As more people turn to large-language models (LLMs) during personal crises, the companies behind them are engineering these AI-powered models to be more personal, empathetic, and understanding when prompts indicate a user is in emotional distress.

The trend is raising concern about how well prepared ChatGPT, Claude, Gemini, and other popular LLMs are to provide appropriate emotional support.

To address this concern, Georgia Tech researchers have developed a new neural framework that examines how different training sources influence empathy and social reasoning within these AI-powered models.

“A lot of people are using LLMs to help process the death of a loved one, or maybe their girlfriend broke up with them, or their boyfriend cheated on them,” said Glenn Matlin, a Ph.D. student in the School of Interactive Computing.

Matlin said the new framework he and a Georgia Tech-led research team have developed audits neural networks with an LLM and estimates the influence documents, transcripts, and other data sources have on a model’s output.

“We’re trying to connect the data to outputs, which is helpful for us in understanding where the information a language model produces comes from,” he said.

Matlin is the lead author of a new paper examining model sourcing for social reasoning and how models mimic human behavior in conversations. The paper will be presented this week at the Conference on Language Modeling in San Francisco.

Matlin said the findings suggest that an objective tone in its output indicates that a model places greater importance on textbooks or similar neutral training sources. An empathetic tone suggests the model values sources with dialogue and conversational examples more.

“We find that models tend to learn better from interpersonal dialogue when it comes to social reasoning,” he said. “Models that are good at social reasoning will rely less on textbooks about human emotions and more on FAQs or customer service dialogue.

“It can read all it wants from texts that explain anger, but that wouldn’t help it determine whether someone is angry.”

Professor Mark Riedl, Matlin’s advisor, said that knowing which sources models value most could enable engineers to alter their behavior.

“The question is, in this vast ocean of documents in this neural network, which ones become important in certain situations, and which ones don’t,” Riedl said. 

“This leads to the potential for interventions in how we train the model. If we want more empathy, we might look to see where it learns empathy and give it more of those documents, fewer of them, or other documents.”

Riedl said the framework can audit an LLM’s sourcing for advice on any topic, from financial to medical.

“We want to help people make more informed decisions about whether they should use AI for a particular task,” he said.

“You might make a different decision about whether to trust the AI’s recommendation if you knew how a skill was learned and which documents influenced that decision. The first step is knowing how it learns, where it learns from, and what documents tend to lead to certain behaviors.”

The findings that Matlin, Riedl, and their collaborators have made so far came through a partnership with the Allen AI Institute (Ai2), which allowed them to audit their open-sourced LLM, Olmo 3.

While not as large as ChatGPT or Claude, Olmo 3’s training data comprises over a billion documents.

Matilin said it’s impossible to know why ChatGPT or Claude behave in certain ways because their parent companies do not disclose their training data. However, Matlin’s study may help researchers to make better guesses about what’s happening behind the scenes.

He added that the study is the first of its kind and scale on an academic level.

“This is the first one that’s ever been done for an open data model of that type,” he said. “We’re providing a view into neural networks that is hard to come by without being in the corporate setting.”

Amazon and the Georgia Institute of Technology announced today the launch of the Georgia Tech–Amazon Science Hub, reinforcing a shared commitment to advancing cutting-edge research through academic–industry collaboration. The Hub builds on extensive existing collaborations between Georgia Tech and Amazon over the past decade. As part of the collaboration, Amazon will provide funding for joint research projects and community-building initiatives designed to accelerate scientific ambitions. 

"The most meaningful breakthroughs happen when intelligent hardware and thoughtful software come together to genuinely help people,” said Tapas Roy, vice president of Device Software and Services at Amazon. “Georgia Tech's world-class expertise in robotics and engineering, paired with Amazon's work in on-device intelligence and artificial intelligence, will accelerate our understanding of systems that can both respond to prompts and anticipate customer needs. I'm excited to see how this collaboration advances the science of how humans and intelligent machines work together.”
 
The Hub’s initial focus areas will explore challenges in autonomous systems, next-generation logistics infrastructure, and low-Earth-orbit communications, drawing on Georgia Tech’s well-known interdisciplinary research approach.
 
Georgia Tech is also joining the Fauna Robotics' ecosystem, using the Sprout research platform to study how humans and robots can live, work, and play side by side. This collaboration opens new ways to study mobility, perception, manipulation, and human-robot interaction. Leadership and alumni recently celebrated the launch at Lab 126, with guest appearances from Sprout and Buzz, the Georgia Tech mascot.

