A Bias Network Approach (BNA) to Encourage Ethical Reflection Among AI Developers

Area of Science:

• Artificial Intelligence
• Sociotechnical Systems
• AI Ethics

Background:

• Current AI bias literature often treats biases in isolation, leading to developer overload or uncritical acceptance.
• This 'isolationist approach' hinders effective bias management and understanding of their interconnectedness.
• A sociotechnical perspective is needed to address the human and systemic factors in AI bias.

Purpose of the Study:

• Introduce the Bias Network Approach (BNA) as a sociotechnical method for AI developers.
• To provide a framework for identifying, mapping, and relating biases throughout the AI development lifecycle.
• To foster dialogue and critical reflection on AI bias among development teams.

Main Methods:

• Developed the Bias Network Approach (BNA), a sociotechnical framework.
• Utilized graphical representations to depict interconnections between biases.
• Conducted a pilot case study on a healthcare waiting list Natural Language Processing (NLP) model in Chile.
• Involved a small AI developer team and external expert guidance.

Main Results:

• The BNA effectively visualizes interconnected biases and their impacts, aiding ethical reflection.
• Implementation of the BNA promoted greater transparency in AI development decision-making.
• The study highlighted the need to address professional biases and material limitations as significant sources of AI bias.

Conclusions:

• The Bias Network Approach (BNA) offers a promising sociotechnical method for managing AI bias.
• Visualizing bias interconnections enhances ethical considerations and transparency in AI development.
• Future research should focus on professional biases and material limitations within AI development processes.