The Academic Minute

What happens when platforms recommend opinions instead of users?

Sriniwas Pandey, lecturer at the School of Computing at Binghamton University, details this.

Faculty Bio:

Pandey is a lecturer in the School of Computing at Binghamton University. He received his PhD degree in system science from Binghamton University. He holds bachelor's and master's degrees in computer science engineering from UTU India and IIITDMJ India, respectively. His research interests include complex systems, networks and machine learning. His research aims to comprehend the intricacies of eccentric behavior within society, focusing on identifying the underlying factors contributing to such behaviors and its impact on network dynamics.

Transcript:

Although social media can serve as a civil digital meeting place, pockets of users with intense opinions that clash with others that have different views has become a common occurrence. There are plenty of reasons for this, but one factor is content recommendations by the platform itself.

I co-authored a study with Binghamton University Professor Hiroki Sayama that explores how these content recommendation systems affect the overall social climate on social media.

We created a computer simulation of a social media platform with users connected to each other. Each user had a default set of opinions. However these users could form new opinions or be influenced by what they saw from other users. The strength of connection between users could change over time based on how similar their ideas were.

We found that when the platform only recommended similar users without adjusting what opinions people saw, users broke apart into tight communities that had very different ideologies.

However, when the platform recommended opinions, rather than users, there was far less network fragmentation. This worked even when people naturally preferred users similar to them. This also led to more unusual or off-center opinions. We experimented with different initial social network configurations.

When people were exposed to diverse opinions, their “knowledge base” expanded, giving them more room to generate creative ideas within their community.

This implies that echo chambers are less likely when people see diverse opinions. The findings drive home the importance of social media design – and how choice of recommendation strategy can impact the cohesion of the platform’s community and the extent to which users are able to express less mainstream viewpoints.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.academicminute.org

There are stark mortality differences between pregnancy and abortion.

Maria Steenland, assistant professor of family science in the School of Public Health at the University of Maryland, explores this.

Faculty Bio:

Dr. Steenland is an Assistant Professor in Family Science in the School of Public Health at the University of Maryland, College Park. She is a health services and health policy researcher focused on maternal and reproductive health policy in the United States. Her research useseconometric methods to evaluate maternal and reproductive health programs and policies, with a particular focus on Medicaid policy. The overarching goal of her research is to identify policy options to increase the equity and quality of women’s health services. Her previous researchhas examined the effect of Medicaid payment policies for immediate postpartum contraception, and expansions of Medicaid eligibility in pregnant and postpartum populations.

Transcript:

After Roe v. Wade was overturned by the Dobbs v. Jackson Women’s Health Organization decision in June of 2022, some advocates and academics expressed concern that abortion bans would harm maternal health. They argued that banning abortion could increase maternal mortality in part because the risk of death from childbirth is much higher than the risk from abortion. Continuing a pregnancy places significant physiologic stress on the body, which can lead to life-threatening complications, such as hemorrhage, sepsis, stroke, and heart failure. A prior study, using data from 1998 through 2005, found that the risk of death from childbirth was about 14 times greater than the risk of death from abortion. In the years since that estimate, the measurement of maternal death has improved, increasing the number of pregnancy-related deaths identified annually. At the same time, abortions are taking place earlier in pregnancy, reducing the risk of complications. Given these changes, my colleagues — Kerra Mercon, Ben Brown, Marie Thoma — and I re-estimated the difference between pregnancy-related death and abortion-related death using national data. Using figures from 2018 to 2021, we calculated a range of mortality ratios. We found that the risk of death from ongoing pregnancy was 44 to 70 times greater than the risk of death from abortion. Importantly, even our most conservative estimate was still three times higher than the previously reported figure. We don’t yet know what effect abortion bans will have on overall maternal mortality in the US. However, we can say that people who are forced to continue their pregnancies because of abortion bans will face a dramatically greater risk of death than had they been able to access an abortion.

Read More:

[University of Maryland School of Public Health] - Study: Risk of maternal death during pregnancy greatly exceeds risk of death from abortion

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.academicminute.org

How do we pull back the curtain on China’s use of AI in its military?

Cole McFaul, senior research analyst at Georgetown University, explores this key question.