The Hub will support:
•    Joint research projects between Georgia Tech faculty and Amazon scientists across scientific disciplines 
•    Regular symposia and workshops to foster collaboration between researchers from both institutions and the broader Atlanta tech community
•    Community outreach programs designed to broaden participation from interdisciplinary scholars and innovators

“Georgia Tech and Amazon have worked together for years across research, technology, and student development," said Tim Lieuwen, Georgia Tech executive vice president for Research. "This new hub puts our partnership to work for Georgia. It connects research, education, and industry in ways that will strengthen our state’s innovation economy and create new opportunities for our students.”

About Georgia Tech
Georgia Institute of Technology is one of the nation’s top public research universities, preparing leaders who advance technology and improve the human condition. The Institute serves more than 56,000 students from all 50 states and over 146 countries and drives innovation and economic growth through research that benefits Georgia, the nation, and the world.

About Amazon in Georgia
Since 2010, Amazon has invested more than $30 billion in Georgia, supporting more than 30,000 full- and part-time jobs. Amazon’s research and technology teams in the Atlanta area work on challenges spanning supply chain optimization, logistics technology, and cloud computing. The establishment of the Georgia Tech–Amazon Science Hub further deepens Amazon’s engagement with one of the nation’s leading research institutions.

About Fauna Robotics 
Fauna Robotics builds Sprout, a development platform designed to put advanced robotics capabilities into the hands of researchers and developers. Sprout gives builders hands-on access to study mobility, perception, motor control, and manipulation with the ambition to deploy robots into human-centric spaces.
 

News Contact

News Contact

The Institute for Data Engineering and Science (IDEaS) at Georgia Tech in collaboration with Amazon Web Services has launched its upskilling initiative with the inauguration of the first cohort of the IDEaS & AWS Accelerator Program. This program is designed to provide the campus research community with hands-on experience with AWS tools and instruction from AWS experts in the application of AI assisted workflows to enable and support their teams. 

The fall cohort of 18 researchers will participate in a free, fully remote, 5-week program running from October the 12th through November the 13th. The fall 2026 curriculum focuses on biomedical and health themes. Topics in the course range from foundational themes such as understanding how to implement effective workflows with agents to technical topics such as AI/ML analysis of data sets, medical imaging and classification with AI/ML, and using AI to optimize infrastructure spending.   

The AWS Accelerator Program will be hosting a spring session, with the topic in discussion. To receive information about the spring workshop, and all of the resources for research from IDEaS, sign up for our weekly update here. 

News Contact

Artificial intelligence (AI) is changing not only how people work with technology but also how emerging technologies intersect with history, culture, and the systems that shape society.

These questions will be at the center of a conversation with Kamau Bobb, senior director of Georgia Tech’s Constellations Center for Education in Computing, on Oct. 16, as part of the CTRL+CM Speaker Series.

Bobb will discuss his new book, The Bullwhip's Algorithm: Artificial Intelligence and America's Racial Caste System, from 11 a.m. to 1 p.m. in the Scholars Event Theater at Price Gilbert Library. Lunch will be provided, and registration is required.

Drawing on the themes of his book, Bobb will examine the connections among AI, public policy, education, and social systems. The discussion will consider both the opportunities and challenges posed by emerging technologies and the responsibilities that accompany their development and use.

Bobb brings experience spanning higher education, government, and industry to the conversation.

He is the founding senior director of Georgia Tech’s Constellations Center for Education in Computing, which opened in 2017. Before joining Georgia Tech, he served as director of STEM Education Strategy and Research at Google, a program officer in the National Science Foundation’s Directorate for Computer and Information Science and Engineering, and director of the STEM Initiative for the University System of Georgia. He also served on President Obama’s My Brother’s Keeper STEM Task Force.

He earned a Ph.D. in science and technology policy from Georgia Tech and master’s and bachelor’s degrees in mechanical engineering from the University of California, Berkeley.

The speaking engagement follows Bobb’s September appearance at the Richard Tapia Celebration of Diversity in Computing in Atlanta. He co-headlined a session with data scientist and Tapia co-founder Valerie Thomas on AI and its impact on underrepresented groups, a topic also explored in his book.

