Ghosts in the Model: AI, Ontology, and the Risk of False Insight

About this episode

What happens when the mortgage industry rushes headlong into AI… without first agreeing on what things actually mean?

In this episode of MISMO Mic’d Up, I’m joined by Greg Alvord for a conversation that goes well beyond buzzwords and vendor decks. We dig into the foundational question most AI discussions skip entirely: How do we know what we’re measuring, and why should we trust the conclusions?

Greg brings decades of experience at the intersection of data, modeling, and mortgage technology, and he doesn’t shy away from challenging comfortable assumptions. We explore how two of the primary mathematical tools behind modern AI—linear regression and neural networks—can both mislead when variables are poorly defined, inconsistently labeled, or chosen simply because they’re easy to capture rather than meaningful to outcomes.

A key theme throughout the discussion is ontology: the disciplined practice of defining concepts, relationships, and meaning before attempting automation or intelligence. Without shared definitions, AI systems can surface patterns that look impressive but are statistically fragile—or worse, entirely coincidental. More data doesn’t automatically mean better insight, and more variables don’t guarantee better predictions. In fact, they often increase the likelihood of false confidence.

Greg walks through why this matters so deeply in mortgage lending, where decisions impact real people, real money, and real regulatory obligations. We talk about how “ghost signals” can emerge in neural networks, why explainability is not optional in a regulated industry, and how inconsistent data definitions quietly undermine even the most advanced tools.

From there, the conversation turns constructive. We discuss how industry standards—particularly those developed through MISMO—provide the scaffolding AI actually needs to scale responsibly. Shared data models, common definitions, and agreed-upon semantics aren’t a brake on innovation; they’re the runway. They enable lenders, vendors, regulators, and investors to move faster together without introducing unnecessary risk.

This episode also explores the human side of AI adoption. Technology doesn’t replace judgment—it amplifies it. That amplification can be powerful or dangerous depending on the integrity of the inputs. Greg offers a grounded perspective on how experimentation, hypothesis testing, and intellectual humility should guide AI development, rather than blind faith in algorithms.

If you’re a lender executive, technologist, compliance leader, or product builder trying to separate real AI progress from statistical noise, this conversation is for you. It’s a reminder that the future of mortgage technology won’t be built by models alone—but by models rooted in clarity, standards, and shared understanding.

Because before AI can be intelligent, the industry has to agree on the language it speaks.