When Mason Chilmonczyk, M.S. ME 2017, Ph.D. ME 2020, arrived at Georgia Tech to pursue graduate degrees in mechanical engineering, his goal was to become a professor. Instead, an unexpected turn in his research led him to entrepreneurship.

Today, he is the chief executive officer of Andson Biotech, a growing biotools startup he co-founded with Andrei Fedorov, associate chair for graduate studies and the Rae S. and Frank H. Neely Chair at the George W. Woodruff School of Mechanical Engineering. The company is commercializing a breakthrough technology Chilmonczyk developed during his doctoral research that simplifies the development and production of cell and gene therapies.

Read the full story on the George W. Woodruff School of Mechanical Engineering website.

News Contact

Nothing rivals the human brain’s complexity. Its 86 billion neurons and 85 billion other cells make an estimated 100 trillion connections. If the brain were a computer, it would perform an exaflop (a billion-billion) mathematical calculations every second and use the equivalent of only 20 watts of power. As impressive as the brain is, neurologists can’t fully explain how neurons work together.

To help find answers, researchers at the Institute for Neuroscience, Neurotechnology, and Society (INNS) are using math, data, and AI to unlock the secrets of thought. Together they are helping turn the brain’s raw electrical “noise” into real insights about how people think, move, and perceive the world.

Fair warning: Prepare your neurons for the complexity of this brain research ahead.

Building AI Like a Brain

What if artificial neurons in AI programs were arranged as they are in the brain?

AI programs would then help us understand why the brain is organized the way it is. This neuro-AI synthesis would also work faster, use less energy, and be easier to interpret. Creating such systems is the goal of Apurva Ratan Murty, an assistant professor of Psychology who is creating topographic AI models like the one above of three domains — vision, audition, and language inspired by the brain. In the near future, he predicts doctors might be able to use these patterns to predict the effects of brain lesions and other disorders. “We’re not there yet,” he says. “But our work brings us significantly closer to that future than ever before.”

Computing Thought and Movement

How cats walk keeps Chethan Pandarinath on his toes. This biomedical engineer uses sensors to analyze how two sets of feline leg muscles — flexors and extensors — are controlled by the spinal cord. Understanding how that happens could help patients partially paralyzed from spinal cord injuries, strokes, or progressive neuro-degenerative diseases get back on their feet again. “My lab is using AI tools that allow us to turn complex spinal cord activity data into something we can interpret. It tells us there’s a simple underlying structure behind the complex activity patterns,” says the associate professor.

Revealing the Brain’s Spike Patterns

“The brain is like a symphony conductor,” says Simon Sponberg. “Individual instruments have some independent control, but most of the music comes from the brain’s precise coordination of notes among the different players in the body.” This physics professor studies the fantastically fast-beating wings of the hummingbird-sized hawk moth (Manduca sexta). Its agile flight movement comes as a result of spikes in electrical activity in 10 muscles. Sponberg found something that surprised him — the brain focuses less on creating the number of spikes than in orchestrating their precise patterns over time. To Sponberg, every millisecond matters. “We are just beginning to understand how the nervous system first acquires precisely timed spiking patterns during development,” he says.

Predicting Decisions Through Statistics

Put a mouse in a maze with food far away, and it will learn to find it. But life for mice — and people — isn’t so simple. Sometimes they want to explore, only want water, or just want to go home. What’s more, animals make decisions based on their history, not just on how they feel at the moment. To dig deeper into the decision-making process, Anqi Wu, an assistant professor in the School of Computational Science and Engineering, is giving mice more options. By using a new computational framework called SWIRL (Switching Inverse Reinforcement Learning), her findings have outperformed models that fail to take historical behavior into account. “We’re seeking to understand not only animal behavior but also human behavior to gain insight into the human decision-making process over a long period of time,” she says.

Modeling the Mind’s Wiring With Math

Connectivity shapes cognition in the cerebral cortex, a layered structure in the brain. The visual cortex, in particular, processes visual data from the retina relayed through the Lateral Geniculate Nucleus (LGN) in the thalamus, and directs it to the correct cognitive domain in the brain. How it does this is the mystery that computational neuroscientist Hannah Choi wants to solve. “The big question I’m interested in is how network connectivity patterns in the architecture of the LGN are related to computations,” says this assistant math professor. To find answers, she shows mice repeated image patterns such as flower-cat-dog-house and then disrupts the pattern. The goal? To grasp how the thalamus’s nonlinear dynamical system works. If scientists and doctors better understand how brain regions are wired together, such knowledge could lead to better disease treatment.

