An artificial intelligence could guide reparations programmes created to redress decades of US housing discrimination against Black homebuyers. By examining the first US housing reparations programme, researchers have shown that algorithms can suggest how large the monetary support provided by such programmes for an area needs to be in order to improve Black people鈥檚 chances of getting a favourable loan to buy a home.
鈥淚f, for some reason, you are excluded from the housing market, you are excluded from one of the primary ways to generate wealth for your family in this country,鈥� says at the Massachusetts Institute of Technology (MIT) .
A long history of government-sanctioned redlining of neighbourhoods颅 鈥� in which services are withheld from people who live in certain areas based on discriminatory factors 鈥� reinforced by 聽has affected millions of Black borrowers trying to buy homes.
Advertisement
This has caused a huge racial wealth gap. In 2015, the Federal Reserve Bank of Boston found that, in the Boston area, white households had
A growing number of cities and states are considering or implementing housing reparations policies. In 2019, city council members in Evanston, Illinois, voted to set up the first US 聽providing housing grants of up to $25,000, so that recipients could buy or repair homes, or get mortgage assistance.
Inspired by the Evanston example, Hosoi and her colleagues have trained an AI model on mortgage-lending data to evaluate the impact of the $25,000 grant for Black residents in Evanston鈥檚 Cook County. 鈥淲hat we are really interested in is enabling the people from the municipal level, like the city council, to use this kind of tool to actually budget the housing reparations scenario,鈥� says聽team member , also at MIT.
The AI estimated that just 27 per cent of Black people who were denied a conventional mortgage in Cook County would now be accepted by using the $25,000 as a deposit to lower their lending risk. On average, a Black person who applied for a conventional mortgage but was denied it would have needed a $33,289 down payment and $173 monthly support to get accepted, found the AI.
Such findings suggest that the effectiveness of the reparations programme in Evanston could be boosted by increasing the amount of financial assistance, along with encouraging local lenders to lower the debt-to-income ratio used in lending decisions, according to the researchers.
That estimate should be considered a general guide rather than being accurate, because the algorithm wasn鈥檛 able to train on credit score data, which is also used in evaluating lending risk, says So. Evanston officials didn鈥檛 respond to requests for comment.
It is a good example of how to develop a 鈥渇air algorithm鈥� to account for historical factors such as discrimination, says at the , a non-profit group based in Oakland, California. He says the research offers a practical approach for policy-makers that goes beyond a purely academic discussion.
鈥淚 think it鈥檚 important that we supplement the moral reasons that we need to address this with data, because there are some people who aren鈥檛 going to be convinced by those moral reasons to close that racial wealth gap,鈥� says Le. 鈥淚t鈥檚 the idea that we can build this pathway to homeownership through reparations, and this is the amount of money we need.鈥�
The team presented the findings at the ACM Conference on Fairness, Accountability, and Transparency in June. Next the group plans to develop AI-powered tools that policy-makers can try for themselves to examine different housing reparations scenarios.
FAccT 鈥�22