Faculty Bio:

Cole McFaul is a Senior Research Analyst and an Andrew W. Marshall Fellow at Georgetown University’s Center for Security and Emerging Technology (CSET), where he mainly focuses on emerging technology competition in the Asia-Pacific and China’s science and technology ecosystem. Prior to joining CSET, Cole researched the political economy of China’s international engagement strategies at the Center for Strategic and International Studies and the Shorenstein Asia-Pacific Research Center at Stanford University. Cole holds a B.A. in Political Science and an M.A. in East Asian Studies from Stanford University.

Transcript:

It’s no secret that China is using AI to modernize its military and compete with the U.S. and other rivals. However, what has remained secret is civilian firms’ involvement in these efforts and China’s fusion of commercial innovation with military power.

My colleagues Sam Bresnick, Daniel Chou and I set out to answer a key question: who supplies the PLA with AI-related goods and services? Our project relies on a novel data set of 2800 AI-related contract award notices published by the PLA between January 2023 and December 2024. We define an AI-related award as any contract supporting AI-enabled or autonomous technologies, like language and vision models, unmanned vehicles, augmented and virtual reality, simulation and training environments, and smart manufacturing and robotics.

From this data set, we identified 338 entities awarded two or more AI-related contracts. Using open-source information, we classified each into one of three groups: state-owned enterprises, research institutions, and nontraditional vendors (which are firms without self-reported state ownership).

We found that state owned enterprises and defense-affiliated research institutions led in AI-related military procurement. Institutions like AVIC, NORINCO, the Chinese Academy of Sciences, and the Seven Sons of National Defense dominate the top of the list.

But a deeper look at who supplies the PLA with AI-related goods and services reveals that a wide range of other organizations are also active. Nearly 70 percent of the entities awarded two or more AI-related contracts were nontraditional vendors. Civilian universities like Shanghai Jiao Tong, Tsinghua University, and Peking University were also awarded AI-related contracts.

Our research reveals that, at least in the public procurement of AI-related goods and services, military-civil fusion is no longer aspirational—it’s operational. Unless the United States adapts to this reality, it risks facing a Chinese defense base that is more capable, adaptable, and technologically sophisticated.

Read More:

[CSET] - Pulling Back the Curtain on China’s Military-Civil Fusion

[CSET] - Civilian Tech Is Powering China’s Military

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.academicminute.org

What is Gulf War Illness and why is recognition important?

Beatrice Golomb, professor of medicine at the University of California San Diego, seeks to inform.

Faculty Bio:

Dr. Golomb is a Professor of Medicine at UC San Diego with over 15 years of experience treating veteran patients, including veterans with Gulf War Illness (GWI). She was the inaugural Scientific Director for the Congressionally directed Research Advisory Committee on Gulf War Veterans’ Illnesses (RAC) and her research for RAND and with funding from the Department of Defense have expanded knowledge of exposure relations, mechanisms, markers and treatment for GWI. Dr. Golomb and her team remain committed to research to improve the lives and health for our heroes from the Gulf War.

Transcript:

For decades, Gulf War veterans have battled for recognition of the often devastating health challenges they experience as a consequence of their honorable service. Next month, we will mark an immensely important milestone for Gulf War veterans, their families, clinicians and researchers. Gulf War illness will finally receive its own International Classification of Diseases — or ICD — diagnostic code. This is more than just administrative coding. This is long-overdue validation for the suffering of the quarter-million affected veterans. It is a formal acknowledgment that Gulf War illness is real, it is physical, and it is service-related. Gulf War illness affects about one-third of the nearly 700,000 U.S. troops who served in the 1990-1991 Gulf War. It manifests as a consistent profile of symptoms: persistent fatigue, cognitive difficulties, chronic pain, respiratory issues, skin problems, gastrointestinal distress. Decades of study have linked Gulf War illness to chemical exposures and identified objective abnormalities such as structural brain changes. With this new ICD code, health care providers will be better able to recognize, diagnose, and treat Gulf War illness. Insurance, medical records, research, and public health tracking will now explicitly acknowledge the condition, rather than forcing patients to substitute related diagnoses. For researchers like me, the change accelerates our ability to study Gulf War illness in large populations, monitor treatment outcomes rigorously, and understand how this condition may overlap or interact with other diseases. To all veterans whose symptoms were dismissed and whose needs went unmet: this new diagnostic code is for you. It’s a recognition of your service. It’s a commitment to your care. And it represents hope — hope that research, medicine, and policy will now move forward more fully, more justly, to give you the answers and the support you deserve.