This story was originally published through the Georgia Tech Alumni Magazine. Read the original publication here.

News Contact

One day after the historic Artemis II launch, the College of Sciences welcomed more than 150 researchers, students, and community members to its signature Frontiers in Science conference. Held on April 2, the full-day event focused on space research guiding discovery and innovation.

As during previous editions, this year’s conference featured more than two dozen scientists, engineers, policy experts, and thought leaders from Georgia Tech and beyond, illustrating how collaboration across fields – from science and engineering to public policy and international affairs – helps to advance strategic research priorities. 

“Frontiers is about discovery and connections across disciplines and generations,” says Susan Lozier, dean of the College of Sciences and Betsy Middleton and John Clark Sutherland Chair. “This edition provided an inspiring glimpse into the future of space exploration and the many ways Georgia Tech is contributing to research and missions seeking answers to what lies beyond our planet.” 

Commitment to Space

Space research is a key institutional priority at Georgia Tech, which is home to numerous academic and research programs in planetary sciences, robotics, mission design, space policy, and other areas. 

The recently established Space Research Institute (SRI) serves as the central hub connecting the broad range of space-related research across campus. Led by Jud Ready, who also serves as principal research engineer at the Georgia Tech Research Institute, SRI has expanded support for space research and commercialization through initiatives such as the CreationsVC Space Fellows Program and Centers, Programs, and Initiatives seed grant program.

SRI’s efforts are in line with Georgia Tech’s long-standing contribution to space exploration. Hundreds of Yellow Jacket alumni work in the space sector, including several graduates who are playing key roles in the Artemis program. To date, more than a dozen Georgia Tech alumni have traveled to space.

Exploring the Final Frontier

The conference featured a series of panels and discussions led by faculty and researchers from the Colleges of Sciences and Engineering as well as the Ivan Allen College of Liberal Arts. 

Sessions explored how researchers are studying the processes and conditions that support planetary habitability, seeking to answer one of humanity’s greatest questions: Does life exist beyond Earth? Speakers also examined how analog fieldwork in Earth’s extreme environments can inform space exploration, and how space research, in turn, can deepen our understanding of our own world.

Additional conversations centered on building better space missions through improved understanding of team and individual resilience, data collection, navigation, and the development of advanced technologies like the robots developed through the NASA LASSIE Project. 

Frontiers also highlighted Georgia Tech’s commitment to preparing the next generation of space scientists, engineers, and leaders. Student training and engagement were recurring themes throughout the day, with speakers emphasizing opportunities for student-led and student-run missions and research. A panel of Georgia Tech alumni shared their own STEM career journeys, challenging the idea of “one right path” to success — and acknowledging the resources and opportunities available at the Institute. 

A highlight of the conference was a fireside chat with Atlanta-native, retired U.S. Army Colonel and NASA Astronaut R. Shane Kimbrough (M.S. Operations Research 1998). Kimbrough, who spent a total of 388 days in space and performed nine spacewalks across three missions, reflected on his career and the evolution of spaceflight. He emphasized the expanding role of public-private and international partnerships in advancing ambitious goals, such as creating a permanent human outpost on the Moon. 

Policy and Public

The conference also explored how policy influences space discovery and innovation, with discussions touching on such issues as space security, access, governance, sustainability — and the influence of technology and science fiction on public perception and policy. 

Panelists described current policy frameworks governing outer space as struggling to keep pace with rapidly advancing technologies and expanding activities. According to these experts, increasing tensions among commercial, research, and recreational uses of space call for greater coordination among private and government entities to balance competing priorities while maximizing opportunities for innovation and exploration. 

The conference was punctuated by a networking lunch connecting attendees with Atlanta’s public astronomy community – including partners at several universities and the Georgia Tech Astronomy Club, which set up telescopes for attendees to safely observe the sun. Later that evening, the Georgia Tech Observatory hosted its Public Night, welcoming the broader Atlanta community to campus for telescope views of Jupiter, the Orion Nebula, and other celestial bodies. 

