Enhancing Equitable Access to AI in Housing and Homelessness System of Care through Federated Learning

Smaller agencies can now access quality AI tools: The paper demonstrates that federated learning allows smaller homeless service agencies—which typically lack sufficient data to build their own AI systems—to benefit from predictive models trained collaboratively with larger agencies, addressing a significant equity gap in AI access across the homelessness services sector.

Privacy is preserved without sacrificing accuracy: The federated learning approach achieved comparable performance to a scenario where all data was fully shared between agencies, but without requiring agencies to exchange sensitive personal information about people experiencing homelessness, protecting individual privacy while enabling collaboration.

Real-world validation in Calgary: The researchers tested their approach using actual data from Calgary's Housing and Homelessness System of Care, demonstrating practical feasibility rather than theoretical potential, which strengthens confidence in the model's real-world applicability.

Better resource allocation for vulnerable populations: By enabling all agencies to access improved predictive AI tools, the system can help human decision-makers more effectively allocate limited supportive housing resources to people experiencing homelessness, potentially improving outcomes across the entire care system.

A model for equitable AI deployment: This work provides a concrete example of how federated learning can be used to distribute AI benefits equitably across organizations with different technological capacities, offering a potential template for other social services sectors facing similar data fragmentation challenges.