Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
125 tokens/sec
GPT-4o
53 tokens/sec
Gemini 2.5 Pro Pro
42 tokens/sec
o3 Pro
4 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Participatory Problem Formulation for Fairer Machine Learning Through Community Based System Dynamics (2005.07572v3)

Published 15 May 2020 in cs.CY, cs.LG, and stat.ML

Abstract: Recent research on algorithmic fairness has highlighted that the problem formulation phase of ML system development can be a key source of bias that has significant downstream impacts on ML system fairness outcomes. However, very little attention has been paid to methods for improving the fairness efficacy of this critical phase of ML system development. Current practice neither accounts for the dynamic complexity of high-stakes domains nor incorporates the perspectives of vulnerable stakeholders. In this paper we introduce community based system dynamics (CBSD) as an approach to enable the participation of typically excluded stakeholders in the problem formulation phase of the ML system development process and facilitate the deep problem understanding required to mitigate bias during this crucial stage.

Citations (61)

Summary

  • The paper introduces a participatory method integrating Community-Based System Dynamics to mitigate bias during ML problem formulation.
  • It employs visual tools like causal loop diagrams to transparently model dynamic societal complexities and stakeholder perspectives.
  • The study demonstrates that engaging marginalized voices early enhances fairness and leads to more robust and equitable ML systems.

Participatory Problem Formulation for Fairer Machine Learning Through Community-Based System Dynamics

The paper presents an innovative approach to improving fairness in ML by addressing a critical yet often overlooked stage in ML system development: problem formulation. Traditionally, this phase involves crafting a strategic goal and translating it into ML problems. However, the authors highlight that biases introduced here can lead to significant unfairness in system outcomes, particularly impacting marginalized communities. As an alternative, they propose the integration of Community-Based System Dynamics (CBSD) to engage stakeholders, especially those typically excluded from these discussions, in the problem formulation stage.

Central to the argument is that current methods fail by not adequately capturing the dynamic complexities of society and by excluding essential stakeholder perspectives. For example, high-stakes domains such as healthcare and criminal justice have often been influenced by implicit biases built into ML algorithms, resulting in unfair outcomes, such as racial bias in risk assessments. A cited paper (Obermeyer et al.) illustrates this point by analyzing how healthcare costs were used as an inadequate proxy for health needs, leading to systematic bias against African-Americans—a consequence of neglecting socio-economic disparities in healthcare access.

The authors explore two major pitfalls in current problem formulation practices: a lack of recognition for the dynamic complexity of real-world issues and the exclusion of diverse stakeholder experiences. These gaps result in ML systems that inadequately address the systemic biases prevalent within problem domains.

CBSD offers a participatory framework that involves stakeholders in developing causal models of complex systems. Unlike other modeling techniques such as Structural Causal Models or Causal Bayesian Networks, CBSD is rooted in system dynamics. This approach emphasizes feedback and the dynamic interplay of variables over time, allowing for a comprehensive understanding of societal and technological complexities. The unique feature of CBSD is its participatory nature, which facilitates the inclusion of marginalized voices, thus offering richer, more accurate problem formulations.

The methodology employs visual tools such as causal loop diagrams and stock and flow diagrams, making the modeling process more transparent and accessible. These diagrams help illustrate and quantify variable interactions and feedback loops, providing stakeholders with a platform to develop deeper insights into causal structures and potential impacts of proposed interventions. CBSD encourages a culture of collaboration, where diverse stakeholders can iteratively refine causal models, critically reflect on assumptions, and propose equitable solutions.

Furthermore, CBSD has successfully been applied in various domains such as public health and social work, proving its potential for impact. The paper argues for extending these practices to the field of ML, particularly for fairness interventions. By centering discussions on causal inferences and their potential biases, CBSD aims to foster a formalized approach to problem formulation that is both inclusive and cognizant of long-term impacts.

In conclusion, this work underscores the necessity of revisiting how problems are formulated in ML contexts. By incorporating CBSD, stakeholders can proactively address biases at the earliest stages, thus paving the way for more equitable and robust ML systems. The authors advocate for a shift towards participatory model development that acknowledges and integrates diverse experiences, reflecting a commitment to fairness and ethical ML practices. This approach not only highlights the technical intricacies of causal modeling but also stresses the importance of community engagement and representation in the creation of just ML interventions. Future developments in AI should increasingly focus on embedding such frameworks to ensure equitable technology deployment.

Youtube Logo Streamline Icon: https://streamlinehq.com