The Observatory Night was a fitting conclusion to a full day focused on Georgia Tech’s commitment and contributions to inspiring future generations of space explorers through research, education, and outreach. 

Experience the Frontiers conference in pictures on the College of Sciences’ Flickr account.

News Contact

As students increasingly turn to artificial intelligence (AI) to help with coursework, some worry that their learning could be compromised. Georgia Tech researchers are working to counter this potential decline with an AI tool they hope will promote learning rather than hinder it.  

TokenSmith is a citation-supported large language model (LLM) tutor that can be hosted locally on a user’s personal computer. The tutor only provides answers based on course materials, such as the textbook or lecture slides.  

Associate Professor Joy Arulraj began the project with support from the Bill Kent Family Foundation AI in Higher Education Faculty Fellowship last year. The fellowship, led by Georgia Tech’s Center for 21st Century Universities, supports faculty projects exploring innovative and ethical uses of AI in teaching.   

Arulraj has enlisted assistant professors Kexin Rong and Steve Mussmann to help build TokenSmith.  

Mussmann said TokenSmith is a synergistic blend of a database system and a machine learning system. The model stores textbooks, textbook annotations by course staff, common questions and answers, a learning state of the student, and student feedback in a structured database system. However, machine learning plays a key role in the answer generation as well as adapting the system to the student, course staff guidance, and user feedback.

"What excites me most is demonstrating how data-driven ML and principled database systems design can reinforce each other — one providing adaptability and flexibility, the other providing structure and traceability — in a way that benefits students," Mussmann said.

Keeping the model local has been an important focus of the project. The team wanted to create an AI tutor that helps students learn from their class resources rather than just giving answers. With each response, TokenSmith cites the origin of the answer in the provided documents.  

“One problem with LLMs is that they can hallucinate and provide wrong answers, but in this controlled environment, we can add these guardrails to make sure it’s actually helpful in an educational setting,” Rong said.  

Rong said she feels that students often undervalue textbooks, and she hopes TokenSmith can motivate students to make better use of them.  

“Textbooks can sometimes be daunting, but maybe if we combine them with the model, students might be more willing to read a paragraph or page in the textbook, and that could help clarify something for them,” she said.  

Running the model locally is more cost-effective and helps preserve the user’s privacy. But running the new tool locally comes with technical challenges.  

One challenge with creating the model is speed. Since it is a locally based model, TokenSmith depends solely on the user’s computer memory.  Tests have also shown that the tutor currently struggles to answer more complex questions. 

“We are interested in pushing the boundaries of these local models so that they give students good answers and also run fast enough to keep students engaged,” Arulraj said.  

News Contact

In recent years, the Centers for Disease Control and Prevention, the Department of Homeland Security, and other authorities have flagged a record number of unauthorized shipments of biological materials. At the same time, global intelligence communities have identified numerous attempts to smuggle sensitive biological samples in efforts of industrial theft or espionage. 

“A small vial of genetically engineered cells can contain multiple millions of dollars’ worth of intellectual property and require several years of work to develop,” said Corey Wilson, a professor in Georgia Tech’s School of Chemical and Biomolecular Engineering (ChBE). “Accordingly, the protection of high-value engineered cell lines has become critically important to the biotechnology industry.”

Wilson and his research team have published their findings in Science Advances demonstrating the effectiveness of their new biological security technology, known as GeneLock™, in protecting high-value engineered cell lines.

GeneLock is a cybersecurity-inspired technology that protects valuable genetic material directly at the DNA level. To demonstrate its strength, Wilson’s team conducted what they describe as a first-of-its-kind biohackathon, detailed in the new paper, to simulate unauthorized access. 

“GeneLock greatly improves our ability to protect high-value engineered cell lines by expanding security from the lab environment to the genetic level,” Wilson said.

Economic Impact

What are the stakes? Estimates place the global market for high-value genetic materials at more than $1.5 trillion, projected to reach $8 trillion by 2035. The use of these materials ranges from advanced medicines and proprietary research enzymes to specialty chemicals and sustainable materials.

