exchange algorithm to align with human values
Rachel Freedman (UC Berkeley); Jana Schaich Borg (Duke University); Walter Sinnott-Armstrong (Duke University); John P. Dickerson (UMD); Vincent Conitzer (Duke University)
Co-designing a real-time classroom orchestration tool to support teacher–AI complementarity
Kenneth Holstein (Carnegie Mellon University); Bruce M. McLaren (Carnegie Mellon University); Vincent Aleven (Carnegie Mellon University)
Keeping Community in the Loop: Understanding Wikipedia Stakeholder Values for Machine Learning-Based Systems
C. Estelle Smith (University of Minnesota); Bowen Yu University of Minnesota); Anjali Srivastava (University of Minnesota); Aaron Halfaker (Wikimedia Foundation); L. Terveen (University of Minnesota); Haiyi Zhu (Carnegie Mellon University)
Critical examinations
Is Machine Learning Speaking my Language? A Critical Look at the NLP-Pipeline Across 8 Human Languages
Esma Wali (Clarkson University); Yan Chen (Clarkson University); Christopher M Mahoney (Clarkson University); Thomas G Middleton (Clarkson University); Marzieh Babaeianjelodar (Clarkson University); Mariama Njie (Iona College); Jeanna Neefe Matthews (Clarkson University);
Participation is Not a Design Fix for Machine Learning
Mona Sloane (NYU); Emanuel Moss (CUNY Graduate Center); Olaitan Awomolo (Temple University); Laura Forlano (IIT)
What If I Don't Like Any Of The Choices? The Limits of Preference Elicitation for Participatory Algorithm Design
Samantha Robertson (UC Berkeley); Niloufar Salehi (UC Berkeley)
Bringing the People Back In: Contesting Benchmark Machine Learning Datasets
Emily Denton (Google); Alex Hanna (Google); Razvan Amironesei (USF); Andrew Smart (Google); Hilary Nicole (Google); Morgan Scheuerman (Google)
The Hidden Assumptions Behind Counterfactual Explanations and Principal Reasons
Solon Barocas (Cornell University); Andrew Selbst (UCLA School of Law); Manish Raghavan (Cornell)
Policy
A Review of the UK-ICO’s Draft Guidance on the AI Auditing Framework
Emre Kazim (University College London); Adriano Koshiyama (University College London)
Invited Speakers
Call for Participation
We would like to experiment with breakout discussions to encourage group discussions around shared
topics of interest. Please fill out our this form if you are
interested in joining or organizing such a discussion.
Call for Papers
The workshop will include contributed papers. All accepted papers will be
allocated either a virtual poster presentation, or a virtual talk slot. We will
not publish proceedings, but will optionally link the papers and talk recordings
on the workshop website.
We invite submissions in two tracks:
-
Research Track. Full papers, works-in-progress, position papers, and case studies. We expect
that these submissions introduce novel ideas or results.
The papers should have up to 4 pages (excluding references, acknowledgements, or appendices), and
be formatted using the ICML submission
template. Papers should be
anonymized.
-
Encore Track. Papers that have already been accepted at other venues.
There are no format requirements for this track. The papers should be accepted
at another recognized archival conference or journal, and be submitted by one
of the paper’s authors.
Submissions are currently closed.
For any questions, please send us an email to participatory-ml-organizers
at lists.mayfirst.org
Timeline
- Submission deadline:
June 22, 2020, AoE
- Notification:
July 1, 2020
- Workshop: July 17, 2020
Organizing Committee
Funding assistance
We are grateful to Open Philanthropy
for providing funding assistance for the workshop. If you want to participate in the workshop and
require funding, please write us an email to participatory-ml-organizers
at lists.mayfirst.org