Announcing new Chief Scientist & Head of Enterprise AI

As Chief Scientist, Prem will lead the technology strategy, architecture, and development for Capital One’s enterprise data, analytics, and machine learning initiatives, including advancing its AI capabilities, tools, and research efforts.  

Prem joins Capital One from Amazon’s Alexa organization, bringing more than two decades of experience leading science, technology, and commercialization efforts in natural language processing, speech recognition, computer vision, forecasting, and other machine learning (ML) applications. Following are excerpts from a conversation with Prem during his initial weeks at Capital One.

I’ve been fortunate to work at the cutting edge of AI and machine learning throughout my career. I began my professional path at BBN [now Raytheon BBN Technologies] where I had the opportunity to learn from some of the most respected leaders in speech and language technologies, and eventually became its Executive Vice President and Principal Scientist for Speech, Language, and Multimedia. Then I went to the University of Southern California to serve as a senior Vice Dean of Engineering in the Viterbi School and as the Executive Director of its storied Information Sciences Institute. I still maintain a faculty appointment in the department of computer science, and continue to advise PhD students. Most recently, I was VP at Amazon leading the Alexa AI organization.

What I enjoy most about working in this field is advancing technology and leveraging those advances to create new capabilities that make a difference in the everyday lives of people at home and at work. When I thought about the next frontier of challenges I wanted to tackle in my career, I kept coming back to the verticals where AI had the most potential to impact the most lives. It came down to a few specific industries, and ultimately finance rose to the top of my list. Money, and especially access to financial services, impacts all of our lives in an enduring way. 

Capital One has been known for its forward-leaning approach to data, analytics, machine learning, and software. As a customer of Capital One, I have been a fan of how they have been at the forefront of the digital transformation of the financial industry—for example, they were one of the first major companies to go all in on public cloud technology. I’d followed the company’s recent advances in AI and ML for years. And as a customer, I’ve long been impressed by how Capital One has consistently translated its investments in technology into delightful experiences for its customers. 

As I spoke with company leaders and associates and became more familiar with Capital One’s people and culture, it started becoming clear that at its core Capital One is a technology company with the skills and risk management of a bank. What also became clear is that every associate I spoke with was inspired by the mission of continuing to change banking for good, and improving the financial lives of millions of customers. The entire company lives and breathes that mission every day. The power of AI in combination with our ubiquitous focus on the mission presents an unprecedented opportunity to continue delivering positive change for all our customers.

Personal finances and banking are an important part of peoples’ lives.

But finances are ultimately an avenue to do so many other things that people find enjoyable and satisfying, whether it’s travel, planning for your kids’ education, caring for your parents, or buying a new car you’ve been saving up for. 

The fact that we can apply data, AI, and machine learning to understand our customers’ needs, goals, and pain points means we’re in an unprecedented position to continue to deliver the right help at the right time—to continue to deliver tremendous value to people in all different spheres of their lives, and we can do it at scale to over 100 million customers. 

I’m also inspired by the opportunity to continue taking a humane, human-centered approach to helping people with their finances through technology. Using data and AI to continue to better understand our customers, we can develop intelligent and dynamically adaptive approaches that serve their individual, unique needs. For example, we could help make personalized, precise recommendations when and where it matters most, at scale. And what’s more—harnessing the emergent capabilities of data and AI can provide the potential to translate those recommendations into actions that help the customers accomplish their underlying goals. I think that’s incredibly powerful!

Let me start with one of the first lessons I learned while developing and deploying conversational AI for call centers in the early 2000s: the fundamental design tenet for any advanced technology, especially AI, should be that the way it is applied in practice should progressively reduce the cognitive burden on human users by shifting it to the system. Even today, we see examples where poorly designed “intelligent” automated systems actually shift the burden to the users. To me, AI is ultimately about empowering people so they can spend more time on more satisfying, more high-value tasks. That tenet is important to keep front and center when we develop and deploy AI for our customers and for our associates who build our customer-facing experiences.  Bottom line—whether it is through AI or any other technology, the central element of our job is to make things simpler and easier for our customers and stakeholders.

Another important learning is that it is important to include a diversity of perspectives right from the start and in every stage of the AI lifecycle—from design, to experimentation, to testing, and deployment. That approach will result in the most robust, most performant, and most broadly useful AI-powered solutions we can deliver.