The Constellations Center’s Collective Impact Creative Technology program organizes the CTRL+CM Speaker Series. The Oct. 16 event is open to those interested in AI, computing, education, public policy, and the social impact of emerging technologies.

Register for the Oct. 16 event.

News Contact

Let’s say you ask ChatGPT a question that stumps it, or a Waymo vehicle encounters something unusual in the road, or an autonomous factory faces an unexpected disruption.

Most people would recognize that something unexpected has happened and adjust accordingly.  

For machines, it’s not always that simple.  

Researchers at Georgia Tech’s H. Milton Stewart School of Industrial and Systems Engineering (ISyE) are addressing this very challenge: how can autonomous systems recognize when something unexpected has happened, determine whether it matters, and decide how to respond?

In a recent paper published in the INFORMS Journal on Data Science, the research team introduced a new framework called Mutual Information Surprise, designed to help machines identify meaningful surprises in complex environments. The work could help lay the foundation for AI agents, robots, and autonomous systems that can better navigate unexpected situations, avoid costly mistakes, and make more informed decisions when faced with situations they weren’t designed or trained to handle.  

"Traditional automation systems are designed to follow instructions. They do not have a sense of surprise. Machines follow a predetermined recipe and produce the expected result," said Yu Ding, the Anderson-Interface Chair and professor in ISyE, who is leading this project. " Intelligent agents, however, must do more than simply follow instructions. They need to recognize when their current understanding or reasoning process has become inadequate. Humans routinely use surprise for this purpose. An autonomous machine needs an analogous computational capability."

This research concept was jointly worked out by Xiao Liu, the David M. McKenney Family Associate Professor, as well as a former postdoctoral fellow in ISyE, Yinsong Wang, and Ph.D. student Quan Zeng.

While several methods already exist to measure surprise in machines, such as the "Shannon Surprise" and "Bayesian Surprise," the ISyE team saw limitations in both approaches.  

Shannon Surprise tends to focus on the rarity of an event, while Bayesian Surprise focuses on how much a new observation changes a system's beliefs.

To illustrate the limitations of existing surprise measures, Ding pointed to the examples of someone noticing a specific type of car parked nearby or a student who consistently earns top grades but suddenly performs poorly on an exam.

Existing surprise measures don't always flag the right kinds of moments to be surprised, Ding said. To the examples given, the Shannon Surprise would flag the specific car parked nearby, although people typically wouldn’t be surprised by that event. Conversely, people would be surprised by the student performing poorly, but the Bayesian Surprise would not flag it.    

"We introduce this new surprise definition because it measures the gain of knowledge and epistemic progression, thus capturing the right moments when a system should be surprised and ignoring the moments when it shouldn't," Ding added.  

The framework could eventually help systems determine when to continue as planned or when to stop and reassess, whether that's an AI assistant struggling to answer a question, a robot encountering a situation it has never seen before, or an autonomous factory detecting a disruption that could affect production and require human intervention.

Looking ahead, the ISyE research team is exploring potential applications in engineering autonomous systems and is developing a proposal for federally funded programs.  

News Contact

Artificial intelligence (AI) is becoming increasingly vital to the nation’s power grid. However, AI tools meant to help energy companies manage the grid, plan for future needs, and respond to problems also create new security and safety concerns.

Saman Zonouz, an associate professor in Georgia Tech’s School of Cybersecurity and Privacy and the School of Electrical and Computer Engineering, is helping address those concerns as a member of a planning committee for the National Academies of Sciences, Engineering, and Medicine.

Since May 2026, Zonouz has served on the planning committee for an upcoming public workshop on integrating AI into the electricity system. The workshop is scheduled for Oct. 26 and will convene experts to discuss how to safely and reliably use AI across the power grid.

The workshop will explore how AI is used today, the challenges of integrating AI into existing power systems, and how people and AI can work together. The event will also examine relationships between AI and cybersecurity, data centers, workforce development, and future research.

For Zonouz, cybersecurity and resilience are key parts of the conversation.

“AI is both a shield and a target for the grid,” Zonouz said.

Zonouz leads Georgia Tech’s Cyber-Physical Systems Security Research Laboratory. The lab’s research, much of it funded by the U.S. Department of Energy, investigates how AI can protect power systems from cyberattacks.

One recent initiative combines AI with detailed knowledge about power system operations to detect attacks in near real time. Another project sought to identify security weaknesses in solar inverters and other devices that connect energy sources, such as solar panels, to the power grid.

