By Gabbie Burton
Contributing Writer

No matter where you look these days it seems as though artificial intelligence is there to meet your gaze. From the mundane to the complex and from achievement to threat, AI has become an inescapable part of modern life. The only question left for humanity to answer is how should we use it? 

The University of Delaware has an answer: For the greater good.

In early July of this year, the University of Delaware opened the First State AI Institute, an initiative to build a state-of-the-art AI infrastructure for the university to use in four key areas: research excellence, enhanced teaching and learning, optimized business operations and ethical AI leadership. 

The Institute is located in the FinTech Innovation Hub on the Science, Technology and Advanced Research (STAR) Campus.

“[AI] is proliferating into literally every discipline,” said Prof. Sunita Chandrasekaran, the Institute’s director and associate professor in the Department of Computer and Information Sciences. “We have had a lot of demand for our services, and are still having demand from the faculty of almost every discipline.”

Chandrasekaran explained that in her ten years at the university, the demand for AI has increased rapidly over the last two to three years, showing an increased need for the university to invest in AI frameworks. Fortunately for the institute, Chandrasekaran is well equipped to meet the increased demand. In addition to her roles as director and assistant professor, she also leads a large grant by the National Science Foundation on Democratizing Access to Research Software Engineering (DARSE) and was co-director of the AI Center of Excellence.

“Prior to leading this institute, we had an AI Center of Excellence,” said Chandrasekaran. “As part of the Center of Excellence, we ran hackathons and worked with faculty, but suddenly in 2023-2024 timeframe, we saw this boost of need for AI, and we couldn’t keep up with the demand, which then led to the start of the Institute.”

The increased need for AI capabilities raised questions including whether or not sensitive data sets from researchers at the university should be entered into third party AI vendors.

“These questions led us to wonder, ‘What if we pulled our own test bed infrastructure at UD? What if we built our own system so we can retain some of the sensitive data set within UD’s infrastructure, and still perform science and open research?” she said. “What if we took a copy of an existing tool but customized it towards faculty needs and their domain science, and found some answers? In the process, we will be doing new research.”

By having their own AI system, the university can safeguard sensitive data sets and personalize the system to best address UD’s unique faculty, student and operational needs including enrollment, finance, development and any other field as it arises. Further down the line,

Chandrasekaran shared that she can envision the university working with outside partners, but the primary focus remains on the university’s work and needs.

With this bold approach comes hard work, building the infrastructure, hardware and workforce to reach the goals of the Institute. Chandrasekaran explained that the Institute is following a timeline and so far they are working on finding leads for their key areas, organizing working groups, gathering data, recruiting research software engineers and more. A main initiative the Institute took on over the summer was the hackathon, which Chandrasekaran said took up the entire month of August.

On top of all the logistical work that is going into building the Institute, there is still a looming concern Chandrasekaran and the whole institute team must factor into the equation: the ethical complexities of AI and how best to mitigate them.

According to the institute’s website, “the Institute’s research will advance an open, human-centered vision of AI — one rooted in transparency, accountability and the public good.” The work to ensure such a model of AI requires diligence from the very beginning.

“It’s been a constant concern,” said Chandrasekaran. “We are trying to make sure that the input data set is not biased. I think that’s where the main issue comes from. Once your data set is biased or ignoring a certain type of input, then your end result - your end model - is biased. Then you’re done, and you can no longer get a neutral model.”

While the Institute is not creating an AI model from scratch, they are retraining an already existing AI system to mitigate this preconceived bias.

“We get a copy of the already messed up model, but then we are retraining it with our own data set, where we can ensure that it is unbiased, and then we use the retrained model for inferences, for results,” she said. “There is some control, if not all of it, because you know what you train the model on.”

While Chandrasekaran and the Institute’s team are doing what they can to create a more ethical AI model, the same may not be true for all AI models. Chandrasekaran shared that the other piece to ethical AI is user knowledge and awareness.

“I think the user needs to know what the right answer is, and that’s the other problem, because half the time we don’t know,” she said. “It is a very tough problem to solve – to know how to create awareness within the community and not just blindly believe what the models give you. You need to have some sense of what the outcome should be, and that’s an awareness we are trying to create internally.”

While the complexities of AI may never cease, and the First State AI Institute is still in the early stages of facing those complexities, Chandrasekaran and the team at the Institute are prioritizing a human and UD focused model. While there is still a lot of work coming up for the Institute, Chandrasekaran repeated a mantra that seemed to reflect her outlook on not only the work of the institute but the larger implications of AI:

“It is an opportunity, not a challenge.”

To learn more about the First State AI Institute at the University of Delaware, visit https://first-state-ai.udel.edu.