Reimagining Dementia Care

University of Notre Dame business professors Ahmed Abbasi and Ken Kelley have different research specialties, but in 2021, they saw that combining their perspectives and disciplines could spur meaningful ideas to impact societal good. Together, they formed the Human-centered Analytics Lab (HAL), a collaboration that brings together expertise from across disciplines to ask: How can we develop and use emerging technologies in ways that will genuinely improve people's lives?

Kelley, the Edward F. Sorin Society Professor of IT, Analytics, and Operations, considers how to measure and model  human elements and psychological attributes that are notoriously difficult to measure and model. Abbasi, the Joe and Jane Giovanini Professor of IT, Analytics, and Operations, focuses on machine learning, artificial intelligence and cognitive science. Both are in the IT, Analytics, and Operations Department, which is an interdisciplinary group focused on the technical side of management. 

One of the HAL team’s latest projects explores how artificial intelligence and social robotics might support people living with dementia and the caregivers and families who accompany them. On September 5, they presented “Robots & Caregiving: A Human-Centered Analytics Approach,” as part of Notre Dame’s 2026 Shamrock Series when the Fighting Irish took on the Wisconsin Badgers at Lambeau Field in Green Bay, Wisconsin.

On stage, three social robots named Rosie, Marcus and Tony joined Abbasi, Kelley, Ph.D. students Sofía Calderón and Ray Alavo, and Kristina Radivojevic Topalovich, HAL postdoctoral research associate, to deliver the presentation.

Kelley, who also serves as senior associate dean for faculty and research, introduced a stark picture of the near future: A tremendous surge in older adults facing dementia diseases will meet a severe shortage of professional healthcare workers. In memory care facilities, one facilitator may be responsible for dozens of residents, providing daily, individual emotional engagement — an increasingly difficult task. The HAL researchers set out to discover whether artificial intelligence can close this care gap; not to replace the warmth of the human connection, but to support it.

The initiative is a collaboration between Milton Village, a specialized adult day center in South Bend. Working alongside leaders including Angel Baginske of Alzheimer’s & Dementia Services of Northern Indiana and Dr. JoAnn Burke, Chair of the Indiana Commission on Aging, the researchers focused on how social robots could facilitate reminiscence therapy, a clinical intervention where structured conversations about past memories help stabilize cognitive function, reduce agitation and bring joy to individuals living with dementia. 

During the presentation, Abbasi called on two robots, Tony and Rosie, to demonstrate the conversation analytics pipeline of listening, processing and responding using football-themed prompts. Tony acted as an AI Marcus Freeman, head coach of Notre Dame football, answering questions from the audience about the upcoming game against Wisconsin. In the two questions asked, Tony’s responses demonstrated how delays in his answers as Coach Freeman could lead to user disengagement.

Next, Abassi, who also serves as director of the Lucy Institute for Data and Society, directed Tony and Rosie to demonstrate voice quality using an iconic scene from the movie “Rudy.” First, Abbasi played a version with stilted, monotone voices. Then the robots recited the scene with re-engineered audio analytics and voice embeddings to better present the emotional pitch and personas.

“That second version scored significantly higher in similarity metrics to human speech standards,” said Abbasi. “Proper voice alignment accounts for a 13% increase in perceived intelligence. Users simply do not want to interact with flat, overly robotic voices.”

Before deploying social robots to interact with vulnerable individuals facing memory loss, the researchers tested the underlying technology under controlled conditions. Calderón described how she led an empirical trial in the lab involving more than 400 Notre Dame undergraduate students with behavioral interview coaching using the structured STAR method: situation, task, action and result. The coaching was facilitated either by human instructors or by social robots running various AI pipeline configurations. The results demonstrated that cloud-connected social robots achieved training outcomes and user satisfaction on par with human facilitators.

Translating these laboratory findings to dementia patients with sensitivity and care presented three challenges: adults with dementia are often less verbal, unable to complete standard survey evaluations and require ongoing, longitudinal relationships rather than one-off interactions. To gauge patient comfort without relying on written surveys, the HAL team introduced real-time computer vision analytics. Using camera inputs mounted directly on the robots, the system tracks 30 facial landmarks twice per second to detect subtle emotional responses, including confusion, sadness, surprise and disgust.

The most complex technical challenge lay in long-term memory retrieval. In dementia care, two distinct forms of false memory intersect: human confabulation (when a patient with cognitive decline unintentionally invents details to fill memory gaps) and AI hallucination (when a large language model confidently generates incorrect details to fill missing context).

To solve this, Radivojevic Topalovich helped engineer a novel memory structure called LLM Wiki, which achieved 100% file retrieval success and outperformed standard AI memory architectures, drastically reducing AI hallucinations.

“We spent about 10,000 lines of code and two years to get to the point where we were ready for ongoing randomized controlled trials,” said Abbasi, introducing a video clip representing the culmination of years of work to the audience.

In the video clip, Rosie engages a dementia patient in a slow dialogue about her youth. When Rosie asks about her early school days, the woman smiles as memories resurface: holding tightly to the handrail while navigating the high steps off the school bus, greeting her principal each morning and learning phonics from her gentle first-grade teacher, Miss Ayers. 

This interaction highlighted the primary objective of the project: relieving emotional and logistical pressure on human caregivers. Caregivers in residential facilities face high turnover, low pay and long shifts, making it difficult for new staff to know every resident's personal backstory. While a busy staff member or family member might find hearing the same story repeated dozens of times frustrating, an AI-driven robot possesses endless patience, pulling up historic photos, favorite music or personal context to encourage meaningful dialogue.

Next, the HAL team will begin a 12-week randomized controlled trial and test Marcus, a mobile robot capable of physical movement and interactive games like checkers. In response to a question from a Shamrock Series presentation attendee about how to ensure social technology respects non-Western values and languages, Abbasi described how his work with another HAL Ph.D. student, Kezia Oketch, and his broader work as director of the Lucy Family Institute is advancing culturally appropriate AI models in Latin America and East Africa.

Kelley concluded the presentation by reflecting on the philosophical imperative inspiring their research, referencing the question Notre Dame President Rev. Robert A. Dowd, C.S.C., chose as the theme for the 2026–27 academic year: What does it mean to live a good life? For HAL researchers, Kelley said, the answer lies in using technology thoughtfully to care for one another, especially when age or illness threatens human connection.

“The technology is remarkable, but what should stay with you is the human connection it facilitates. Technology alone won't solve the demographic challenge, but people using technology thoughtfully can make a meaningful difference to everyone providing care for a person with dementia,” he said.