Larry Deschaine: From Hype to Engineering Reality Q&A

Q1. You've built a huge body of work around hybrid classical–quantum AI — what's the clearest sign you've seen that this is becoming practical, not just experimental?

A. The clearest sign I see is that we don't have to invent the infrastructure anymore. Last year, when I watched students and interns build a working hybrid system in a ten-week period, I knew we'd crossed a threshold. And industry partners are asking better questions — less "can quantum solve this?" and more "where in our workflow does it actually help?" That shift from hype to solid engineering is encouraging.

Q2. Where do you think people are getting a little too optimistic about quantum AI right now — and where are they not paying enough attention?

A. I question when "quantum advantage" gets treated as a metric or focused milestone, because its value is often domain-specific; a nuance that gets lost in the headlines. I focus on hybrid classical-quantum systems, paying close attention to classical orchestration and the quantum steps. Classical methods solve equations that represent physics, whereas quantum computations solve the quantum system. I'm more pragmatic about using the right technique to solve the problem at hand, regardless of the technique, including understanding and quantifying the uncertainty in the answers.

Q3. You've applied this across so many areas — climate, manufacturing, health — was there one application that really surprised you?

A. What surprised me most was how naturally these hybrid methods fit the natural sciences — tornado formation, hurricane intensification, and air pollution release migration. These are problems in which every additional minute of warning saves lives, and they involve complex, coupled physics that leverage the best of the classical and quantum methods integrated into solutions.

Q4. From your experience, when does a hybrid classical–quantum approach actually start to outperform a traditional model?

A. The traditional model is the classical model, which becomes part of the hybrid approach. The quantum parts can supplement classical models by providing unique features and insights, especially through entanglement and superposition viewpoints. First, the developers are instructed to code up an initial classical solution that serves as the performance floor. Then they assess whether applying quantum can add value. What I tell them is to apply quantum to a problem when it makes sense, where you can rationalize why a quantum step could add something the classical approach can't get to as efficiently or effectively. And if you're unsure, code up some test cases. Part of the art of this is knowing where a quantum computing approach can be valuable and then testing it to verify those insights.

Q5. When people hear "quantum in production," it still feels abstract — what does a real, working system look like today?

A. It looks like any serious engineering project. The hybrid systems I build combine classical and quantum methods, integrate their functionality, and often blend their outputs using an ensemble. The quantum parts are boxes in a system diagram.

Q6. You've worked with frameworks like PennyLane, Qiskit, and Cirq — how do you think about choosing between them?

A. In my courses, many capstones are intentionally built across all three frameworks because each one has some distinct capabilities and intricacies. I think it's important for developers to be prepared by being fluent in multiple frameworks; they'll likely encounter all of them and others during their career, depending on the industry, use case, hardware, and the development team. We build, we benchmark, and we cross-implement. Which one gets used is a decision made based on the problem at hand. In many of my capstones, I benefit from using all three frameworks for different tasks, rarely relying on just one.

Q7. If someone is a data scientist or engineer today, what skills should they start building to prepare for this space?

A. On the classical side, the foundations are linear algebra, machine learning, optimization, probability and statistics, and the physical and engineering basics of the use case. The quantum side adds its own foundation — enough quantum mechanics training is needed to genuinely understand superposition, entanglement, amplitudes, shots, and measurements. From there, one needs to know deeply how a quantum circuit transforms information, qubit by qubit, gate by gate, shot by shot. Quantum intuition is earned through working problems and following the math even when it's slow — there's no shortcut to it. It's a part that no one can hand to you. Understand what the experimental results reveal. Also, study the historical thoughts of the quantum developers, understand the great discoveries and the conflicts – back then and now. This will help one frame one's mind to work intelligently and knowledgeably in this field.

Q8. What's a common misconception you see when people first get into quantum machine learning?

A. The biggest one is that quantum machine learning is "regular ML, but on a quantum computer." It isn't — quantum models capture patterns differently, see differently, sense differently, learn differently, and fail differently, and designing a useful one is closer to a physics experiment than tuning a deep classical network. The intuition you need is earned by following the math and understanding the experimental observations, even when it's hard, mind-bending, and somewhat confusing. Develop your own set of codes to gain understanding, then deep-dive into the results by modifying and experimenting with them.

Q9. As you've benchmarked quantum against classical methods, where are you starting to see early signs of real advantage?

A. I want to be careful with the word "advantage" because it often gets abused. My work is hybrid, so it's not a matter of classical or quantum, but rather classical and quantum. What I see is that after developing 40+ capstones across numerous fields, the ones I presented to the VICEROY Scholars last week, show reproducible, statistically defensible improvements using the hybrid approach. But I'm not using the buzzwords "quantum advantage" or "quantum supremacy". I'm looking at rigorous analysis of how and where to apply these tools to generate improvements where the classical approach breaks down. In finding application spaces where computations and decisions developed using hybrid approaches are valuable and practical. In the use cases I've been exploring, the hybrid approach often delivers the best performance, showing the right kind of progress for the field right now.

Q10. How does your faith support your work at the forefront of emerging technologies?

A. My faith is the foundation. Growing up, I learned how many Catholic clergy contributed to science — such as Gregor Mendel, who contributed to genetics, and Georges Lemaître, who developed what is now known as the "Big Bang" explanation of the universe's beginning. As a Native American Catholic clergy, I think a lot about stewardship — that we're responsible for what we build and how it affects people and creation. I look towards quantum computing to understand how nature, how creation, works. When I'm working on tools that forecast severe weather or defend against cyberattacks on a power grid, I'm not just optimizing a loss function; I'm thinking about families in the storm's path, and about patients on ventilators when the grid goes down. I've never found faith and rigorous science to be in tension, and I rely on and need both to do my work well.

Q11. Looking ahead, what do you think will matter most over the next five years — better hardware, better algorithms, or better hybrid approaches?

A. All three matter, and they're more dynamically interconnected than people sometimes recognize — better hardware enables developing and running algorithms that weren't practical or possible. Developing better, stronger, and more resource-intensive algorithms for larger or more complex applications reveals where we need hardware improvements. Better hybrid practice exposes where the binding constraint is. At any given point, one or more of them will be on the critical path; right now, it's all three. They feed each other. What matters is that we keep working on all three in a synergistic, resolute way, cross-pollinating the needs, pain points, and lessons learned.

Q12. You are presenting at Quantum.Tech this year — what are you looking forward to learning, and who are you looking forward to meeting?

A. What I'm most looking forward to is cross-industry conversation, it's a place where pharma, logistics, energy, defense, and others are all asking similar questions about where this technology fits their use cases. On the people side, I want to spend time with the practitioners running real production pilots and applications, the people – like me – who live in the quantum trenches, who have succeeded and failed. The conversation I most want to have – most near and dear to my heart - doing this almost 50 years now - is about the workforce pipeline. I feel that quantum training should start in high school, as mine did, or ideally earlier. And I want to talk about mentorship; every meaningful step in my career came primarily from someone who poured inspiration into me early on in my journey. That meeting I had with Richard Feynman at the 1981 MIT Endicott meeting, the meeting where quantum computing was launched, was a day that changed my life. I want to reach out to find and encourage people committed to doing the same – mentoring and encouraging the next generations.