AI use cases are starting to emerge (that transcend the “AI for Enterprise” marketplace) and illuminate the promise of artificial intelligence to solve tough societal problems. The California Policy Lab and the University of Chicago Poverty Lab “have used County data to predict homelessness among single adults receiving mainstream County services.” This AI-based computer model predicts who will become homeless in L.A. How can your organization leverage these insights into your AI strategy? This post is designed to highlight the “lessons learned” from the systems thinking of an applied AI technology solution to the broken system that is the national homelessness crisis.
The Future of Trust, New Paradigms of Trust, and Designing Trust into Systems are also consistent areas of analysis here at OODA Loop. The What Next? section at the end of this post focuses on how trust in systems plays a role in people seeking help – and how predictive analytics can help social service programs restore trust in systems that have failed those most in need.
Predictive Analytics Applied to “Broken Systems” in the “Era of Code”
“…how we deal with some of these systems that are broke or are getting broken or starting to fail…is about looking for the edge cases.”
Over the course of 2023, we have been tracking artificial intelligence extensively, along with providing analysis of the evolution of ChatGPT, Deep Learning, Large Language Models, Neural Language Models, and Natural Language Processing. What is interesting about this “edge case” of AI-based predictive analytics applied to homeless policy and resource management is that it intersects with the overall theme of the OODA Almanac 2023 – “Jagged Transitions” – including “the adoption of disruptive technologies while still entrenched in low-entropy old systems and in the face of systemic global community threats and the risks of personal displacement” – and the following forecasting themes from the 2023 Almanac:
The System is Broke: “Within OODA, we use the phrase “The System is Broke” as shorthand for the dysfunction and decline of government, private sector, and cultural systems. In 2022, it seemed like we invoked the phrase on an increasingly frequent basis to describe issues in U.S. healthcare, transportation, financial policy, high education, supply chain disruption, pandemic response, and governance.”
The homelessness crisis is symptomatic of the intersection of the broken systems that include the opioid epidemic and the broken mental health care system.
The Code Era: “We are entering into the Era of Code. Code that writes code and code that breaks code. Code that talks to us and code that talks for us. Code that predicts and code that decides. Code that rewrites us. Organizations and individuals that prioritize understanding how the Code Era impacts them will develop an increasing advantage in the future. If a tool like ChatGPT is available to all, the advantage goes to those who know how to derive the most value from the tool. Organizations and societies that look to utilize these tools rather than ban them will maximize this advantage. Instead of banning ChatGPT in schools, we should be teaching students how to use the capability to better advance their knowledge, and ability to learn, and communicate ideas. We will also rapidly enter into an age of regulatory arbitrage around the acceptable use of new technologies.”
Besides this evolving regulatory environment. applied technology use cases at the county, state and federal level – like this AI-based system already in the field and in use by the Los Angeles County’s Homeless Initiative – will also be on the rise and will continue to be compelling, solutions-based, scalable nationally – and will undoubtedly impact both regulatory thinking and future policy.
Predicting and Preventing Homelessness in Los Angeles
“The sheer scale of the data makes it ideal for the application of predictive analytics, which is the use of statistical models to make predictions about the future based on patterns and interrelationships between current and historical data.”
Report Summary
In 2019, The California Policy Lab (CPL) and the University of Chicago Poverty Lab have used County data on multi-system service use to predict homelessness among single adults receiving mainstream County services.(1) By identifying people at high risk of first-time homelessness or returns to homelessness and understanding risk factors associated with future homelessness, the County can more effectively target its homelessness prevention efforts to ensure limited resources are going to those most likely to benefit from them.
Methodology & data sources
Using Los Angeles County data (13), the research team has developed a model for predicting homelessness in the County. The data sources for the project are derived from the Enterprise Linkage Project (ELP), which holds over 85 million service utilization records on 1.9 million single adults from seven agencies covering health services, benefits payments, law enforcement, and homeless services. The sheer scale of the data makes it ideal for the application of predictive analytics, which is the use of statistical models to make predictions about the future based on patterns and interrelationships between current and historical data. For example, for this research we predicted whether single adults experienced a new homeless spell in the 12-month outcome window of calendar year 2017, using data derived from calendar years 2012-16 as the prior service period.