Historically, the performance of AI solutions has tended to degrade over time unless the underlying models are retrained on an ongoing basis. But recent advances in deep learning have unlocked the possibility of self-learning—a capability that allows AI systems to automatically learn through everyday interactions and not just maintain but improve their performance over time. This is one of the most exciting emergent areas for industry applications of AI in my mind. Harnessing this opportunity requires companies to take a thoughtful approach to designing an overall framework, so that every interaction improves the models, and every improvement drives greater customer satisfaction across a range of use cases.  

Capital One has already made great strides in its technology journey, including in the data, AI and machine learning spaces. I am looking forward to continuing to build out a world-class AI organization and infrastructure that anticipates and prepares us for the future. The key priorities as I see them are: 

• Accelerating the delivery of new and differentiated experiences for associates and customers
• Continuing to strengthen the data- and technology-focused culture that already thrives at Capital One 
• Ensuring Capital One remains a prime destination for the top AI and tech talent

I am also energized by the opportunity to continue to build out a world-leading enterprise AI framework and a robust, scalable foundation for fast development and delivery of meaningful customer-facing experiences and features.

I am passionate about enriching the diversity of talent in technology organizations, including promoting national, impactful models for industry and academic partnerships to advance AI research and training programs. A great example of this is the Columbia University and Amazon Summer Undergraduate Research Experience (SURE) Program that I helped launch while at Amazon. In AI, more representation, diverse background and experience, and multi-sector collaboration across AI will lead to more innovation, greater responsibility, and more powerful outcomes. Capital One has an array of applied AI/ML university research partnerships, and I am very excited about the opportunities for strengthening and expanding this work.

Natural language processing (NLP) has been an enduring thread of interest throughout my career. And over the past few years, NLP and LLMs have perhaps become synonymous in popular perception.

First and foremost, I am most interested in the potential that LLMs have to transform every aspect of our professional and personal lives—from how we develop code and applications, to how we discover and consume information, to how we interact with systems and even with our environment. Ultimately, AI is about empowering people to do things that they couldn’t do before, or that were just far too burdensome to do.  And it is about democratizing access to opportunities and capabilities - whether it is in finance, or education or healthcare, or any space.

But to paraphrase the parable of Damocles, with the great power of LLMs, comes the great responsibility of using them appropriately. One of my Ph.D. students’ research in this area drove important new advances in characterizing the biases and prejudices of large language models and how to improve them. Her work expanded my own understanding of this aspect of LLMs. Over the past few years, we’ve made big strides in understanding the challenges and gaps in this space, and things have improved a lot, but the problem is nowhere near solved. This area is one of those where interpretable and explainable AI becomes particularly important; especially in the form of readily usable tools and diagnostics we can use to get insight into the models’ inference processes and better understand how to improve their performance.

Another opportunity that excites me about advances in emerging AI is about bringing behavioral and transactional aspects of customer interactions to foundation models. Such models hold the promise of transforming so much for the customer in ways that haven’t been possible before. There are also fascinating challenges at the intersection of AI, machine learning, and experience design. By the way, when we talk about shifting the cognitive burden from the human to the AI system, the affordances that are created through flexible, agile user experience design form an important lever in optimizing that shift.

When executed responsibly and effectively, AI has the potential to help each of us find our full potential. For example, if you love music, perhaps generative AI could help you become a world-class singer (at least, I hope it does that for me one day!). 

Listening to music, spending time with my family and friends, and mentoring people inside and outside of work—all of these are activities that energize me and keep me busy.

We’ve moved into a world where it’s not necessary to “break into” a career in AI, per se. All of us either already are or we soon will be interacting with AI in some capacity in our daily lives, and many of us are already doing that in our jobs. So if you’re really interested in this space, my main advice is to identify your passion and look into how AI can enrich your pursuit of your passion.  

For example, if you are an educator, you may want to talk to experts and identify ways in which AI can be a leveler that helps improve educational outcomes for all students.  More broadly, whether your passion is journalism, creating novel content, growing a business, or advancing medicine—you should be actively thinking about how to responsibly harness the power and advances of AI to accelerate that passion. In this new phase we’re entering, it’s important to stay true to your passions, to what energizes you, while remaining open to learning—even actively seeking to learn.  

I am deeply optimistic that broad societal engagement with AI will have a democratizing impact on the development of future generations of AI technology, ensuring broad access to the benefits of AI.

And, I’ll just say, if you want to do all of these things: deliver business value and exciting new customer experiences, be on the forefront of change and opportunity, and work to deliver something better for the world—come work with us! My organization at Capital One is hiring.