When Redfin CEO Glenn Kelman briefly spoke with Sam Altman, he warned him to tread lightly when it comes to infusing artificial intelligence into real estate search.
“You should be careful about housing,” Kelman told the OpenAI CEO. “It’s not like other searches. There’s so much regulation around it — and for good reason.”
Kelman recalled the conversation during a virtual panel discussion hosted by real estate startup Pacaso on Tuesday.
In May, OpenAI began rolling out ChatGPT plugins in partnership with Redfin and Zillow Group to boost how people search for homes online.
But one month later, in response to a June report by real estate news site Inman discussing the discriminatory risks posed by chatbots, and after reviewing a 28-page report from Redfin, OpenAI swiftly deactivated all real estate plugins, according to Kelman.
The Redfin and Zillow plugins are no longer available on ChatGPT. Both were disabled in June.
“We agree with their decision and are working together with OpenAI to ensure ChatGPT’s responses to users’ questions meet fair housing standards — an issue we raised with them and are optimistic that they’ll be able to address,” said Redfin spokesperson Angela Cherry.
Redfin created the document for OpenAI to show examples of the “ideal way” that ChatGPT would respond to a user question about housing, in line with fair housing regulations, Cherry said.
Zillow spokesperson Claire Carroll said the company is focusing on “other AI priorities.”
“The generative AI space is fast-moving, and we’re excited about the ways it will help our customers and partners,” she said.
The rollout of the plugins sparked questions about how the new technology plays into the longstanding issue of racial discrimination in real estate, and the possibility of violating fair housing laws. The Inman report from June highlighted how users could ask ChatGPT which neighborhoods were associated with specific demographics.
“There’s no great solution to it. And it’s a broader problem than just real estate search.”
Kelman was asked how Redfin avoids racial bias or discrimination when building AI models to identify buyer likelihood. He said the company does not provide demographic information.
“They don’t mind reading another book,” Kelman said of training AI models. “We’ve just taken some books off the shelf around user information.”
Redfin in 2021 announced it would not include neighborhood crime data on its real estate search platform.
“We were considering this because we’re very much focused on answering all the questions people have when they’re considering a home purchase, and we know that one of these questions is whether they’ll feel safe in a given home or neighborhood,” Redfin wrote in a blog post at the time. “But the data available don’t allow us to speak accurately to that question, and given the long history of redlining and racist housing covenants in the United States there’s too great a risk of this inaccuracy reinforcing racial bias.”
Redfin settled a lawsuit last year with the National Fair Housing Alliance and nine other organizations, which alleged the company’s minimum home price policy violated the Fair Housing Act by discriminating against buyers and sellers in communities of color.
Increasing use of automation and larger amounts of data have spurred debate about discrimination in online marketplaces such as eBay and Airbnb.
Kelman emphasized that Redfin has deeply considered bias in AI systems. “There’s no great solution to it,” he said. “And it’s a broader problem than just real estate search.”
AI’s biggest impact in real estate so far has been its ability to predict prices and recommend homes, Kelman said. In regard to recent advancements in generative AI, he said he mostly sees real estate agents use ChatGPT as a so-called “spam machine” to help with generating emails to send to customers.
Kelman was joined in Tuesday’s discussion by Spencer Rascoff, the former Zillow Group CEO and co-founder of Pacaso, a real estate startup that sells fractional shares of vacation homes. The former business rivals talked about the state of the market, real estate-related legal cases, and return-to-work’s impact on housing.
Kelman said the current housing market is unlike the 2008 financial crisis. Back then, sellers were pressured to sell their houses at much lower prices because they couldn’t afford their mortgages, making overall home prices cheaper, he said.
Kelman said there isn’t “widespread distress” in today’s market, meaning fewer homes are available for sale amid already-low inventory. At the same time, a recent Redfin study found that more than 90% of homeowners have mortgage rates below 6%, keeping them locked into their current homes to avoid today’s elevated rates.
Kelman said changes in the adjustable rate mortgage rate or a sudden economic swing, either up or down, could force movement in the housing market. But for now he said the market remains sluggish, with only 4 million units transacting annually, down from 6 million in 2021.
“It’s going to be a real grind,” he said.
Two class-action lawsuits could change the structure of real estate. Rascoff called out the “elephant in the room,” referring to a pair of lawsuits related to buyer-agent commission structures. The lawsuits claim that forcing a seller’s agent to share half of their commissions with the buyer’s agent keeps commissions artificially high in the U.S.
“My guess is that most judges are hesitant to cut the real estate industry in half and say all the buyer’s agents are out of work,” Kelman said about the lawsuits. “But it’s possible the DOJ could follow up and do something more aggressive.”
Return-to-office is having an effect on some housing markets. Kelman said there’s somewhat of a housing rebound in places like Seattle, as large companies are calling employees back to the office. Many workers who moved to more affordable regions like Florida and Texas during the pandemic are now returning, he said. Large employers, including Redfin and Amazon, have summoned employees back to their workplaces.
“A bunch of our Seattle agents are busy with people,” Kelman said. “Mostly for Amazon.”