Using predictive analytics, the research team has created models to predict two types of new homeless spells (NHS): returns to homelessness (RTH), in whichthe individual is not homeless in the six months prior to the outcome window, and first-time homelessness (FTH), in which the individual has no record of homelessness prior to the outcome window. (14)
Notes from the Report
(1) For the purposes of this project, “mainstream County services” include services provided by LA County departments reporting data to the Enterprise Linkage Project. Those departments include the Department of Health Services, Department of Mental Health, Probation, Sheriff’s Department, Department of Public Health (Substance Abuse Treatment & Control), and Department of Public Social Services.
(13) The research team and LA County take data privacy extremely seriously and there are multiple measures in place to ensure that privacy. Individual County agencies participating in the Enterprise Linkage Project (ELP) run an encryption code that scrambles personally identifiable information such as names, birth dates, and social security numbers of the individuals in their data. The data is then uploaded to a secure server for inclusion into the ELP. The California Policy Lab has a data sharing agreement with the County CEO providing access to this de-identified data for the purposes of this project. The research team also used Homeless Management Information System (HMIS) data provided by the Los Angeles Homelessness Services Authority (LAHSA). The County encrypts the personally identifiable data in the HMIS using the same method that is applied to the rest of the ELP, and then shares the data with the research team. The research team does not have access to any information that would re-identify the individuals in the data set.
(14) Because predictive analytics requires prior risk factors in order to make predictions about the future, only those County clients who have had interactions with County services prior to the outcome window (approximately 70% of individuals experiencing new homeless spells, and just over 50% of individuals experiencing first-time homelessness) can be included in the model.
REPORT: Predicting and Preventing Homelessness in Los Angeles 
PRESS RELEASE: Report: LA Could Better Target Homeless Prevention Services with Predictive Analytics
Is there a way to predict who will become homeless? These UCLA researchers say yes
“Only 1 in 10 people who seem like they are going to become homeless — actually become homeless.”
As reported by the LA Times: “Most [homeless] prevention programs don’t take such statistics into account, erring on the side of helping as many people in need as possible. But to be truly effective and cost-effective, a program would have to be able to identify that one person who will become homeless with reasonable accuracy. Until now, there’s been no way to do that. Researchers at UCLA’s California Policy Lab and the University of Chicago Poverty Lab, however, are changing that by analyzing millions of interactions between Los Angeles County’s social services agencies and residents…UCLA reported that it has improved the odds of identifying who will become homeless to 1 in 2.
Buoyed by the results, L.A. County homeless officials are launching a $3-million pilot program to put the laboratory’s work to a real-world test. The departments of health, mental health, children and family services and public social services will comb a list of people compiled by UCLA and target them for preventive services…the researchers used predictive analytics to model hundreds of potential risk factors for homelessness. Chief among them were interactions with the county’s social services agencies:
- Half of the people who became homeless for the first time in 2017, and 70% of those who were previously homeless and had lost their housing again in 2017, were already clients with the county.
- Those at the highest risk were found to have come from areas of high poverty and had been involved with multiple agencies, from mental health services to the jail.
- Given the overall correlation, the researchers were able to predict each person’s likelihood of becoming homeless in 2017 by examining all county records of single adults over the four preceding years.
- Among the 3,000 ranked as most likely to lose housing, nearly half did — a rate 48 times the average county client.
- Among the top 1% of that group — or about 19,600 people — more than a third became homeless.
- Researchers concluded that as many as 6,900 episodes of homelessness could have been prevented by addressing that 1%.”
A computer model predicts who will become homeless in L.A. Then these workers step in
“Since publication of the 2019 report, CPL has partnered with Los Angeles County on a pilot program to identify and reach out to L.A. residents at the highest risk of homelessness.”
Also from the LA Times (June 2022):
When her phone rang in February, Mashawn Cross was skeptical of the gentle voice offering help at the end of the line.
“You said you do what? And you’re with who?” the 52-year-old recalled saying.
“How did you get my name to start with?” Cross asked.
The answer is an unusual mobilization of data analysis to try to head off homelessness before it starts. Cross is part of a rare effort by L.A. County to marry predictive modeling — a tool used to forecast events by tracking patterns in current and historical data — with the deeply personal work of homelessness prevention. The county found Cross and scores of other people through a predictive tool developed by UCLA researchers, which pulls data from eight L.A. County agencies to help outreach workers focus their attention and assistance on people believed to be at gravest risk of losing their homes. L.A. County has struggled to keep up with the number of people who become homeless annually, even as it steps up efforts to get people into housing. Figuring out whom to help is crucial because millions of residents seem vulnerable yet avoid homelessness, said Janey Rountree, founding executive director of the California Policy Lab at UCLA.
