When asked about the connotation of words such as artificial intelligence, machine learning, and blockchain, most would point to lofty, technical, and even futuristic concepts: Robotics. Self-driving cars. Cryptocurrency. For the most part, these highly functioning digital technologies are considered to be designed for the advanced, by the advanced, with the purpose of breaking the boundaries of human capability. But recently, these up and coming technologies have become relevant in a decidedly unglamorous, often overlooked issue that has plagued society for decades: homelessness.
One of the largest blockades to practical homelessness solutions is the unavailability of the kind of data that is necessary to provide support to the unhoused; the cyclical nature of homelessness results in gross miscalculations in number, there is an inherent mistrust that prevents the unhoused from identifying themselves, and often the unhoused have no access to personal records, IDs, or the like . Without the knowledge of how many people are homeless at a given time, what their needs are, or how to best get a hold of them, agencies and support centers are unable to provide critical resources like food and medical care. And, while retroactive aids like shelters, food pantries, and needle exchanges are effective and highly necessary, they are limited in their scope both in terms of how many people they can reach and the extent of services they can provide due to limited resources.
Thus, there exists a niche where digital technology in the form of machine learning can and has been used to recruit or disperse the relevant information. Using advanced algorithms to gain a better understanding of homeless demographics and connect the unhoused to a wider spectrum of resources can help to bolster the support they get from the entities who are established to help them. And in a variety of cases, it has already been shown to increase the positive impact on their communities.
In Austin, Texas, researchers have implemented blockchain technology successfully to aid the city’s unhoused communities. Blockchain, which uses digital ledger technology to create a compounding network of encrypted transactions, allows “an individual’s control over their information” that can be verified and stored . Though its most sensationalized usage has been in cryptocurrency, blockchain was used in Austin to create a digitized and attainable medical and identification record for unhoused people who may otherwise have lost access to these important documents . This “digital identity” uses biometric features, such as fingerprints, to connect an individual’s records to social services, hospitals, and assistive agencies that they may need to access . All information is private on the blockchain, but accessible and shareable with individual consent. This is possible due to the necessary combination of an individual’s private key and a public source’s key to say “yes, I am this person who wants to share my information with this service.” Only with both, the former ensuring privacy and the latter acting as a verifying agent, can information be released or added .
This Austin-based project not only challenges the notion that blockchain is only functional for elitist currency exchange, but illuminates a promising future of increased access to necessary resources for the unhoused. Though only implemented on a relatively small scale over a period of 12 months for the initial study, the researchers yielded promising results: with more than 200 unhoused participants, more than 50% lacked some form of photo ID, insurance, or otherwise. With the blockchain prototype implemented, however, they saw both a willingness to participate in the system from all sides (unhoused individuals, care centers, etc), and a successful distribution of identifying information .
Blockchain, then, proves to be truly advantageous when functionalized for the benefit of society: with this technology, uhhoused people can retain privacy (as it is only functional with their consent), keep track of identification numbers and medical history, and seek help without the fear of being turned away for lack of these things. The proper medical care and social service access that can be gained in this manner would not only expedite aid and response time in crises for those living on the street if the technology were broadened to a state or country-wide scale, but could even increase proactive homelessness response by acting as a resource for those in tumultuous situations who are at risk of homelessness .
The city of London, Ontario, has also turned to digital technology in the form of artificial intelligence to combat issues related to homelessness. The city’s Information and Technology Services division devised an algorithm that predicts whether a person is at risk, in 6 months time, of chronic homelessness. This is defined as spending at least 180 out of 365 days of the year in a shelter . The algorithm used data from the Homeless Individuals and Families Information System and created a related vector of variables for each individual. In the vector, it defined certain parameters for each individual in the study as dynamic (like the number of services a person used in a set of designated time steps) or static (income, demographic data), and fed the complete result into a Python model neural network that would output the prediction of chronic homelessness . Logically, they wanted the system to account for as many cases of predicted future homelessness as possible, so slightly overestimating was more desirable than underestimating. With this in mind, precision and recall statistical optimizations were also calculated to limit the amount of false negative outcomes, that is, people not predicted to end up chronically homeless that ultimately do .
The neural network AI was additionally bolstered by a “local interpretable model agnostic explanations algorithm” that had the sole purpose of making the program more digestible to a common user . It did this by taking the end prediction (at risk or not at risk) and displaying all of the steps that the code went through to finalize the result: this includes naming the variables or combinations thereof that were flagged by the machine, as well as an statistically significant determinations. The final model, in addition to its comprehensibility, boasted an accuracy rating of 0.971 .
This kind of machine learning algorithm has previously been used in predicting undesirable outcomes in things like finances, but this research group reframed the idea of pinpointing undesirable outcomes to center around the social issue of chronic homelessness. This AI’s purpose is for a more humanized application, which generates a need for reinforced functionality measures and increased trust in the algorithm.
This is achieved by taking the typically uninterpretable “black box” neural network model and adding features to logically lead users through its decision making. The London project, upon its inception, has already seen success in a local shelter that adopted this AI technique to guide their resource allocation priority . And the impact of this AI application could be far-reaching with even greater implementation: if people at risk for chronic homelessness can be correctly identified on a larger and more accurate scale than current simple trend-based models, more intervention measures could be sought out in advance for these individuals to prevent this worst-case from happening. This could lower the number of chronically homeless and lessen burdens on social services and shelters.
