Vassilis Digalakis Jr. joins CISE as he tackles AI’s biggest flaws
In today’s rapidly changing artificial intelligence landscape, Assistant Professor Vassilis Digalakis Jr. (Questrom (OTM)) is building a different kind of future.
Digalakis, who joined CISE as a Faculty Affiliate in February 2026, isn’t just interested in making AI faster or more powerful. For him, the real challenge is making it human-like. He views AI as a system of values that needs to be carefully engineered.
“There is no agreed-upon definition of trustworthy AI,” Digalakis noted. “What I like to do is view trustworthy AI as a set of core principles that AI systems should have.”
For him, that means a checklist of values, including accountability, transparency, fairness, safety, robustness, and human oversight.
His journey to the forefront of “Responsible AI” began at the Technical University of Crete, where he studied Electrical and Computer Engineering. During his PhD at MIT, however, he pivoted into Operations Research. Digalakis described the field as the bridge between mathematics and the reality of business and society.
The convergence between his technical tools and Operations Research training allows him to move beyond just writing code to studying the entire pipeline, from the moment data is collected to the second a human makes a decision based on it.
Studying the end-to-end process shows that a model’s success isn’t measured solely by its predictive power but also by its reliability in real-world situations.
“In the fields of machine learning and AI, you typically have this mindset that you’re building an algorithm, you study it, you perhaps build theory around it,” he explained. “But more often than not, it’s kind of disconnected from the broader society. In operations research, you study the entire pipeline from data all the way to decisions.”
One of the most pressing problems Digalakis addresses is what he calls “structural stability.”
In high-stakes fields like healthcare, AI models are often retrained as new data comes in. However, traditional models can be unreliable—a small change in the data might cause the AI to completely change its logic.
This instability arises because most models are designed to maximize accuracy at any given moment. Models often over-adjust to minor noise or outliers in a new dataset. When it tries to optimize for these small points, it may take a completely different path to reach a conclusion.
During his research working with doctors, Digalakis saw firsthand how this change undermined trust.
“If the model changed unexpectedly, the stakeholders, the physicians in this case, could see the change. So they would audit us, and ask, ‘Why did this happen?’”
For example, these shifts could mean that a patient who was flagged as “high-risk” one day for certain symptoms might suddenly be downgraded to “low-risk” the next day. Even if their medical charts hadn’t changed, it looks like the AI model was changing its diagnosis. When the logic changes, practitioners may be hesitant to trust the model.
To solve this, Digalakis develops algorithms that are “stable by design.”
These models are engineered to evolve smoothly over time, changing their logic only when necessary. “Stability constraints” are incorporated into the learning process, which penalize the model if its structure or parameters deviate too drastically from its previous version without a significant boost in accuracy. In a hospital setting, that consistency between models can be the difference between life and death.
At BU, Digalakis is now passing this on to his students through a course on Data Ethics and Responsible AI.
His goal is to produce graduates who don’t just know how to use AI, but who know when to question it.
“I want them to remember that when they’re dealing with AI systems… they should always view what they do with a lot of critical thinking,” he says. “You should definitely not blindly trust their outputs.”
He emphasizes that AI models are mirrors of the data they are given.
By teaching his students to recognize these flaws, he is preparing them to pursue careers in which AI is a tool for progress.
As AI continues to extend into more areas of daily life, Digalakis is looking toward AI governance.
“Often we build great models, but we have no established and holistic framework for deciding what model to deploy, when to retrain the model, and when to change the model,” he observes.
For Digalakis, the goal remains to ensure that as AI systems become smarter, they also become more reliable and more trustworthy.
By shifting his focus from raw power to human interest, he is helping to ensure that the next generation of artificial intelligence is built on a foundation of integrity rather than just efficiency.
Vassilis Digalakis Jr is an Assistant Professor of Operations & Technology Management (OTM) at the Questrom School of Business at Boston University (BU). He is also affiliated with the Systems Engineering Division (SE) at the College of Engineering, the Center for Information and Systems Engineering (CISE), and the Hariri Institute for Computing, Computational Science and Engineering. Vassilis works on trustworthy AI: how to design AI systems that behave reliably, transparently, and appropriately for their context of use. He combines machine learning, optimization, and operations research to build such systems and to study how their deployment and governance shape adoption, downstream decisions, and real-world outcomes, particularly in healthcare and sustainability. His research has been published in journals such as Operations Research, Manufacturing & Service Operations Management, and IEEE Transactions on Knowledge and Data Engineering, and has received awards, including the INFORMS Pierskalla Award and the Harold Kuhn Award.