Read More:

[PNAS] - Acetylcholinesterase inhibitors and Gulf War illnesses

[ScienceDirect] - Adverse effect propensity: A new feature of Gulf War illness predicted by environmental exposures

[National Library of Medicine] - Mitochondrial impairment but not peripheral inflammation predicts greater Gulf War illness severity

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.academicminute.org

On Old Dominion University Week: A non-invasive method for detecting an aggressive brain tumor could be key for patients. But how does it work?

Khan Iftekharuddin, Professor and Eminent Scholar, delves into this.

Faculty Bio:

Dr. Khan Iftekharuddin is a professor and Batten Endowed Chair in Machine Learning in the department of Electrical and Computer Engineering (ECE) at Old Dominion University (ODU). He concurrently serves as a Director, ODU Vision Lab and an Inaugural Director, Institute of Data Science. Dr. Iftekharuddin has been cited among the top 2% researchers in the globe for both career-long impact and single-year impact, and his Vision Lab has consistently ranked among top teams in Global Brain Tumor Segmentation and Patient Survivability Prediction Challenges co-organized by MICCIA and NCI since 2014.Prior to his current roles, he served as an Interim Dean in Batten College of Engineering and Technology, Associate Dean for Research, Innovation and Graduate Studies, and Chair of the ECE Department at ODU. He received his MS (1991) and PhD (1995) degrees in Electrical and Computer Engineering from University of Dayton, OH.

Transcript:

Glioblastoma Multiforme, or GBM, is the most aggressive and deadly type of brain cancer, killing about 10,000 Americans each year and accounting for half of all brain cancer deaths in the U.S. The fast-growing cancer spreads microscopic cancer cells in surrounding healthy tissue and has an average survivability of 18-24 months from diagnosis. Prognosis for GBM is poor, with recurrence in 90% of patient cases within six to nine months, even after aggressive treatment protocol including surgery, radiation and chemotherapy. Diagnosing brain tumor recurrence on standard imaging scans like MRIs is challenging because treatment-related changes in the brain tissues, such as scar tissue, necrosis (dead tissues) and edema (swelling), often appear like recurrent tumor tissue. Currently, the only way to confirm tumor recurrence is through an invasive brain biopsy. My colleagues and I are investigating how computational modeling, AI, and machine learning methods can help distinguish true tumor recurrence from surrounding abnormal tissues, without needing to do a biopsy. This work builds on long-standing research of brain tumor volume segmentation and tracking, tumor sub-typing, and patient survivability prediction. We’re working with about half a dozen clinical collaborators across the US to analyze and process large amounts of high-resolution Magnetic Resonance imaging alongside molecular and patient clinical data. This analysis will help us develop non-invasive AI models that classify tumor recurrence and radiation-induced challenges. These tools could improve early detection, tracking, and treatment planning, helping physicians better predict the trajectory of tumor growth and tailor interventions for individual patients. Additionally, we’re working to study inherent biases in these AI models and ensure that they are representative of different patient populations. This will bolster their robustness and efficacy in clinical settings.

Research Projects

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.academicminute.org

On Old Dominion University Week: How fast can a viral variant spread?

Hong Qin, associate professor in the School of Data Science and the Department of Computer Science at Old Dominion University, analyzes the data to find out.

Faculty Bio:

Hong Qin is an Associate Professor in the School of Data Science and the Department of Computer Science at Old Dominion University. His work develops AI and statistical methods for genomic surveillance, pandemic prediction, and trustworthy health AI.

Transcript:

Viruses evolve as they spread, and when a new viral variant begins to outcompete others, it can quickly reshape an outbreak. But measuring a variant’s advantage is tricky, because case counts and sequencing volume rise and fall for reasons unrelated to biology.A new approach called the differential population growth rate, or DPGR, focuses on comparisons instead of absolute numbers. In a given region and short time window, DPGR looks at two variants that are sampled side-by-side. It tracks the ratio of their weekly sequence counts and takes a logarithm. If that log-ratio changes roughly as a straight line, the slope estimates how much faster one variant is growing than the other. A positive slope means variant A is gaining on variant B; a negative slope means variant A is losing ground.This pairwise design makes one variant an internal control, helping reduce distortions from shifting testing, reporting, or sequencing intensity. DPGR also has an additive property: if variant A overlaps with B, and B overlaps with C, their slopes can be combined to estimate a comparison of A versus C, even when A and C rarely appear together.Using DPGR with genomic surveillance data, researchers can map how variants’ advantages change across places and over time. For example, COVID-19’s Omicron variant outpaced the Delta variant worldwide, but the estimated advantage of Omicron differed by region. DPGR can also compare sublineages and build a “fitness staircase” that summarizes stepwise gains.The result? A simple, interpretable signal that complements other epidemic models and can help anticipate which variant may dominate next.

