Fri, Oct 30, 2020 page9

• Using AI, a Canadian city predicts who might become homeless

  The system, implemented in London, Ontario, assesses individual factors and helps the city assist vulnerable residents, while providing reasonable data security, a professor said

  • By Chris Arsenault / Thomson Reuters Foundation, Toronto
  • • 
    • 
    • 

  As makeshift tent cities spring up across Canada to house rough sleepers who fear using shelters due to the COVID-19 pandemic, one city is leveraging artificial intelligence (AI) to predict which residents are at risk of becoming homeless.

  Computer programmers working for the city of London, Ontario, 170km southwest of the provincial capital Toronto, said that the new system is the first of its kind anywhere — and it could offer insights for other regions grappling with homelessness.

  “Shelters are just packed to the brim across the country right now,” said London’s homeless prevention manager Jonathan Rivard, who works on the AI system.

  

  Illustration: Mountain People

  “We need to do a better job of providing resources to individuals before they hit rock bottom, not once they do,” Rivard said.

  Canada is seeing a second wave of COVID-19 cases, with the Ontario Government warning that the province could experience the “worst-case scenarios seen in northern Italy and New York City” if trends continue.

  Homeless people are at higher risk of being infected and infecting others during the pandemic, due to weakened immune systems, and poor access to shelter and sanitation, health experts say.

  The AI system analyzes the personal data of participants to calculate who faces having nowhere to sleep for an extended period, said Matt Ross, an information technology (IT) expert who helped the city build the program.

  As a test, the Chronic Homelessness Artificial Intelligence (CHAI) system tracked a group of individuals for six months before its formal launch in August.

  Over that period, CHAI saw a 93 percent success rate in predicting when someone would become chronically homeless, Ross said, adding that it is now meeting or exceeding that rate.

  By using AI to anticipate who is likely to become chronically homeless, the city can prioritize how it works with those individuals, and try to provide them with safe housing and access to health services they might need, Rivard said.

  Chronic homelessness refers to someone who has been staying in a shelter for 180 days or more within a12-month period, Rivard said.

  Those individuals use 12 times more resources than people who are occasionally homeless, and addressing their situation can in the long run save time and money, he said.

  City staff are working with local shelters, community groups and homeless people on how to best use the data, Rivard added.

  Annually, more than 230,000 people in Canada experience homelessness — “about 35,000 on any given night,” advocacy group Canadian Alliance to End Homelessness president Tim Richter said.

  Richter blames government cuts to affordable housing and other programs in the late 1980s and early 1990s for what he calls the “explosive growth” of mass homelessness over the past 30 years.

  When city officials first suggested using a computer program to predict chronic homelessness, it “raised some red flags” related to privacy, said Peter Rozeluk, executive director of Mission Services of London, a nonprofit that runs homeless shelters.

  “I suppose whenever anyone uses the term ‘AI,’ it can seem dystopian, simply because of how the media and Hollywood has depicted artificial intelligence,” Rozeluk said.

  After discussing the proposal with officials, he said that he supports its general goal of getting better data to aid in decision making.

  The AI program is only applied to consenting individuals, and participants can quit the program at any time, Ross said, adding that their data would be removed from the model.

  His team of data scientists do not have access to the real names of individuals involved.

  Instead, each person is given an identification number which is run through the system along with other data, including age, race, gender, military status, the kinds of city services they have accessed and how often they sleep in shelters.

  Unlike most other AI systems, which produce their final conclusions without revealing the steps taken to get to them, London’s technology can explain how and why it reached assessments about an individual’s risk level, Ross said.

  Building the system cost about C$14,000 (US$10,480), funded by the city’s IT department, Ross said, adding that CHAI therefore does not take away resources from front-line services for homeless people, such as shelters.

  So far, the system has identified at least 88 people at risk of chronic homelessness, in a city of about 400,000 residents, Rivard said.

  According to the model’s predictions, a single man who has stayed in shelters, is older than 52 and has no local family is often at high risk of becoming chronically homeless, especially if he is a veteran or a First Nation member, Rivard said.

  While the AI provides information about an individual’s risk of becoming long-term homeless, all decisions related to deploying services are kept in human hands, he added.

  Two unaffiliated computer science experts and a privacy lawyer said that the program seems to take the necessary steps to protect users’ personal information.

  “It looks like they have put a lot of thought into doing it right,” said Teresa Scassa, a law professor at the University of Ottawa who studies AI and privacy.

  The designers have ensured that the data put into the system is standardized and accurate, and meets national guidelines on the ethical use of automated decisionmaking, she said.

  Amulya Yadav, an assistant professor at Pennsylvania State University who studies AI and homelessness, said that the initiative is an example of how machine learning is “being democratized.”

  “The barriers to entry are being reduced,” Yadav said. “I really hope they pull it off well, and it’s the first of many.”

  However, Scassa, Yadav and other experts worry about what might happen to sensitive data on vulnerable residents.

  “It is paramount to think about not just what our data is used for, but what can our data be used for in the future — and assume whoever holds the data has no scruples,” said Paulo Garcia, an assistant professor at Ottawa’s Carleton University.

  If a new government came into power looking to cut costs, for example, this information could potentially be used to determine who is taking up large amounts of resources and where funding could be slashed, Scassa said.

  Rozeluk, who works on the front lines of Canada’s homelessness crisis, has a different concern.

  Predicting when someone might become chronically homeless is less important than providing actual housing, he said.

  Studies have been done for decades on the issue, and the consensus is clear, Rozeluk said.

  “The solution to homelessness is safe, adequate, affordable housing ... and providing support afterwards,” he said.
• Most Popular

  • 1

    Palmeiras hopes for club cup win

  • 2

    Magnitude 6.4 earthquake hits Hualien County

  • 3

    Taiwan to see nine extended holidays in 2026

  • 4

    Pegatron ranks last in Greenpeace report

  • 5

    Hegseth outlines Chinese threat