Aurelia Butler

Many policy discussions on AI safety regulation have focused on the need to establish regulatory “guardrails” to protect the public from the risks of AI technology. In a new paper published in the journal Risk Analysis, two experts argue that, instead of imposing guardrails, policymakers should demand “leashes.”
Director of the Penn Program on Regulation and professor at University of Pennsylvania Carey Law School, Cary Coglianese and University of Notre Dame computer science doctoral candidate Colton R. Crum explain that management-based regulation (a flexible “leash” strategy) will work better than a prescriptive guardrail approach, as AI is too heterogeneous and dynamic to operate within fixed lanes. Leashes “are flexible and adaptable — just as physical leashes used when walking a dog through a neighborhood allow for a range of movement and exploration,” the authors write. Leashes “permit AI tools to explore new domains without regulatory barriers getting in the way.”
The various applications of AI include social media, chatbots, autonomous vehicles, precision medicine, fintech investment advisors, and many more. While AI offers benefits for society, such as, to pick but one example, the ability to find evidence of cancerous tumors that well-trained radiologists can miss, it also can pose risks.
In their paper, Coglianese and Crum offer three examples of AI risks: autonomous vehicle (AV) collisions, suicide associated with social media, and bias and discrimination brought about by AI through a variety of applications and digital formats, such as AI-generated text, images, and videos.
With flexible management-based regulation, firms using AI tools that pose risks in each of these settings — and others — would be expected to put their AI tools on a leash by creating internal systems to anticipate and reduce the range of possible harms from the use of their tools.
Management-based regulation can flexibly respond to “AI’s novel uses and problems and better allows for technological exploration, discovery, and change,” write Coglianese and Crum. At the same time, it provides “a tethered structure that, like a leash, can help prevent AI from ‘running away.'” More

Researchers gave participants face tattoos that can track when their brain is working too hard. Published May 29 in the Cell Press journal Device, the study introduces a non-permanent wireless forehead e-tattoo that decodes brainwaves to measure mental strain without bulky headgear. This technology may help track the mental workload of workers like air traffic controllers and truck drivers, whose lapses in focus can have serious consequences.
“Technology is developing faster than human evolution. Our brain capacity cannot keep up and can easily get overloaded,” says Nanshu Lu, the study’s author, from the University of Texas at Austin (UT Austin). “There is an optimal mental workload for optimal performance, which differs from person to person.”
Humans perform best in a cognitive Goldilocks zone, neither overwhelmed nor bored. Finding that balance is key to optimal performance. Current mental workload assessment relies on the NASA Task Load Index, a lengthy and subjective survey participants complete after performing tasks.
The e-tattoo offers an objective alternative by analyzing electrical activity from the brain and eye movement, in processes known as electroencephalography (EEG) and electrooculography (EOG). Unlike EEG caps that are bulky with dangling wires and lathered with squishy gel, the wireless e-tattoo consists of a lightweight battery pack and paper-thin, sticker-like sensors. These sensors feature wavy loops and coils, a design that allows them to stretch and conform seamlessly to the skin for comfort and clear signals.
“What’s surprising is those caps, while having more sensors for different regions of the brain, never get a perfect signal because everyone’s head shape is different,” says Lu. “We measure participants’ facial features to manufacture personalized e-tattoos to ensure that the sensors are always in the right location and receiving signals.”
The researchers tested the e-tattoo on six participants who completed a memory challenge that increased in difficulty. As mental load rose, participants showed higher activity in theta and delta brainwaves, signaling increased cognitive demand, while alpha and beta activity decreased, indicating mental fatigue. The results suggest that the device can detect when the brain is struggling.
The device didn’t stop at detection. It could also predict mental strain. The researchers trained a computer model to estimate mental workload based on signals from the e-tattoo, successfully distinguishing between different levels of mental workload. The results show that the device can potentially predict mental fatigue.
Cost is another advantage. Traditional EEG equipment can exceed $15,000, while the e-tattoo’s chips and battery pack costs $200, and disposable sensors are about $20 each. “Being low cost makes the device accessible,” says author Luis Sentis from UT Austin. “One of my wishes is to turn the e-tattoo into a product we can wear at home.”
While the e-tattoo only works on hairless skin, the researchers are working to combine it with ink-based sensors that work on hair. This will allow for full head coverage and more comprehensive brain monitoring. As robots and new technology increasingly enter workplaces and homes, the team hopes this technology will enhance understanding of human-machine interaction.
“We’ve long monitored workers’ physical health, tracking injuries and muscle strain,” says Sentis. “Now we have the ability to monitor mental strain, which hasn’t been tracked. This could fundamentally change how organizations ensure the overall well-being of their workforce.” More