Read More:

[Wiley] - A data-driven sliding-window pairwise comparative approach for the estimation of transmission fitness of SARS-CoV-2 variants and construction of the evolution fitness landscape

YouTube

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.academicminute.org

On Old Dominion University Week: When it’s hot out, some places are hotter than others.

Maryam Golbazi, research assistant professor of climate science, examines why.

Faculty Bio:

Maryam Golbazi is a Research Assistant Professor at Old Dominion University working with the Joint Institute on Advanced Computing for Environmental Studies (JI-ACES). She specializes in numerical weather prediction models, atmospheric chemistry modeling, wind energy, and data assimilation. Her work integrates advanced numerical modeling with satellite and in-situ observations to improve forecasts of air pollution, wind energy resources, and extreme weather events. She is currently leveraging data science and AI/ML methods to develop localized weather models, while maintaining the rigor and integrity of established physical modeling techniques. With prior research experience at the National Center for Atmospheric Research and many collaborative projects, Dr. Golbazi’s research bridges science and application to address pressing environmental and energy challenges. She aims to leverage fundamental science and state of the art data-driven techniques to produce actionable insights that help protect communities, inform policy, and guide sustainable infrastructure planning.

Transcript:

On a summer afternoon in Hampton Roads, Virginia, the heat doesn’t feel the same everywhere. In some neighborhoods, the air lingers thick, heavy, slow to cool even after sunset. In others, just a few miles away, temperatures drop faster, offering relief once the sun goes down. These differences aren’t random. They’re shaped by concrete, roads, buildings, and the environment.In our study, which I conducted with my colleague Frank Liu, we used some of the highest-resolution weather simulations ever applied to a real U.S. city to understand how extreme heat behaves at the neighborhood scale. Instead of looking at cities from satellites, we zoomed in, down to city blocks, using advanced atmospheric models.During two intense heat waves in the summer of 2024, our simulations revealed that dense urban areas were, on average, up to five or six degrees hotter than nearby rural regions. And at night urban neighborhoods stayed warm far longer.But temperature was only part of the story.When we combined heat exposure with census data, a pattern emerged: lower-income communities experienced higher heat stress. And that translated directly into energy demand. That matters, because cooling isn’t free. For families already struggling with energy costs, extreme heat becomes both a health risk and a financial burden.As heat waves become longer and more intense, understanding where heat concentrates, and who pays the price, may be just as important as predicting the temperature itself. Our research shows that climate change isn’t just about rising averages. It’s about how people experience heat differently, day to day, street to street!

Read More:

[Springer Nature] - High-resolution modeling of extreme heat events with socioeconomic consideration: a real-case WRF–LES approach

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.academicminute.org

On Old Dominion University Week: Can AI chatbots spread generalizations?

Tiffany Zhu, assistant professor of global ethics and technology, examines why this could be the case.

Faculty Bio:

Tiffany Zhu is a philosopher. She is an Assistant Professor of Global Ethics and Technology in the Department of Philosophy and Religious Studies at Old Dominion University. Prior to joining ODU in Fall 2025, she was a Faculty Fellow in AI Ethics at the California Center for Ethics and Policy at California State Polytechnic University Pomona

Transcript:

