A recently published research by the housing charity Shelter showed that there are over 320,000 homeless people in the U.K. At the same time, London rough sleeping hit a new record high, with an 18 percent rise between 2018 and 2019. The mayor of London, Sadiq Khan, described this as a “national disgrace” and blamed the crisis on welfare reforms and a lack of investment in social housing. As true as this statement is, what does this mean for those who are left to exist without a home? Can we, as a society and with the help of new technologies, fill our government and councils’ shoes and come up with a solution to this seemingly unsolvable growing problem?
Machine learning might be the solution. StreetLink, a homelessness charity that enables members of the public to connect people sleeping rough with local services that can support them, along with a team of data scientists backed by the Alan Turing Institute, has recently been looking into how machine learning could help improve the decision-making process that goes on in homelessness support. Very soon, AI could decide what to do when passersby report to StreetLink as they spot a person sleeping on the streets.
While AI interfering in charitable projects to deepen human interactions can sound worrying to some, in this case it could mean being more supportive of homeless people, and some cases even save their lives. Since StreetLink started offering its services across the U.K. in 2012, it has encountered the same problem: humans are too vague and sometimes unreliable when they call the company to report of rough sleepers.
When someone calls StreetLink to alert its team and network about a homeless person seen on the streets, they are requested to give a detailed description. Whoever is calling needs to include the rough sleeper’s gender, clothes, location and condition. Most of the time, the description is not complete enough. After receiving the call, StreetLink’s support team needs to go over the information again and make the tough call of whether or not to alert local response teams to find the rough sleeper.
At the moment, because the decision and analysis of the information given by callers is made by humans, just one in seven alerts processed by StreetLink actually results in the homeless person being found. When the weather conditions are particularly bad, there is a spike in the number of alerts, which puts StreetLink’s staff under an overwhelming amount of work. Machine learning could simplify and optimise the whole process.
Using information from past decisions, Alan Turing Institute’s team of data scientists has created a machine learning model that automatically categorises alerts. Once put into place, it would give StreetLink an immediate sense of which alerts should be prioritised—essentially taking over the decision making process from the staff. This model is set to be implemented for the first time in London over the coming months.
Meanwhile, in the U.S., Eric Rice had already seen the potential of machine learning in solving homelessness. As a professor of social work and co-founder of the University of Southern California Center for Artificial Intelligence in Society (CAIS), where he and engineering professor Milind Tambe developed predictive models for public health interventions, Rice works primarily on issues of housing and HIV prevention for homeless youth.
In Los Angeles, homelessness is measured on a much larger scale than what we are witnessing in the U.K.—an estimated 53,000 people experience homelessness on a given night, including 3,000 between the ages of 13 and 24. Just like StreetLink’s machine learning model, Rice’s one predicts and picks up the most at-risk cases. Rice, as well as members of many other community-based organisations, then go to the ‘selected’ rough sleeper to provide help.
CAIS has recently gained traction with a $1 million state grant to deploy its California HIV Research Program through 2020. The U.S. might soon adopt his model, but Rice is thinking bigger already. He shared with OZY that he’s now working on developing predictive models for suicide prevention on college campuses and endangered animal poaching in Africa.
StreetLink’s project would be a smaller-scale version of what is already being used in Los Angeles. And while the concerns surrounding AI making choices for marginalised communities should not be cast aside, it is clear that the world needs the help of innovative technologies, even if only to make these problems seem less insurmountable.