In response to this challenge of allocation resources to the youth, the prevalent method is a Coordinated Entry System. Housing resources are collected across organizations within a community and persons in need of housing are assessed against a criteria. This approach allocates a disproportionate amount of housing to the youth. However, this prioritization process remains heavily paper-driven, with human decision-makers authorized to make decisions based on very simplistic rubrics.
In the research paper “Evidence From the Past: AI Decision Aids to Improve Housing Systems for Homeless Youth”, data was analyzed from over 10,000 homeless youth assessed using the Next Step Tool(NST). The NST scores youth based on self-reported history related to housing and homelessness, risks, socialization, daily functioning, and wellness. Many communities use these scores to as the sole heuristic and place youth accordingly into various housing programs. Perhaps the most precise answers to the stated research questions are the key findings in the study: while having some predictive value for housing outcomes, the NST score alone did not turn out to be the best predictor. The researchers found an inverse pattern between NST scores and housing success; this means that youth with more excellent vulnerability scores went less on achieving housing success without interventions. The finding, to some extent, validates current practice in prioritizing high-risk youth for housing assistance.
The study also identified that AI models, including several other relevant variables from assessment, could predict more detailed and precise ways of housing outcomes. These models were able to identify different factors that predicted success for various types of housing interventions. The study showed that permanent supportive housing was not successful for youth with a trauma history, but with rapid re-housing, youth didn’t do as well either.
These insights signal a high potential for AI decision aids to make more tailored and practical housing recommendations. Taking into account a greater variety of factors and their intricate interactions, AI could help social workers to make more informed decisions about which of the possible housing interventions would be most successful for individual youth.
AI in this process must not strive to replace human decision-making but rather augment it. Their idea is for AI decision aids to provide case workers with subsistence insight and recommendations so humans can still make final decisions about housing placements.
A system like this has clear benefits. Perfect matching of scarce housing resources are successfully placed. Better long-term outcomes for such vulnerable young people will curtail their exposure to the risks of homelessness and enhance an opportunity wherein they can build up a stable, productive life.
AI decision aids in this respect are, however, not without their challenges. There might exist some resistance from housing providers who may show reluctance to rely on an automated system for such critical decisions. More importantly, ethical concerns have to be raised and dealt with in detail to ensure that the AI systems do not perpetuate or worsen any bias already prevalent in housing allocation.
The authors proposed working with the Department of Housing and Urban Development and local communities to develop easy-to-use interfaces between these AI decision aids. These interfaces would offer easier access and interpretation by housing providers regarding the predictions and recommendations coming out of AI but leave human judgment in the frame of decision-making untouched.