Currently, the protection of high-value cell lines depends on physical safeguards such as restricted lab access and secure facilities, Wilson explained.

“The key weakness of physical security measures is once circumvented, there are typically no measures in place to protect valuable cells from theft, abuse, or unauthorized use,” Wilson said. 

“Once a sample leaves the building, the DNA it carries typically remains fully functional. This is like placing an unlocked cellphone in a desk drawer. Anyone who gains access to the drawer can view sensitive content on the phone­­­­­­­—or in this case will have full access to the valuable cell line.”

Genetic Passcode Protection

The GeneLock biological security technology developed by Wilson and his team places a passcode on engineered cells, akin to those used on ATM machines and protected cellphones.

Instead of leaving a valuable gene in readable form, the team scrambles the DNA sequence of interest. The scrambled genetic asset remains in a nonfunctional state unless the living cell where it resides receives the correct sequence of chemical inputs. Those inputs act as a molecular passcode.

“Only the right combination, delivered in the right order, rearranges the DNA into a working form,” Wilson said.

Biohackathon Security Test

To evaluate the technology, the researchers organized a blue team and a red team in what they describe as an ethical biohackathon. The blue team designed the encrypted DNA sequence, while the red team was challenged to discover the correct chemical passcode through experimentation in a gray box exercise, meaning the red team had partial knowledge of the system but did not have access to the internal designs. 

“This approach for testing security strength is commonly used in cybersecurity,” Wilson explained. 

The blue team engineered the system inside Escherichia coli, or E. coli, a bacterium widely used in biotechnology. The protected asset was a fluorescent protein gene selected as a measurable stand-in for commercially valuable targets. When the correct chemical sequence was applied, the fluorescence turned on. Without the correct passcode, the gene remained scrambled and the cells could not fluoresce green. 

“In practice, most DNA sequences produce valuable proteins or chemicals that are essentially invisible to the human eye, requiring specialized devices or experiments to observe,” Wilson said. “If the biohackathon were conducted with a standard commercially valuable target, the penetration testing would have taken more than 10 times longer to complete, years instead of months.”

The biohackathon results showed a dramatic reduction in risk. GeneLock reduced the probability of unlocking the genetic asset by random search to about 1 in 85,000 (a 0.001% chance), assuming the unauthorized user had access to the required chemical inputs.

Without access to those inputs, “the likelihood of success by chance becomes effectively negligible,” said Dowan Kim (Georgia Tech PhD 2024), co-lead author of the study.

Commercial Uses and What’s Next 

Although the researchers used a non-commercial fluorescent protein as a test case, the implications extend much further. Many biotechnology companies rely on proprietary engineered strains. New England Biolabs, for example, produces more than 265 non-disclosed enzymes in E. coli, each representing a high-value cell line. 

Protein-based drugs are also manufactured in living cells, and proprietary metabolic pathways are used to produce specialty chemicals, bioplastics, and high-value ingredients. 

“In each case, the genetic blueprint inside the cell represents intellectual property that can be protected by our technology,” said Ishita Kumar, a PhD candidate in ChBE and co-lead author of the study.

While the team’s current focus is on protecting intellectual property in the form of high-value cells, future iterations aim to strengthen biological security more broadly. 

“We are currently developing protection measures to mitigate unauthorized use or release of sensitive cell lines that can be potentially hazardous to human health or the environment,” Wilson said.

“As it stands, GeneLock represents an important shift in biological security, enabling, for the first time, protection of valuable cells at the genetic level, even after physical security measures have been bypassed,” he added. 

The work is already moving toward commercialization. The team filed a provisional patent application with the U.S. Patent and Trademark Office in February 2026 and is forming a company to deploy the technology.

This research was funded by a grant from the National Science Foundation.

CITATION:

Dowan Kim, Ishita Kumar, Mohamed Hassan, Luisa F. Barraza-Vergara, Christopher A. Voigt, and Corey J. Wilson, “Protecting cells at the genetic level and simulating unauthorized access via a biohackathon,” Science Advances, 2026.

