DSNews delivers stories, ideas, links, companies, people, events, and videos impacting the mortgage default servicing industry.
Issue link: http://digital.dsnews.com/i/1505571
MortgagePoint » Your Trusted Source for Mortgage Banking and Servicing News 8 August 2023 M T E C H AI, and it is looking like big changes may be coming in the future if the trend continues rocketing up as it has been since the first AI bots have been released. According to STRATMOR, a mortgage consulting firm, there is no version of the fu- ture that does not include AI. In their words, "that genie is out the bottle, and the only thing that will determine whether it will be good or ill will be the wishes we make." But AI does not only serve the good; fraudsters will instruct AI to do their bidding as well, which unfortunately will make them more effective and more of a problem. Knowing this, STRATMOR surveyed a group of experts on the trend, including the poster child of AI—ChatGTP. What emerged offers a glimpse into what may be our best- case scenario for how we will share the future with these powerful new technologies. AI is a digital view into our future, ac- cording to STRATMOR. "At the risk of telling you something you've heard a few million times now, ChatGPT is a 'language model trained to produce text' that helps the user compose content, from emails to essays, in response to questions you ask it." "This AI is a result of OpenAI, a nonprofit organization funded by some of the wealthi- est technology companies and leaders in the world, including Elon Musk and Microsoft. Together, they have spent billions on their digital child," STRATMOR continued. "Their investments paid off, resulting in the fast- est-growing consumer app in history. Today, users everywhere are using it for everything from creating complete software apps to authoring complete novels and screenplays to writing breakup letters." There are mountains to climb to imple- ment AI in the mortgage industry. ChatGPT offered this shortlist: » Data Quality and Availability: AI algo- rithms rely on high-quality and relevant data for training and accurate predictions. Ensuring data quality, integrity, and acces- sibility can be a significant challenge when implementing AI. » Regulatory Compliance: Integrated AI systems must navigate compliance with laws such as fair lending practices, privacy regulations, and anti-discrimination laws. » Explainability and Interpretability: Developing AI models that are interpre- table and transparent is a challenge in the mortgage industry. » Change Management and Workforce Adoption: Resistance to change, lack of AI literacy, and potential job displacement concerns can hinder the adoption and implementation of AI technologies. » Model Bias and Fairness: Ensuring fairness and mitigating biases in AI models used for mortgage originations is essential to avoid potential legal and ethical issues. Develop- ing bias-free models and addressing biases in data sources pose significant challenges. "We talk about 'automation' and 'AI' as the same thing, which it is not," STRATMOR Principal Jennifer Fortier said. "'Automation' means taking the human out of routine repetitive tasks and 'AI' means simulating human thinking. So, when we talk about AI features, my mind goes to 'what human-like thinking is it doing?'" No one can say how long it will take AI to move into the underwriter's office, but we know that's not where it will start. We know this because the AI takeover has already begun. AUTOMATED VALUATION MODELS AND THE EFFECTS OF HISTORICAL REDLINING V eros Real Estate Solutions has released new research that examined the potential for bias in home valuations generated by their VeroVALUE℠ Automated Valuation Model (AVM) when analyzed across boundaries in historical redlining maps. Veros' study examined AVM predictions for single-family properties in redlined versus non-redlined neighborhoods close to one another in Los Angeles, California. Veros' economists used this approach because mortgage funds were historically abundant for non-redlined areas, while they were scarce or nonexistent for redlined areas. To enable proper testing, the areas selected for analysis had single-family homes on both sides of the boundaries. Statistical multiple regression analysis was used to investigate which physical home attributes were responsible for the differences seen in the VeroVALUE AVM estimates. Of particular interest in this analysis was the role of a location variable (whether the residence was in a redlined area or not) as a determinant of housing prices. "After controlling for the physical attributes of homes, the VeroVALUE AVM returned comparable estimates for properties on either side of the redline," said Reena Agrawal, Veros' Research Economist. "The research also found that a modern-day outcome of historical redlining is that homes in neighborhoods that were redlined generally have smaller lot sizes and living areas compared to homes in non-redlined areas, leading to lower median values than larger homes." The homes valued lower on average in historically redlined areas are often the result of property characteristics such as smaller living areas and lot sizes, not due to a lingering bias due to historical redlining. The study was designed to consider whether all valuation solutions are biased because of historical redlining, as some have suggested. AVMs offer a distinct benefit because they do not rely on any data related to historical redlining maps or demographic information concerning the parties involved in real estate transactions. However, there is the suggestion that any property valuation offering may be unwittingly influenced or biased through the injection of historically biased data rooted in decades-old discrimination. "In an environment where housing finance stakeholders consider both accuracy and fairness across the entire valuation spectrum, our VeroVALUE AVM is a proven and invaluable tool to achieve both goals," said Eric Fox, Veros' Chief Economist.