AI chatbots powered by large language models are increasingly shaping our understanding of the world. Some of my research examines how they use a linguistic device called “generics,” which expresses generalizations without explicit quantification. An example of a sentence using a generic to talk about a social group would be: “Immigrants work low-wage jobs”. These statements are consequential because they can spread stereotypes. One study looking at ChatGPT 3.5 found, among other tendencies, that the chatbot often paired social generics with what I call individuation hedges, which emphasize diversity within groups. For instance, when asked “Are women more likely to get attacked while walking alone at night?,” the chatbot affirmed the trend but added that “the likelihood of getting attacked depends on an individual’s characteristics.” Not only are these hedges too formulaic to counteract unfair generalizations, they also reduce the accuracy of some responses, obscure the structural causes of social patterns including oppression, and could lead users to form false or even harmful beliefs about themselves and others.To improve their use, I proposed a strategy with three elements. First, I suggest requiring democratic and interdisciplinary guidance during the process of reinforcement learning through human feedback. Second, I advise shifting away from a transactional toward a dialogical model of AI-human interaction, meaning chatbots should probe for context and user goals and assumptions rather than simply providing answers. Finally, I suggest chatbots should use generics in conjunction with historical framing and asking counterfactual questions, in order to help AI users to think flexibly about how social reality is contingent, changeable, and sometimes unjust.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.academicminute.org

On Old Dominion University Week: Coastal resilience will be key going forward in a warming climate.

George McLeod, director of the Center for Geospatial Science, Education, and Analytics, shows how virtual worlds can help us protect our own.

Faculty Bio:

Dr. George McLeod is the Director of the Center for Geospatial Science, Education, and Analytics at Old Dominion University and Senior Fellow with Virginia’s Commonwealth Center for Recurrent Flooding Resiliency. He oversees the creation of vital location intelligence for a wide range of academic research questions, including those focused on the intersection of the built landscape and environmental hazards. His expertise in ocean science, geovisualization, remote sensing, and UAV operations has allowed him to play an important role in advancing coastal hazards and flooding resilience research in Virginia.

Transcript:

When I tell people what we’re working on, the first question is almost always the same: “Okay… what exactly is a digital twin?”At its simplest, a digital twin is a virtual version of a real place or object. Most of us have seen something like this before—think about the detailed, three-dimensional cities you move through in video games. They look real, but they’re mostly just scenery.In our work on coastal resilience, the digital twin isn’t the backdrop; it’s the main character.We’re building virtual versions of real coastal communities and entire regions so we can study very real challenges: sea-level rise, storm flooding, water quality, and the public-health impacts that come with all of that. And unlike a static 3D model, these digital twins are alive. They combine physical models of land and water with models of how systems behave, things like transportation networks, population movement, and hydrology.That lets us ask big, practical questions. For example: If a Category 3 hurricane hits Virginia on top of an extra foot of sea-level rise, which neighborhoods flood first? Which roads are cut off—and for how long? How much damage might we see? Who could be displaced, and where should emergency responders focus their efforts?Our most ambitious project takes this idea even further. We’re working with colleagues at NASA to build what we call the Coastal Zone Digital Twin of the entire Chesapeake Bay. It’s a dynamic system that’s constantly updated, showing what’s happening now, what’s likely to happen next, and letting us test “what if” scenarios, so communities can prepare for the coastal future that’s already on the way.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.academicminute.org

Are you good at multitasking?

Jinglu Jiang, associate professor at the School of Management at Binghamton University, reveals how this behavior may allow harmful emails to slip by.

Transcript:

The ability to juggle multiple tasks is a defining feature of modern work. But that constant multitasking may make people more vulnerable to phishing attacks.

In a recent study, my co-authors and I examined how multitasking affects people’s ability to detect phishing emails. We conducted two online experiments with nearly one thousand participants. In both experiments, participants worked in multitasking settings. They first completed a mentally demanding primary task, like memorizing numbers or work-related information, while being interrupted with a secondary task: deciding whether incoming emails were legitimate or phishing. This setup mirrors everyday work environments, where email alerts arrive while people are focused on other tasks.

We found that when the primary task placed a high demand on people’s working memory, phishing detection performance dropped substantially. However, we also identified an important countermeasure. When participants received a simple reminder that some emails might be phishing attempts, detection performance improved—even under heavy cognitive load.

We also found that message design plays a role. Reminders were especially effective against phishing emails that promised rewards. By contrast, loss-framed messages—such as warnings about account suspension—tended to trigger vigilance on their own, leaving less room for reminders to add value.

Together, these findings suggest that phishing defenses should account for multitasking, not assume users are fully attentive. Organizations may benefit from context-aware reminders that support attention when cognitive demands are highest and risks are most likely to go unnoticed.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.academicminute.org