Alum Spotlight: Jay Budzik (PhD ’03)

Jay Budzik’s (PhD ’03) career in AI and machine learning (ML) had an auspicious beginning.

He joined the Illinois Math and Science Academy during high school in 1992, about a year after the first website launched. While still in high school, he was a research assistant at the Fermi National Accelerator Laboratory, where he recalls helping web usage spread after a visit from web inventor Tim Berners-Lee. He ported web servers and web browsers to UNIX platforms and helped Fermilab launch its first internal search engine.

In 1995, Budzik joined Kris Hammond’s Intelligent Information Laboratory (InfoLab) at the University of Chicago as an undergraduate research assistant, helping develop a web search engine that answered questions in plain English. The system was featured at the 1996 Democratic National Convention in Chicago.

In 1998, Budzik joined Northwestern as a computer science PhD student and Hammond joined as a visiting professor before becoming a full professor of computer science in 1999. Budzik was awarded a Presidential Fellowship for his research, the highest distinction awarded by the University to graduate students.

Now, Budzik is a senior vice president at Fifth Third Bank, a component of the S&P 500 and in the top 10 in the US by assets. He manages a team of AI scientists working on a variety of initiatives across consumer products and central operations.

How did your path lead from your Northwestern degree to where you are today?

At Northwestern, my focus was on applying AI to create new capabilities with impact. I’ve used those skills and ways of thinking throughout my career: find a problem, figure out how to measure progress, and use science to move the needle. I still ask myself: How can I use AI to solve a business problem and how will I measure whether it works? That process and thought pattern started at Northwestern.

At Northwestern, what was your AI and ML research focus?

Coming into grad school, Kris Hammond, Professor Larry Birnbaum, and the InfoLab were focused on showing what was possible and how AI could make an impact. We delivered what was probably the internet’s first question answering system: you could ask a question in plain English and get an answer. We made recommender systems that could suggest a restaurant in a new city based on examples of restaurants you liked at home.

The lab focused on helping consumers and businesses leverage vast information resources available online. We were part of an avant-garde in AI using statistical methods to drive intelligent agent workflows and systems. We focused on how to make practical systems that exhibited some level of intelligence without having to solve AI-complete problems, leveraging things like search technology and ML, with a bit of knowledge representation and classical AI mixed in, to make new services that delivered first-of-their-kind user experiences. It was exhilarating to be at the forefront. Our work was featured in the news and in the international press.

The idea that we could have impact by creating new technologies that would work in the world, that people could use, and that could have commercial applications in new industry categories captured my attention. Our work promised to help people gain broader perspectives and to create better outcomes for individuals, companies, and society. Northwestern supported our engineering mindset and had wonderful collaborators—it was such a fun time.

Was there a faculty member or project that still shapes how you think about the field?

A lot of my training was in symbolic AI, including inference, backward chaining, and knowledge representation. Early AI pioneers, some of whom are still on Northwestern’s faculty, had amazing vision, but the technical approaches of the 1980s and 1990s hit a ceiling. The internet offered vast troves of data, so we shifted toward statistical techniques, including ML and information retrieval, that could make use of that data. That allowed us to build machines that could automatically construct representations and enable intelligent behavior. We made the dream of intelligent assistants practical, even though their capabilities were somewhat limited compared to today’s standards. I carry that with me today.

My dissertation project was a software agent that looked at what you were reading or writing in Microsoft Office applications. It automatically analyzed the text and used clues from the structure of the document to identify key themes. It then found related information from a variety of backend systems such as web search, research libraries, personal files, and the company intranet. It evaluated each item to see if it was worth presenting. We built a flexible architecture to plug in different applications and enable different business logic. Our goal was to make knowledge workers more efficient and more aware of the information resources they have at their disposal. We founded a company to commercialize it.

At Northwestern, I learned how to invent and develop technologies that solve practical problems, and importantly, how to evaluate a system to determine whether it is performing at a level that would allow it to become commercially viable.

This is an engineering problem, and Northwestern is a great engineering school. Modern AI systems present many engineering challenges, including how to evaluate them and close the gap so that they can be used on more than just toy problems. My Northwestern experiences taught me to do this. In those days I worked closely with industry partners to identify use cases and understand how they would assign value to the capability. Then we would go back and prove it out in the lab and in the field. Once we had something ready, we had to tell the story of how the technology creates value. This gave me the background I needed to start, build, and exit several technology companies, manage turnarounds, and do what I do today.

What is your current professional focus? And what are you proud of?

I’ve started and sold technology companies in publishing and marketing services and now I’m working on applied AI in the financial services industry. Before joining Fifth Third, I worked with my team at Zest AI to invent a process to de-bias lending models, using adversarial training methods to reduce the negative impact to women and people of color from data reflecting historic bias and discriminatory practices like redlining (in which whole neighborhoods were excluded because of their demographics). By making credit scores fairer, we impacted trillions of dollars of money movement across hundreds of institutions and helped historically underrepresented and underserved groups receive their fair share. The models also were more accurate, which made the banks and credit unions who used them more money along the way.

At Fifth Third Bank, I now see that program operate from the inside, where we plan to safely expand our balance sheet by billions as a result of upgrading the models we use in our lending businesses to use AI instead of traditional methods. I’m also working on automating the many manual paper processes using the next-generation models, allowing us to expand and scale document processing in ways that we couldn't have contemplated before.

What's the one piece of advice you'd give a current CS student?

AI models are very capable, but don't try to be all things to all people. Focus on a specific problem that you're excited to solve that connects with something people care deeply about and measure your progress on how well you are able to solve their problem. When the problem is specific, you can use functional constraints to get a handle on how to solve it, you know how to evaluate your solution, and you can measure progress. When the problem is important, people will be excited to help you solve it.

Beyond core CS coursework, what skills do you think matter most for succeeding in an AI career?

Understanding the economics, both the cost of what you're doing and financial outcomes on the other side, is really important if you want to make your project successful and motivate people to work with you. Connect the technology to the business outcome.

Another key is storytelling and big-picture thinking. Narratives are powerful, and if you can cultivate the skill of connecting highly technical work to a narrative that people can understand and want to join, you will find that people will help and follow you. And, because Northwestern offers things like the Walter P. Murphy Cooperative Engineering Education (Co-op) Program, capstones, and client-centered practicum courses, students have a wealth of opportunities to develop and practice these important skills. Focus on how you can use your technical superpowers to make a positive impact and how you can make the connection between technology and impact easy for everyday people to understand.