A new study suggests a creative approach to allocating scarce housing resources among homeless youth so that the process is fair and efficient. A group of researchers at the University of Southern California has developed a data-driven framework using artificial intelligence to design housing allocation policies that improve current practice.
The challenges of the homeless youth population in the U.S. are grim: hundreds of thousands have to turn to shelters or be on their own in the streets. Housing assistance programs can very significantly improve outcomes for these vulnerable youth, but demand far exceeds available supply. Both make for difficult choices about how to decide on priorities and allocate scarce resources.
Currently, most communities use a system in which the youth are scored based on vulnerability to decide who should be prioritized. While this is well-intentioned, it is not related directly to outcomes and runs the risk of continuing the same, not-so-fair deal between different groups.
The researchers at USC put forward a much more sophisticated approach with the help of machine learning and optimization techniques in designing personalized allocation policies. More specifically, they aim to create an efficient housing allocation that maximizes the overall probability of successful outcomes for youth to achieve stable housing, while allowing policymakers to specify fairness constraints. Lastly, policies to be created should be easy to understand and explain; it is critical for adoption.
The researchers tested their approach using real-world data on over 10,000 homeless youth. Their AI-designed policies outperform the status-quo approach quite significantly.
When fairness is optimized across vulnerability levels, their policies achieve significantly better outcomes for high-risk youth and leave overall efficiency roughly the same.
Their approach, regarding racial equity optimization, drastically reduced disparities in outcomes compared with current practices across racial groups. Although fairness improved, General Efficiency showed little cost for this. The AI identified innovative ways of shifting resources around in this way to improve fairness without cutting into effectiveness. This work is an example of how AI and data science could be mobilized in efforts to tackle some of the most pressing current social challenges. Better-designed policies can help more equitably and effectively serve our most vulnerable populations once infused with the power of machine learning and optimization. However, further research and real-world testing are called for. The researchers hope to generalize their framework for enhancing resource allocation in other domains—from organ donation to public housing. As AI capabilities continue to advance, we’re likely to see more such applications to enhance fairness and efficiency in complex social systems.