Regardless of the challenge, Zonouz and his lab work to identify and fix these weaknesses before attackers can exploit them. Despite their best efforts, however, Zonouz said that some AI tools meant to secure a system can create new ways for attackers to target a power grid.

The lab has demonstrated in several studies that AI systems can be tricked by false or manipulated data. Attackers can also target the software and systems that AI depends on. As AI becomes more advanced, some systems may even be able to modify grid operations.

This raises an important question: What happens if an AI system fails or is attacked?

“That is why the right frame is resilience,” Zonouz said.

For Zonouz, the goal is not to create an AI system that can never fail or be attacked. It’s to build a trustworthy power grid that can continue operating safely even if an AI system fails or is compromised.

Zonouz will bring his perspective to the National Academies workshop. In addition to helping organize the event, he will lead a session on cybersecurity and resilience. He will also help summarize the workshop’s key takeaways during the closing session. The workshop will be held virtually in late October.  

“Lessons learned from the electricity grid can be applied to other critical infrastructure as AI monitoring and control tools continue to advance and become more common,” he said.

News Contact

In-home assistive robots might be able to clean and organize your home, but if they can’t tell you where they put your keys or wallet, they can become a major inconvenience.

To improve robot performance and advance memory capabilities needed for reliable home use, Georgia Tech researchers have developed a new benchmark to evaluate the memory capabilities of vision-language models (VLMs).

That includes Gemini 2.0-flash and GPT-4o.

In-home assistive robots could clean and organize a home while the owner is away, but without enhanced memory capabilities, they could cause more problems than they solve.

They could, for example, cause a major inconvenience if they move important objects like wallets and keys but are unable to tell the owner where they put them.

A new benchmark created by Georgia Tech researchers tests the memory capability of vision-language models (VLMs), including Gemini 2.0-flash and GPT-4o, to determine whether they could improve robot performance.

Zsolt Kira, an associate professor in Tech’s School of Interactive Computing, said that most active VLMs can process only a few hundred images at a time before forgetting them.

“There wasn’t a benchmark that could test these specialized capabilities that we might want from a robot,” Kira said. “Our purpose in creating one is to spur research in this area so that others can develop methods to solve this problem.”

The team’s benchmark consists of memory evaluations based on 60 tasks that require continuous engagement and contextual and environmental awareness. Three task categories — spatial, temporal, and multi-goal — are highly difficult.

“You can chat with a language model like ChatGPT, and it remembers all the text from your previous conversations with it because the text data is small in terms of storage,” Kira said. 

“Embodied agents and robots observe a lot of videos, and those tend to be a lot harder to recall. If you ask it where your diary is, it must efficiently store images of your diary and where it is.

“That’s just a few frames out of a million that it recorded since yesterday, and it must know where that is relative to the spatial layout of the house.”

Karmesh Yadav, a Ph.D. student in Kira’s lab, said users don’t want to repeat information about their preferences after deployment. They also expect the robot to maintain long-term memory.

“In reality, you would want it to have a memory of the entire time it has been in your house, whether that’s days or years,” Yadav said.

The team’s benchmark evaluates memory performance over an average one-day period. It challenges the models with commands like “navigate to an object you did not interact with yesterday” or “navigate to a room you did not visit yesterday.”

The results show that current state-of-the-art models have a long way to go before they can be used in embodied agents. The best model achieved roughly a 50% success rate on high-level difficulty tasks.

Surprisingly, this model wasn’t Gemini or GPT VLMs. It was an open-source reasoning model from Qwen. 

“If you use an open-source reasoning model, even one that’s small in size, it can match the performance of closed-source non-reasoning models,” Yadav said. “Reasoning models can look at the video and translate it into text. It describes what’s going on in the video in words and determines the right frame to choose in its final output.”

Kira said that converting images into descriptive text that language models easily understand could be one way to circumvent the memory storage dilemma. Another way could be to assign a hierarchy that teaches the model to identify irrelevant or redundant images and erase them from storage.

Whatever future progress researchers make in those areas, Kira and his students are confident that their benchmark will remain relevant. 

“This benchmark is scalable,” said Ph.D. student Yusuf Ali. “If two years from now, videos are easy for a model to understand and store, you can use the same infrastructure we propose and scale up the problem’s difficulty.”

Yadav and Ali are co-first authors of a paper on the benchmark, presented in early September at the 2026 European Conference on Computer Vision in Sweden.