News Contact

Manufacturing is undergoing a significant transformation as artificial intelligence reshapes how industrial systems operate, adapt, and scale. The H. Milton Stewart School of Industrial and Systems Engineering (ISyE) has launched its Manufacturing and AI Initiative, which brings together faculty expertise in statistics, optimization, data science, and systems engineering to address emerging challenges and opportunities in modern manufacturing.

ISyE researchers are applying AI to complex manufacturing environments, including multistage production systems, asset management, quality improvement, and human‑centered manufacturing. Faculty leaders emphasize the importance of contextualizing large volumes of manufacturing data so AI can support reliable decision‑making, efficient operations, and sustainable outcomes. At the same time, the initiative acknowledges challenges such as data integration, system complexity, and the need to balance automation with human involvement. Together, these efforts position ISyE at the forefront of shaping AI‑powered manufacturing systems that are innovative, resilient, and socially responsible.

Read the full article in ISyE Magazine 

News Contact

A new study by EPIcenter affiliate Jamal Mamkhezri examines how public preferences for solar‑energy policy have shifted over a six‑year period in New Mexico, offering one of the first long‑term repeated cross‑section analyses of willingness to pay (WTP) for renewable‑energy attributes. Using identical discrete choice experiment (DCE) tasks from surveys conducted in 2017 and 2023, Professor Mamkhezri evaluates how households value increases in Renewable Portfolio Standards (RPS), changes in rooftop versus utility‑scale solar shares, monthly credit‑banking rules, water usage in electricity generation, and smart‑meter information delivery options.

Across more than 1,100 combined respondents, the study uncovers selective temporal stability in energy preferences. Some attributes—such as support for higher RPS targets, reductions in water use, and preferences for online smart‑meter information—remain relatively stable over time. In contrast, others shift considerably: WTP for increasing the rooftop solar share declines by more than 40%, while WTP to protect monthly credit banking rises more than 200%, reflecting heightened awareness of net‑metering debates and rapid growth in rooftop solar adoption.

Importantly, the study reveals that environmental attitudes, measured through New Ecological Paradigm (NEP) scores, once strongly predicted preferences for rooftop solar and smart‑meter technologies in 2017, but these relationships fade or even reverse by 2023—signaling a shift as these technologies transition from niche, identity‑driven goods to mainstream infrastructure. Meanwhile, environmental attitudes continue to robustly shape preferences for RPS increases and water‑use reductions in both survey waves.

Read Full Story on the EPIcenter Webpage

News Contact

A recent review by EPIcenter faculty affiliate Constance Crozier (School of Industrial and Systems Engineering, Georgia Institute of Technology) and Matthew Liska (School of Physics, Georgia Institute of Technology) explores the growing role of data centers in providing flexibility, the ability to shift or reduce electricity use in response to grid conditions, to the electric grid as renewable energy penetration and AI-driven computing demand surge. The authors highlight that data centers, particularly those supporting high-performance computing and AI workloads, are projected to consume nearly 10% of U.S. electricity by the end of the decade, presenting both challenges and opportunities for grid stability.

The paper examines various strategies for enhancing the flexibility of data center energy use. One approach is to use backup power systems, such as uninterruptible power supplies, to support the grid during emergencies. Another method involves rerouting computing jobs to different data centers in other locations to balance energy demand. The authors also discuss implementing smart scheduling techniques that shift workloads to off-peak hours, reducing strain on the grid. Additionally, they highlight adjusting processor speeds by lowering CPU (central processing unit) and GPU (graphics processing unit) clock rates to limit power consumption when needed. Finally, the paper suggests pre-cooling data center equipment to limit the energy required for cooling during peak demand periods. Notably, experimental evidence shows that underclocking GPUs can cut power consumption by 40% with only a 22% performance loss, suggesting technical feasibility for demand-response interventions.

Despite these technical options, the authors find that real-world cost considerations and reliability concerns limit widespread adoption. Data center operators generally do not change their behavior in response to electricity prices, as job revenue far outweighs energy costs under normal conditions. For example, a GPU rented at $2 per hour consumes only $0.04 worth of electricity at average prices, making curtailment unattractive except during extreme price spikes. Surveys indicate that operators are reluctant to compromise reliability or deploy backup systems for ancillary services. Consequently, price-based incentives alone are unlikely to drive meaningful flexibility.

