The world currently sits in the next age of technological revolution: AI. The rapid pace at which AI models and tools are popping up is only accelerating as the world embraces this next phase of its evolution.
As AI continues to evolve and grow, so does its use in real estate. More and more agents are adopting AI in order to increase their productivity, while more and more clients are using AI to help them understand the housing process.
There is a potential problem that the industry faces from this movement, however, and that is the question of how the technology aligns with Fair Housing regulations. Real estate professionals have to fall into compliance with the Fair Housing Act, but does AI adhere to that compliance? How can the industry safeguard against violations of fair housing in AI use?
In the Wisconsin REALTORS® Association webinar “Fair Housing and AI: What Practitioners Need to Know,” National Fair Housing Alliance (NFHA) Chief AI Officer and Head of the Responsible AI Lab Dr. Michael Akinwumi dove into some of the concerns over the fair housing violations of steering and discrimination when it comes to using AI in real estate.
Akinwumi opened by explaining that we are no longer at a stage of questioning whether or not to adopt AI, it has already been adopted by the industry—noting that 42% of Realtors® use it weekly and 69% use it monthly. He said that the question that stands now is now surrounding how to responsibly use AI.
Akinwumi said that for AI adoption to comply with the Fair Housing Act “it must also be driven by accuracy, compliance and fairness.”
He continued that AI itself is not accountable to the Fair Housing Act, but the real estate professionals using it are.
“It’s very, very important that you know that your license does not transfer to the AI vendor,” he explained. “This is where your power as a human agent, your influence based on the domain experience that you have comes to play when it comes to aligning the systems with the Fair Housing Act.”
Akinwumi added that the question agents should be asking is that “if you know the reasoning, the interpretation, then how do you make sure that your professional training, your professional experience addresses these issues?”
He explained that when real estate professionals are trained in the Fair Housing Act, they understand that “discretion in translating a preference into a place is exactly what the Fair Housing Act regulates in human agents.”
“That means if you bring your own agency as a human to how you use these systems, then there’s a lot you can do in making sure that these systems are compliant with the Fair Housing Act,” Akinwumi continued.
Careful consideration of language
The basis of Akinwumi’s message was careful consideration of language when using AI, whether that’s in prompting or in evaluating the responses it generates.
He explained that AI models are “not discriminatory by rule,” but rather are “discriminatory by interpretation.” Basically, the way in which AI interprets the information it is handed through prompts and the information it gathers are where discrimination and steering originate, and not from the general coding of an AI in itself.
So, the question stands of how to prevent this discrimination and steering from occurring, and from reaching clients?
Akinwumi said that in order to avoid fair housing violations, agents need “visibility into how the model interprets your prompt.”
“For example, if you’re using ‘good school, safe area or safe community, low crime area,’ what the model is trying to do is really to interpret what is going on,” he continued. “What does each of those word combinations mean? They could actually serve as proxies or protected classes, both at a federal level and then also at a state or local level.”
Akinwumi said that’s why there needs to be careful consideration in prompt wording, because “if we want to make sure that these proxies do not constitute disparate treatment in how the models are responding to your prompt, you want to have visibility into that interpretative layer of the model.”
Akinwumi also said there needs to be consideration when providing information about clients, as identifying info can easily sway an AI model into steering or discrimination.
He explained that in testing AI models, adding in identifying factors such as lifestyle, race or location can add to the steering or discrimination a model generates in its response.
While these programs are promoted, marketed and advertised as personalized algorithms, Akinwumi said that “the bad news is that the more you tell the system, the worse it becomes when it comes to racial steering.”
“Why? The model reads the same words (good schools, safe community, no crime) differently depending on who is asking, depending on the profile that you’re building or including in your prompt,” he explained.
To combat some of this issue users can ask AI models to state their reasoning for their responses, but Akinwumi said this only addresses some of the problem. In order to confront the whole issue, real estate professionals must enact their training in fair housing to “make an informed judgment, which means that you want to make sure that you are asking for additional information that you can use to evaluate your decision about if or what to use out of the responses you’re getting.”
Building a safeguarding model
Akinwumi also dove into some of the NFHA’s research on the topic, noting that the organization recently audited several different AI models with several different buyer profiles in order to see the different verbal cues that AI may give out which indicate violations of the Fair Housing Act.
Through this research, Akinwumi said he and the NFHA have observed many instances of steering and discrimination in the responses from AI.
Out of its research, the NFHA built a model to help test and understand the ways in which AI can violate fair housing in its language. Akinwumi said that this specific model was designed to answer the question of if the recommendations coming out of AI are influenced through the knowledge of an agent or a client’s race, ethnicity or location.
The model has a user select a profile of location, race and whether you are a buyer or renter, then provides an auto generated prompt that will be submitted to an AI based on this data. Real estate professionals must then review the AI’s response for whether it is discriminatory, and explain their reasoning.
He invited agents to test out the model themselves and determine the verbal cues and language that indicate violations of fair housing, helping to prepare for future AI usage in their jobs in the process.
The goal ahead for the model is to be able to create a tool that, when prompted, can inform users of fair housing violations in their AI usage.
As for continued research, Akinwumi said the NFHA is continuing to evaluate and study AI models in order to seek out solutions for avoiding fair housing violations, expecting a white paper to be released in the next year outlining some of these solutions and recommendations.