Researchers have found it surprisingly complicated to guess who will slide into homelessness and who will avoid it…the California Policy Lab and the University of Chicago Poverty Lab said a decent prediction would require at least 50 factors — and that the best models would require “somewhere between 150 to 200.” The predictive model now being used in L.A. County uses an algorithm that incorporates about 500 features, according to the UCLA team. As of 2020, the UCLA team found that few of the people identified by its predictive modeling — under two dozen in two years — were getting services specifically meant to prevent homelessness under Measure H, an L.A. County sales tax approved by voters.
So the county decided to give them a call. In July, the newly formed Homeless Prevention Unit began to reach out to people deemed at highest risk by the predictive model, cold-calling residents like Cross. The UCLA analysis is done with data stripped of identifying details, which the county matches up with names and information to find possible clients.
What Next? Trusting Systems
“We have clients who have understandable mistrust of systems…They’ve experienced generational trauma. Our clients are extremely unlikely to reach out for help.”
Los Angeles is using AI to predict who might become homeless and help before they do
An NPR report this week (October 2023) provided the most recent results from the pilot program in Los Angeles: “In just over two years L.A.’s pilot prevention program has worked with 560 people. Data shows a large majority have stayed housed so far, but the program is conducting a more formal long term study. It includes a randomized control trial that’s tracking people with similar needs who do not receive assistance. Key questions are:
- Whether a few months of help and a few thousand dollars can make a lasting difference; and
- Whether they are targeting the right people, meaning those who would actually end up on the streets but for this intervention.
- “For example, here in Los Angeles, you might have 2 million people on public assistance, all of whom seem vulnerable. But only 1% to 2% of them will ever experience homelessness,” says Janey Rountree, executive director of the California Policy Lab at UCLA, which developed the program’s AI prediction tool.
- Rountree expects to publish the study results in 2026, which is also when the program’s funding runs out; most of its $31 million budget came from pandemic aid. She hopes there will be a strong case that it should be scaled up, and can be a model. In fact, San Diego County has already voted to create its own similar program using predictive analytics.
“It’s not clear yet whether this experiment can keep people housed long term.”
Depending on its long-term results, Los Angeles’ proactive approach could add much needed evidence for what works to prevent homelessness, says Beth Shinn, an expert on the issue at Vanderbilt University and also an adviser to the L.A. program. “Much of the public equates eviction prevention with homelessness prevention,” she says. But while being evicted can certainly have devastating consequences, Shinn says research has not linked it to a significant rise in unsheltered homelessness. She says housing vouchers have shown the strongest success for keeping people housed, at least for families. A small-scale prevention program in New York City has also had good results. But Shinn says the analytics driving Los Angeles’ program are targeting people at much higher risk for losing housing. “You may have more failures there,” she says, but the reward could also be greater. “You make the most difference for the people who are at highest risk.”
Additional Resources
OODA Almanac 2023 – Jagged Transitions: This is the 3rd installment of our OODA Almanac series which are intended to be a quirky forecasting of themes that the OODA Network think will be emergent each year. You can review our 2022 Almanac and 2021 Almanac which have both held up well. The theme for last year was exponential disruption, which was carried through into our annual OODAcon event. This year’s theme is “jagged transitions” which is meant to invoke the challenges inherent in the adoption of disruptive technologies while still entrenched in low-entropy old systems and in the face of systemic global community threats and the risks of personal displacement.
Food Security and Inflation: Food security is emerging as a major geopolitical concern, with droughts and geopolitical tensions exacerbating the issue. Inflation, directly linked to food security, is spurring political unrest in several countries. See: Food Security
AI Discipline Interdependence: There are concerns about uncontrolled AI growth, with many experts calling for robust AI governance. Both positive and negative impacts of AI need assessment. See: Using AI for Competitive Advantage in Business.
Emerging NLP Approaches: While Big Data remains vital, there’s a growing need for efficient small data analysis, especially with potential chip shortages. Cost reductions in training AI models offer promising prospects for business disruptions. Breakthroughs in unsupervised learning could be especially transformative. See: What Leaders Should Know About NLP
