Old roots and new branches in AI

In one room, I watched AI sift through vast amounts of data, guiding an underwriter to a faster, sharper decision. In another, I watched a banker win a client's trust with kind words and empathy no algorithm could replicate. Both moments happened at Rocket Mortgage, and together they crystallized something I'd spent years circling in academia: the companies that win with AI won't be the ones that adopt the most. They'll be the ones that match the right tool to the right problem, combining generative AI, traditional AI, automation, and human judgment with intention instead of chasing the hype.

A front-row seat to the AI debate

My name is Ken Reid, and I'm a Senior Data Scientist here at Rocket. I hold a Ph.D. in computer science and focused my research on real-world operations research powered by AI techniques known as metaheuristics. Before Rocket, I had a stint as a genomics researcher, bringing AI expertise to one university, then I helped as a data scientist bringing GenAI to the researchers at another. Across those roles, I had a front-row seat to a wide spectrum of AI perspectives, and one conviction emerged above all: The most powerful thing about AI isn't any single technique. It's knowing which tool to reach for, and when.

In academia, I watched the AI debate play out at its extremes. On one end, critics argued that large language models (LLMs) were fundamentally harmful: environmentally costly, intellectually corrosive, and a threat to rigorous thinking. On the other, many insisted we need bigger, more powerful LLMs at any cost. But there is a third camp: researchers pointing to models becoming smaller yet more capable, far less damaging environmentally while trailing only slightly behind the bleeding edge of performance. To borrow an analogy: We don't need to return to wood-burning stoves and candles, nor do we need carbon-emitting coal and oil plants. We want the middle ground of hydro, solar, and wind power. Small, efficient LLMs are our renewable energy.

But what struck me most was watching what happened when researchers stopped arguing about AI and started working with it. At its best, AI wasn't a replacement for thought: It was a catalyst and a constructive adversary. I saw this firsthand when I co-authored a paper on the early effects of LLMs across a dozen disciplines with a dozen scholars. The sciences, humanities, and business all found unexpected value when they engaged thoughtfully rather than dismissively. That experience shaped a conviction I carry into my work today: The benefits of AI are greatly amplified when we master its diverse range of capabilities responsibly, guiding it with our expertise.

The dichotomy of old and new

My own research lives in what people now call "traditional AI": rule-based systems, optimization techniques, metaheuristics, and structured approaches that predate the current generative AI wave. These methods are the foundations of where AI is today and the backbone of government, industry, and infrastructure

Traditional AI gives us precision, explainability, and mathematical rigor, often at a fraction of the computational cost. Generative AI gives us flexibility, creativity, and the ability to work with unstructured information. They are more powerful and responsible together than alone. The wisest approach is to reach for traditional methods where they excel and deploy right-sized generative models where flexibility matters.

Seeing the dichotomy in action at Rocket

Here at Rocket, I've seen first-hand what careful, intentional AI investment looks like in practice. It's not hype-driven adoption or reflexive resistance. It's the harder work of balance: matching the right tool to the right business problem.

I had the privilege of shadowing team members across different parts of the business I would otherwise never see, and what I witnessed reinforced my beliefs.

Human expertise amplified by AI

Shadowing Rocket Mortgage’s Triple Crown Rocket Ready Underwriter Korey Streck showed me how tools like Pathfinder and Rocket Logic don't replace human judgment: They amplify it. Korey's expertise and attention to detail remained central to every decision. The AI handled the heavy lifting of information retrieval and pattern recognition, freeing Korey to focus on the nuanced, human elements that truly matter in underwriting. It was a partnership, not a handoff.

I saw the same dynamic with Rocket Mortgage Triple Crown Bankers Matthew Sitto and Sam Shabrang. Watching them work with advanced communications software and Rocket Logic was like watching skilled musicians with finely tuned instruments. Technology handled routine tasks and surfaced relevant information instantly. But the empathy, the problem-solving, the genuine care for clients came from Matthew and Sam. The tools made them more efficient, but compassion made them effective.

Traditional AI doing what it does best

On the quantitative side, our Trade Data Science Team builds hedging models using traditional machine learning and statistical approaches. These aren't flashy. They're precise and reliable. When you're managing risk at scale, you need models you can explain, validate, and trust. The old roots run deep for good reason.

The same goes for Best Execution, where exact solver optimization models determine the best execution strategies. These are classic operations research techniques: mathematical optimization that guarantees optimal solutions within defined constraints. No hallucinations. No ambiguity. Just rigorous, provable results. Simplicity is genius. Sometimes the most elegant solution is the one that's been working for decades, and that success is measured in tens of millions of dollars.

