Skip to main content

Not All AI Is Developed Equally

May 30, 2025
Technology
Special to National Mortgage Professional

Be wary of new systems that rely on basic consumer-focused AI systems

Whatever your views on Elon Musk, he has clearly mastered using technology to build businesses.  In fact, in a recent interview on CNBC, Musk highlighted that only customized and highly complex neural network-based systems have the capacity to deliver utility in high-stakes use cases.   

A neural network is a type of machine-based AI learning that organizes and intelligently connects nodes of data to better allow the system to gather, interpret and make decisions. This concept is not new, but with the development of high-speed chips, this technology has witnessed massive advancements, which of course, generates market excitement.  

Mortgage lenders, too, are excited and have rushed to enter the AI race, but most lack the scientific foundation and experience to engineer complex AI systems. Some lenders, who are trying to spur investments into their companies, are touting new AI initiatives which are a mere repackaging of an LLM such as Google's Gemini and, therefore, are ‘superficial’, quoting Mr. Musk. 

One exception to this is AngelAi, which I created as a  neural network-based AI mortgage system in 2018. The long, hard road to the ‘self-driving’ cars is analogous to the development of the ‘self-driving’ mortgage: both arrived after decades-long investment in neural net research and development.  

With hundreds of thousands of registered users generating millions of monthly hits, AngelAi is proven, and currently powers most of the loans written at Sun West Mortgage.  As other mortgage lenders rush to say they have an AI solution, let us point out that creating a mortgage AI is more complex than just knowing what prompts to send to ChatGPT, Gemini, Llama, or CoPilot. Not all AI is created equal, and the assumption that every large language model (aka LLM, generative AI, GenAI) can assess risks and process loan applications is a fundamental misunderstanding of how AI works. 

The Misconception: One AI To Rule Them All 

Too often, AI is spoken about in sweeping, generalized terms, as if every system is interchangeable. Some fear that AI will take over human decision-making in critical fields like finance and approvals, while others assume that any AI capable of generating text must also be able to assess mortgage applications. This conflation is misleading and dangerous. A recent paper highlighted how LLM-based AIs exhibit racial bias in mortgage underwriting decisions. 

For over 40 years, we have been first to the market with consumer empowering mortgage technology, and we have invested hundreds of millions of dollars in Ai to ensure it removes all traces of bias. Now, we are proud to bring it to the marketplace. We have achieved “self-driving” mortgages by creating neural networks, which are capable of highly complex problem-solving, vs using LLMs/GenAI, which are trained on massive data sets and are prone to hallucinations and are engineered for creativity. We explained it in depth during these two podcasts, linked here and here

Reality: Specialization Matters 

AI models are trained for specific tasks. The large language models (LLMs) powering conversational AI — like chatbots or search engines are optimized for natural language processing and text generation. They don’t have built-in capabilities to assess creditworthiness, or navigate the complexities of underwriting, or to drive a car as Musk has explained. 

Meanwhile, AngelAi’s model has been developed with complex neural networks for fraud detection, credit risk assessment, document review, and automated underwriting and trained with financial regulations, data integrity, and predictive analytics. That’s why we can provide an exclusive 100% warranty and stand by every response. 

These distinctions are crucial. A chatbot can help answer general and broad consumer questions about mortgage matters but cannot replace the decision-making frameworks lenders use to evaluate risk. The self-driving car problem is high-stakes and analogous to the challenges faced in the mortgage industry, in which, also, there is no room for mistakes.   

Every seasoned loan officer knows to expect the unexpected. Think of Ai facilitating a self-driving mortgage where the passenger -- in this case, the borrower -- is taken from point A of the loan application to point B, loan closing, autonomously, while rapidly reacting and adjusting to different and unexpected events that happen along the way. The Ai needs to be error-intolerant and instantly react with absolute precision.  

The Risks Of Misapplying AI 

Large language models cannot provide a transparent, traceable decision-making process — a fundamental requirement set by the FDIC last year for banks leveraging AI-driven decisions. The FDIC guidance states that any bank relying on AI must be able to justify how the system arrived at its conclusions, ensuring accountability and compliance. In high-stakes applications where life, property, or financial security are at risk, AI cannot operate on the principle that outcomes alone justify the methods — it must be explainable and auditable. 

The Path Forward: Responsible AI Development 

For the mortgage industry, the focus should not be on whether AI can replace human lenders, but rather on how specialized AI can support and enhance lending processes. 

Smart AI tools — trained on financial data and designed within regulatory frameworks—can improve fraud detection, streamline underwriting, and enhance customer experience. But understanding the limitations of AI is just as important as recognizing its potential. The future of AI in mortgage lending depends on informed discussions, responsible implementation, and clarity in its capabilities. Beware those who are setting goals and rushing to keep up.  

 As we shouldn’t expect a generative-AI-powered chatbot to drive a car, we shouldn’t expect a LLM /GenAI to navigate consumer through the largest investment of their lives and to make sound lending decisions. The future of AI in mortgage lending depends on informed discussions, responsible implementation, and a clear understanding of its capabilities. 

About the author
Pavan Agarwal is the founder/creator of AngelAi and CEO of Sun West Mortgage.
Published
May 30, 2025
More from
Tech
AI And The Mortgage Revolution

Chris Burks, Head Of Growth at Zeitro, on embracing change

Jun 26, 2025
Fed Chairman Aligns More Closely With Trump Admin On Crypto

Powell said he sees ‘change in tone’ around digital assets, expects more crypto activity

Jun 26, 2025
FHFA Chief Orders Fannie And Freddie To Prepare For Crypto In Mortgage Underwriting

In sweeping declaration on social media, FHFA Director William J. Pulte says Fannie Mae and Freddie Mac must begin preparing to count cryptocurrency as an asset in mortgage underwriting

Beware The Latest ‘Long Con’ Scam

Real estate professionals of all kinds warned to beware of crypto scams exploiting ‘trust relationships’ with clients

Jun 23, 2025
ServiceLink Adds Three To Its National Sales Team In Origination Division

Chris Miller, James Lee, and Mandy Powell to help forge new lender partnerships, drive client growth

Jun 23, 2025
“Technology” is a Catch-All Word

Tech promises efficiency, but adoption, ROI, and borrower satisfaction tell a more complex story

Jun 19, 2025