Title: How Societal Context Understanding Enhances AI Product Development
Introduction
AI-related products and technologies operate within a societal context, which encompasses various social, cultural, historical, political, and economic factors. However, translating this complex and qualitative context into quantitative representations poses a significant challenge for machine learning and responsible AI product development. The problem understanding phase of AI product development is crucial as it influences how problems are formulated and the subsequent decisions related to datasets and ML architectures. Neglecting the societal context can lead to the propagation of biases and fragile solutions.
The Problem Understanding Chasm
The gap between the understanding of complex problems by product developers and society stakeholders is referred to as the problem understanding chasm. Product developers often lack the necessary tools and knowledge to effectively consider the societal context during development. Consequently, they have a shallow quantitative understanding, while users and stakeholders possess a deep qualitative understanding. This divergence can have real-world repercussions, such as the racial bias found in a healthcare algorithm that failed to account for critical socio-structural factors.
Bridging the Problem Understanding Chasm
To bridge this gap responsibly, AI product developers require access to community-validated and structured knowledge about societal context. Societal Context Understanding Tools and Solutions (SCOUTS), a research team within Google Research’s Responsible AI and Human-Centered Technology (RAI-HCT) team, aims to empower developers with this knowledge throughout the product development lifecycle. By integrating structured societal context knowledge, AI developers can create responsible and robust solutions to complex societal problems.
The Need for a Societal Context Reference Frame
To organize structured societal context knowledge, SCOUTS collaborated with other RAI-HCT teams and external experts to develop a taxonomic reference frame. This reference frame leverages complex adaptive systems theory and identifies three key elements: agents, precepts, and artifacts. Agents represent individuals or institutions, precepts encompass beliefs, values, stereotypes, and biases, and artifacts include language, data, technologies, societal problems, and products. Among these elements, precepts are considered crucial in understanding societal context.
Eliciting Causal Theories from Communities
SCOUTS employs community-based system dynamics (CBSD) to foster collaboration and elicit causal theories from communities. System dynamics methodology enables qualitative and quantitative articulation of causal theories using causal loop and stock and flow diagrams. With CBSD, community groups can learn the basics and begin building models collaboratively, enabling them to describe the problems they face directly. This approach has been applied in healthcare diagnostics, addressing data gaps and incorporating critical factors like cultural memory and trust in medical care.
Conclusion
Understanding and considering the societal context in AI product development is crucial for responsible and equitable solutions. SCOUTS’ mission is to provide developers with scalable and trustworthy societal context knowledge throughout the development lifecycle. By bridging the problem understanding chasm, AI developers can create solutions that address complex societal problems effectively and responsibly.