SAF: Stakeholders' Agreement on Fairness in the Practice of Machine Learning Development

Area of Science:

• Computer Science
• Artificial Intelligence
• Ethics

Background:

• Machine Learning (ML) systems often exhibit biases, leading to unfair outcomes.
• Complete bias mitigation in ML remains a significant challenge.
• Ethical principles like justice and fairness are difficult to implement in ML development.

Purpose of the Study:

• To explain the inherent limitations in completely mitigating bias in Machine Learning.
• To propose an end-to-end methodology for integrating fairness into ML development.
• To facilitate ongoing agreements between ML developers and stakeholders regarding fairness.

Main Methods:

• An iterative, pro-ethical process for ML development is presented.
• The methodology challenges power dynamics in ML fairness decision-making.
• Guidance is provided for identifying, mitigating, and monitoring bias throughout the ML lifecycle.

Main Results:

• The proposed methodology enables ML teams to address bias at every development stage.
• It supports the translation of ethical principles into practical ML development.
• Users receive clear explanations of bias-related trade-offs.

Conclusions:

• Bias mitigation in ML is an ongoing process, not a one-time fix.
• Stakeholder collaboration and transparency are crucial for achieving fairness in ML.
• The methodology offers a practical framework for responsible ML development.