Likewise, USC’s Center for AI in Society has developed an algorithmic system designed to take aim at another great danger of homelessness: health and health awareness, specifically in youth. Young people who are unhoused have a greater risk of contracting HIV than those who are not. Additionally, due to the homeless youth communities oftentimes being unreachable and harboring distrust for many authoritative figures, it is difficult to pass on the necessary learning and resources to bring these numbers down and keep them as healthy as possible.
CAIS aimed to address this issue using AI . Their system, called “CHANGE”, selects the people most likely to be influential among youth experiencing homelessness; this could be based on how widely they socialize, how often they come into contact with other unhoused youth, or how frequently they visit social services meant to aid their circumstances. The model chooses these kids, and the researchers then invite them to participate and disseminate information about safe sex practices and HIV transmission to their peers. Their algorithm achieves this by modeling the youth as mathematical nodes which have an inferred probability of transferring information to their neighbors . They use optimization and maximizing arguments (based on the “influence” circumstances listed above) to seek out which nodes have the most potential for spread of information. After using this algorithm to identify the youth, they also employ a digitized strategy and probability code to infer how many of these selected “peer leaders” will follow through with training and execution.
After using their machine learning algorithm and disseminating the information, CAIS quantified the effectiveness of their AI by surveying youth on new and past sexual practices. They saw a statistically significant increase in reports of safe sex after implementation of their peer leadership program: participants lowered their odds of unsafe sex by upwards of 31%. . This means that their technology, with continued use, could lower instances of HIV among the homeless thus decreasing medical costs, public shame and stigma, illness, and even death. Using AI that was likely born to predict simple probabilities, CAIS was able to truly impact homeless health in a positive manner.
These advanced digital technologies have only very recently become commonplace to hear about in everyday life, but so far only in relatively narrow industrial sectors. As sophistication, knowledge, and accessibility increase, however, it would be ignorant to believe their power would continue to not see utilization in common practices and societal issues.
It is natural to consider an issue such as homelessness to be somewhat unsolvable, and not without reason. Math, finances, and similar subjects are often taken to be concrete, absolute, and containing an achievable solution; easily applicable to AI. Homelessness, on the other hand, is without boundaries, constantly fluctuating, and affected by so many parameters, both individual and terminal, that it would not be thought possible to truly quantify, and thus impossible to truly breach solutions in a quantifiable way. But researchers in Austin, London, and Los Angeles have gathered verifiable proof that with this technology, there is a way to interpret the complexities with data and mathematics, and generate preventative and retroactive homelessness measures that have the potential to be widespread and effective. Humans are problem solvers by nature, and though machine learning is implied to be made for increasing human capability, homelessness is arguably one of the biggest problems we face. So why not let the bounds of human capability be human centered in itself?
 Institute of Medicine (US) Committee on Health Care for Homeless People, “The methodology of counting the homeless,” Homelessness, Health, and Human Needs., 01-Jan-1988. [Online]. Available: https://www.ncbi.nlm.nih.gov/books/NBK218229/. [Accessed: 29-Jan-2022].
 T. Mercer and A. Khurshid, “Advancing Health Equity for People Experiencing Homelessness Using Blockchain Technology for Identity Management: A Research Agenda,” Journal of Health Care for the Poor and Underserved, vol. 32, no. 2, pp. 262–277, 2021, doi: 10.1353/hpu.2021.0062.
 A. Khurshid and A. Gadnis, “Using Blockchain to Create Transaction Identity for Persons Experiencing Homelessness in America: Policy Proposal,” JMIR Research Protocols, vol. 8, no. 3, p. e10654, Mar. 2019, doi: 10.2196/10654.
 “Blockchain for Digital Identity: Real World Use Cases,” ConsenSys. https://consensys.net/blockchain-use-cases/digital-identity/ (accessed Jan. 23, 2022).
 “Using AI to Assist Those Experiencing Homelessness in Austin,” GovTech, Jul. 16, 2020. https://www.govtech.com/health/Using-AI-to-Assist-Those-Experiencing-Homelessness-in-Austin.html (accessed Jan. 23, 2022).
 B. VanBerlo, “An open-source interpretable machine learning approach to prediction of chronic homelessness,” Medium, Jun. 02, 2021. https://towardsdatascience.com/an-open-source-interpretable-machine-learning-approach-to-prediction-of-chronic-homelessness-8215707aa572 (accessed Jan. 23, 2022).
 B. VanBerlo, M. A. S. Ross, J. Rivard, et al, “Interpretable Machine Learning Approaches to Prediction of Chronic Homelessness,” arXiv:2009.09072 [cs], Sep. 2020, Accessed: Jan. 23, 2022. [Online]. Available: http://arxiv.org/abs/2009.09072
 “Eradicating HIV Among Homeless Youth Using AI – USC CAIS,” USC Center for Artificial Intelligence in Society. https://cais.usc.edu/news/eradicating-hiv-among-homeless-youth-using-ai/ (accessed Jan. 23, 2022).
 B. Wilder et al., “Clinical trial of an AI-augmented intervention for HIV prevention in youth experiencing homelessness,” arXiv:2009.09559 [cs], Nov. 2020, Accessed: Jan. 23, 2022. [Online]. Available: http://arxiv.org/abs/2009.09559
 A. Khurshid, V. Rajeswaren, and S. Andrews, “Using Blockchain Technology to Mitigate Challenges in Service Access for the Homeless and Data Exchange Between Providers: Qualitative Study,” J Med Internet Res, vol. 22, no. 6, p. e16887, Jun. 2020, doi: 10.2196/16887.