Read more on the EPIcenter Webpage
Listen to a podcast on the research here

News Contact

While people use search engines, chatbots, and generative artificial intelligence tools every day, most don’t know how they work. This sets unrealistic expectations for AI and leads to misuse. It also slows progress toward building new AI applications. 

Georgia Tech researchers are making AI easier to understand through their work on Transformer Explainer. The free, online tool shows non-experts how ChatGPT, Claude, and other large language models (LLMs) process language. 

Transformer Explainer is easy to use and runs on any web browser. It quickly went viral after its debut, reaching 150,000 users in its first three months. More than 563,000 people worldwide have used the tool so far.

Global interest in Transformer Explainer continues when the team presents the tool at the 2026 Conference on Human Factors in Computing Systems (CHI 2026). CHI, the world’s most prestigious conference on human-computer interaction, will take place in Barcelona, April 13-17.

[Related: GT @ CHI 2026]

“There are moments when LLMs can seem almost like a person with their own will and personality, and that misperception has real consequences. For example, there have been cases where teenagers have made poor decisions based on conversations with LLMs,” said Ph.D. student Aeree Cho.

“Understanding that an LLM is fundamentally a model that predicts the probability distribution of the next token helps users avoid taking its outputs as absolute. What you put in shapes what comes out, and that understanding helps people engage with AI more carefully and critically.”

A transformer is a neural network architecture that changes data input sequence into an output. Text, audio, and images are forms of processed data, which is why transformers are common in generative AI models. They do this by learning context and tracking mathematical relationships between sequence components.

Transformer Explainer demystifies how transformers work. The platform uses visualization and interaction to show, step by step, how text flows through a model and produces predictions.

Using this approach, Transformer Explainer impacts the AI landscape in four main ways:

• It counters hype and misconceptions surrounding AI by showing how transformers work.
• It improves AI literacy among users by removing technical barriers and lowering the entry for learning about AI.
• It expands AI education by helping instructors teach AI mechanisms without extensive setup or computing resources.
• It influences future development of AI tools and educational techniques by providing a blueprint for interpretable AI systems.

“When I first learned about transformers, I felt overwhelmed. A transformer model has many parts, each with its own complex math. Existing resources typically present all this information at once, making it difficult to see how everything fits together,” said Grace Kim, a dual B.S./M.S. computer science student. 

“By leveraging interactive visualization, we use levels of abstraction to first show the big picture of the entire model. Then users click into individual parts to reveal the underlying details and math. This way, Transformer Explainer makes learning far less intimidating.”

Many users don’t know what transformers are or how they work. The Georgia Tech team found that people often misunderstand AI. Some label AI with human-like characteristics, such as creativity. Others even describe it as working like magic.

Furthermore, barriers make it hard for students interested in transformers to start learning. Tutorials tend to be too technical and overwhelm beginners with math and code. While visualization tools exist, these often target more advanced AI experts.

Transformer Explainer overcomes these obstacles through its interactive, user-focused platform. It runs a familiar GPT model directly in any web browser, requiring no installation or special hardware. 

Users can enter their own text and watch the model predict the next word in real time. Sankey-style diagrams show how information moves through embeddings, attention heads, and transformer blocks.

The platform also lets users switch between high-level concepts and detailed math. By adjusting temperature settings, users can see how randomness affects predictions. This reveals how probabilities drive AI outputs, rather than creativity.

“Millions of people around the world interact with transformer-driven AI. We believe that it is crucial to bridge the gap between day-to-day user experience and the models' technical reality, ensuring these tools are not misinterpreted as human-like or seen as sentient,” said Ph.D. student Alex Karpekov. 

“Explaining the architecture helps users recognize that language generated by models is a product of computation, leading to a more grounded engagement with the technology.” 

Cho, Karpekov, and Kim led the development of Transformer Explainer. Ph.D. students Alec Helbling, Seongmin Lee, Ben Hoover, and alumni Zijie (Jay) Wang (Ph.D. ML-CSE 2024) and Minsuk Kahng (Ph.D. CS-CSE 2019) assisted on the project. 

Professor Polo Chau supervised the group and their work. His lab focuses on data science, human-centered AI, and visualization for social good.