This is the dichotomy in practice: generative AI where flexibility and language understanding matter, traditional AI where precision and explainability are non-negotiable.

Being honest about the hype

The examples above prove that our balance is working. But across the broader landscape, that balance is under pressure. Generative AI dominates the conversation right now, capturing headlines, investment dollars, and imagination. Much of that excitement is warranted. But it has also created a distraction. I've seen institutions reach for LLMs to do the work of optimization, prediction, and structured analysis when smaller, more precise, and more efficient tools already exist and do it better.

Thoughtful AI adoption also improves client experience, and the consequences of getting it wrong are real. Multiple cases across industries highlight the risks of deploying AI systems without appropriate guardrails, oversight, and clarity when simpler solutions, such as rules-based systems, may be more reliable.

The techniques that don't make headlines are often the ones that make money. Traditional machine learning, optimization, metaheuristics, and data science have a proven track record of delivering precision and reliability, and they are constantly being refined by both academia and industry. When everyone is looking at the newest technology, it's easy to overlook the proven tools quietly delivering results and keeping wheels turning every day.

Across any organization, years of experience carry well-earned respect. The people who have been doing this work for years have pattern recognition and contextual understanding that no model can replicate. But even the sharpest intuition has limits: it can't easily account for thousands of variables shifting simultaneously, detect slow-moving trends buried in noise, or scale one person's judgment across an entire operation. This is exactly where traditional AI can serve as a collaborator, surfacing patterns the human eye might miss, pressure-testing assumptions in real time, and providing evidence to either confirm or challenge a gut feeling. The question isn't "Should we trust data or experience?" It's "What if our most experienced people had even better tools to validate what they already suspect?"

But two forces work against this. The distraction of GenAI hype pulls attention toward the newest thing, while the comfort of "we've always done it this way" resists change entirely. They pull in different directions, but they lead to the same outcome: we leave performance, efficiency, and money on the table.

At Rocket, we have powerful GenAI tools, and I've seen them in action. But what sets Rocket apart is the commitment to matching that power with equal investment in the broader AI and data science toolkit: optimization, predictive modeling, statistical analysis, and the culture to use the right tool at the right time.

The real competitive advantage

The real competitive advantage in AI isn't having the most advanced models; it's having the judgment to match the right tool to the right problem. Across any industry, decisions are made every day that could benefit from optimization, predictive modeling, or structured analysis. These problems don't appear on anyone's AI roadmap because they've never been framed as AI problems. They might look like scheduling inefficiencies, pricing questions, resource allocation challenges, or process bottlenecks. Solving them doesn't require a large language model. It requires people who understand the full breadth of what AI and data science can do and have the curiosity to look for those opportunities wherever they exist.

The Rocket way: Ethical and forward-moving

What ties all of this together is culture. At Rocket, we don't adopt technology for its own sake. We adopt it to help people: our team members and the clients we serve. Every tool I encountered was deployed thoughtfully, with guardrails in place and humans firmly in the loop. We do the right thing, even when no one is watching. This is captured perfectly in our quality strategy hub: “AI drafts, automation enforces, humans decide.” It is a simple rule with a deep philosophy behind it: Let generative AI, traditional AI, and automation each do what they do best, and keep human judgment firmly in command.

The team members I shadowed weren't competing with AI. They were collaborating with it. Korey's judgment is guided by Pathfinder. Matthew and Sam's compassion is amplified by software. The Trade Data Science team's precision is built on proven statistical foundations. Best Execution's rigor is rooted in decades of operations research. Each one a testament to what happens when the right people meet the right tools.

That collaboration reflects something deeper about how we work here. We stride forward boldly, but we do it ethically and intentionally. We are trailblazers. And trailblazing means more than chasing the newest technology. It means having the knowledge and experience to use what works, the curiosity to explore what's next, and the wisdom to know the difference.

The road ahead

The companies that will lead with AI in the years ahead won’t be the ones that adopted the flashiest tools. They’ll be the ones who built teams capable of reaching for the right approach, whether that’s a large language model, a statistical model, an optimization solver, or a simple rules-based system, then pairing it with automation and human judgment to bring it all together responsibly.

That’s what excites me about working at Rocket. It’s a place where a researcher can shadow a Rocket Mortgage underwriter and see AI augmenting human expertise in real time. Where generative AI and traditional methods aren’t in competition; they’re complementary parts of a coherent strategy. Where the culture doesn’t just tolerate this kind of balanced thinking – it demands it.

The future of AI isn’t about choosing between old and new, or between machine and human. It’s about responsibly combining generative AI, traditional AI, automation, and human judgment, then trusting the right people to know which to reach for, and when.

We’ll figure it out together.