For more information about the project, click here.

News Contact

Artificial intelligence (AI) can prompt a student to reconsider an answer. It can also help a healthcare professional practice responding to a challenging workplace event.

Two Georgia Tech projects explore those distinct possibilities. College of Lifetime Learning researchers Meryem Yilmaz Soylu and Jeonghyun “Jonna” Lee collaborated with School of Biological Sciences lecturer Lesley Baradel to study AI-facilitated Socratic dialogue in large courses. In a separate project, Karen Peterson, a principal lecturer in the College of Lifetime Learning, used AI to develop clinical scenarios, pronunciation feedback, and vocabulary exercises for healthcare communication courses.

Both projects are featured in Georgia Tech’s first Academic AI Playbook, which shares practical examples of how faculty are using AI to support specific learning goals.

Encouraging Students to Think More Deeply

“One of the things that interested me most was the opportunity to extend meaningful academic conversations beyond the time and space available in a classroom,” Baradel said. “Using Socratic Mind in my classroom showed me how AI can facilitate reflection and discussion at a scale difficult to achieve through one-on-one interactions alone.”

Socratic Mind is an AI-powered conversational assessment tool developed by researchers in Georgia Tech’s College of Computing and College of Lifetime Learning. The team piloted the tool with more than 700 students in online and in-person courses. Their research examined whether AI could help students engage more deeply with course concepts by prompting them to explain their reasoning, apply their knowledge, and identify gaps in their understanding.

In Using AI-Facilitated Socratic Dialogue for Formative Learning in Large Undergraduate Courses, Baradel, Yilmaz Soylu, and Lee document the long-term collaboration among the instructor, researchers, and students to better understand how the tool may influence learning at cognitive and behavioral levels.

“Being part of the AI Playbook is especially meaningful because it allows us to share the idea behind the project and what we learned from putting it into practice with colleagues across Tech,” Yilmaz Soylu said. “Readers will see outcomes from research guided by student feedback and interaction data, which continually refined the activity and helped us better understand what makes AI-supported dialogue significant for learning.”

Building Communication Skills by Practice

In AI-Supported Healthcare Communication Training, Peters describes how she used AI to provide healthcare professionals and nurses with opportunities to develop communication skills through learning activities. Those skills include AIDET (Acknowledge, Introduce, Duration, Explanation, and Thank You) and SBAR (Situation, Background, Assessment, and Recommendation). She used generative AI to identify relevant healthcare vocabulary, develop instructional materials, create realistic clinical scenarios, and design activities tailored to students’ vocabulary gaps.

“One of the most important lessons from this work is that AI should be used to support a specific learning objective, not simply because the technology is available,” Peterson said. “When implemented intentionally, AI can help create more personalized opportunities for practice and feedback while supporting the learning goals that matter most.”

Keeping Learning at the Center

Results from the Socratic Mind pilot indicated that AI could promote greater engagement when it challenges students to think through and revise their ideas rather than simply providing answers. Peterson’s work expanded access to authentic healthcare communication experiences, giving learners opportunities to engage in workplace interactions that are difficult to replicate with traditional instructional materials. 

“Beyond the classroom outcomes, the Socratic Mind project generated valuable insights into how students engage with AI-guided educational experiences,” Lee said. “Those findings contribute to a growing body of research that helps educators better understand when and how generative AI can support meaningful learning and inform its thoughtful use in higher education.”

Together, these projects demonstrate that AI’s value in education goes beyond efficiency or automation. Whether prompting students to reflect more deeply on course concepts or helping healthcare professionals practice high-stakes communication, the most effective applications are grounded in clear learning goals. AI can enrich teaching and learning when intentionally designed to support meaningful engagement, practice, and feedback. 

“Rather than replacing existing teaching practices, AI prompted me to reconsider which aspects of the learning experience benefit most from human expertise and which can be enhanced by technology,” Peterson said. “It also opened new possibilities for designing learning experiences that are both responsive and scalable.”

Interim Vice Provost for AI in Education David A. Joyner and Nimisha Roy, a lecturer in the School of Computing Instruction, edited the 2026 edition of the Georgia Tech Academic AI Playbook. The authors of the featured articles, Peterson, Baradel, Yilmaz Soylu, and Lee, hope their work will inspire new ideas for using AI thoughtfully and purposefully to support teaching and learning.

News Contact

News Contact