Acceptance at CHI 2026 stems from the team winning the best poster award at the 2024 IEEE Visualization Conference. This recognition from one of the top venues in visualization research highlights Transformer Explainer’s effectiveness in teaching how transformers work.

“Transformer Explainer has reached over half a million learners worldwide,” said Chau, a faculty member in the School of Computational Science and Engineering. 

“I'm thrilled to see it extend Georgia Tech's mission of expanding access to higher education, now to anyone with a web browser.”

News Contact

Voice-activated, conversational artificial intelligence (AI) agents must provide clear explanations for their suggestions, or older adults aren’t likely to trust them.

That’s one of the main findings from a study by AI Caring on what older adults expect from explainable AI (XAI).

AI Caring is one of three AI Institutions led by Georgia Tech and funded by the National Science Foundation (NSF). The institution supports AI research that benefits older adults and their caregivers.

Niharika Mathur, a Ph.D. candidate in the School of Interactive Computing, was the lead author of a paper based on the study. The paper will be presented in April at the 2026 ACM Conference on Human Factors in Computing Systems (CHI) in Barcelona.

Mathur worked with the Cognitive Empowerment Program at Emory University to interview 23 older adults who live alone and use voice-activated AI assistants like Amazon’s Alexa and Google Home.

Many of them told her they feel excluded from the design of these products.

“The assumption is that all people want interactions the same way and across all kinds of situations, but that isn’t true,” Mathur said. “How older people use AI and what they want from it are different from what younger people prefer.”

One example she gave is that young people tend to be informal when talking with AI. Older people, on the other hand, talk to the agent like they would a person.

“If Older adults are talking to their family members about Alexa, they usually refer to Alexa as ‘she’ instead of ‘it,’” Mathur said. “They tend to humanize these systems a lot more than young people.”

Good Explanations

The study evaluated AI explanations that drew information from four sources of data:

• User history (past conversations with the agent)
• Environmental data (indoor temperature or the weather forecast)
• Activity data (how much time a user spends in different areas of the home)
• Internal reasoning (mathematical probabilities and likely outcomes)

Mathur said older users trust the agent more when it bases its explanations on data from the first three sources. However, internal reasoning creates skepticism.

Internal reasoning means the AI doesn’t have enough data from the other sources to give an explanation. It provides a percentage to reflect its confidence based on what it knows.

“The overwhelming response was negative toward confidence scores,” Mathur said. “If the AI says it’s 92% confident, older adults want to know what that’s based on.”

This is another example that Mathur said points to generational preferences.

“There’s a lot of explainable AI research that shows younger people like to see numbers in explanations, and they also tend to rely too much on explanations that contain numerical confidence. Older adults are the opposite. It makes them trust it less.”

Knowing the Context

Mathur said that AI agents interacting with older adults should serve a dual purpose. They should provide users with companionship and support independence while reducing the caretaking burden often placed on family members. 

Some studies have shown that engineers have tended to favor caretakers in the design of these tools. They prioritize daily tasks and routines, leaving some older adults to feel like they are merely a box to be checked.

She discovered that in urgent situations, older users prefer the AI to be straightforward, while in casual settings, they desire more conversation.

“How people interact with technological systems is grounded in what the stakes of the situation are,” she said. “If it had anything to do with their immediate sense of safety, they did not want conversational elaboration. They want the AI to be very direct and factual.”

Not Just Checking Boxes

Mathur said AI agents that interact with older adults are ideally constructed with a dual purpose. They should provide companionship and autonomy for the users while alleviating the burden of caretaking that is often placed on their family members. 

Some studies have shown that engineers have strayed toward favoring caretakers in the design of these tools. They prioritize daily tasks and routines, leaving some older adults to feel like they are a box to be checked.

“They’re not being thought of as consumers,” Mathur said. “A lot of products are being made for them but not with them.”

She also said psychological well-being is one of the most important outcomes these tools should produce. 

Showing older adults that they are listened to can significantly help in gaining their trust. Some interviewees told Mathur they want agents who are deliberate about understanding their preferences and don’t dismiss their questions.

Meeting these needs reduces the likelihood of protesting and creating conflict with family members.

“It highlights just how important well-designed explanations are,” she said. “We must go beyond a transparency checklist.”

News Contact