Medical professionals have been using artificial intelligence (AI) to streamline diagnoses for decades, using what are called diagnostic decision support systems (DDSSs). Computer scientists at Massachusetts General Hospital (MGH), a founding member of the Mass General Brigham healthcare system first developed MGH’s own DDSS called DXplain in 1984, which relies on thousands of disease profiles, clinical findings, and data points to generate and rank potential diagnoses for use by clinicians. With the popularization and increased accessibility of generative AI and large language models (LLMs) in medicine, investigators at MGH’s Laboratory of Computer Science (LCS) sought to compare the diagnostic capabilities of DXplain, which has evolved over the past four decades, to popular LLMs.
Their new research compares ChatGPT, Gemini, and DXplain at diagnosing patient cases, revealing that DXplain performed somewhat better, but the LLMs also performed well. The investigators envision pairing DXplain with an LLM as the optimal way forward, as it would improve both systems and enhance their clinical efficacy. The results are published in JAMA Network Open.
“Amid all the interest in large language models, it’s easy to forget that the first AI systems used successfully in medicine were expert systems like DXplain,” said co-author Edward Hoffer, MD, of the LCS at MGH.
“These systems can enhance and expand clinicians’ diagnoses, recalling information that physicians may forget in the heat of the moment and isn’t biased by common flaws in human reasoning. And now, we think combining the powerful explanatory capabilities of existing diagnostic systems with the linguistic capabilities of large language models will enable better automated diagnostic decision support and patient outcomes,” said corresponding author Mitchell Feldman, MD, also of MGH’s LCS.
The investigators tested the diagnostic capabilities of DXplain, ChatGPT, and Gemini using 36 patient cases spanning racial, ethnic, age, and gender categories. For each case, the systems had a chance to suggest potential case diagnoses both with and without lab data. With lab data, all three systems listed the correct diagnosis most of the time: 72% for DXplain, 64% for ChatGPT, and 58% for Gemini. Without lab data, DXplain listed the correct diagnosis 56% of the time, outperforming ChatGPT (42%) and Gemini (39%), though the results were not statistically significant.
The researchers observed that the DDSS and LLMs caught certain diseases the others missed, suggesting there may be promise in combining the approaches. Preliminary work building off these findings reveals that LLMs could be used to pull clinical findings from narrative text, which could then be plugged into DDSSs — in turn synergistically improving both systems and their diagnostic conclusions. More

Interactive robots should not just be passive companions, but active partners-like therapy horses who respond to human emotion-say University of Bristol researchers.
Equine-assisted interventions (EAIs) offer a powerful alternative to traditional talking therapies for patients with PTSD, trauma and autism, who struggle to express and regulate emotions through words alone.
The study, presented at the CHI ’25: Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems held in Yokohama, recommends that therapeutic robots should also exhibit a level of autonomy, rather than one-dimensional displays of friendship and compliance.
Lead author Ellen Weir from Bristol’s Faculty of Science and Engineering explains: “Most social robots today are designed to be obedient and predictable — following commands and prioritising user comfort.
“Our research challenges this assumption.”
In EAIs, individuals communicate with horses through body language and emotional energy. If someone is tense or unregulated, the horse resists their cues. When the individual becomes calm, clear, and confident, the horse responds positively. This ‘living mirror’ effect helps participants recognise and adjust their emotional states, improving both internal well-being and social interactions.
However, EAIs require highly trained horses and facilitators, making them expensive and inaccessible.

Ellen continued: “We found that therapeutic robots should not be passive companions but active co-workers, like EAI horses.
“Just as horses respond only when a person is calm and emotionally regulated, therapeutic robots should resist engagement when users are stressed or unsettled. By requiring emotional regulation before responding, these robots could mirror the therapeutic effect of EAIs, rather than simply providing comfort.”
This approach has the potential to transform robotic therapy, helping users develop self-awareness and regulation skills, just as horses do in EAIs.
Beyond therapy, this concept could influence human-robot interaction in other fields, such as training robots for social skills development, emotional coaching, or even stress management in workplaces.
A key question is whether robots can truly replicate — or at least complement — the emotional depth of human-animal interactions. Future research must explore how robotic behaviour can foster trust, empathy, and fine tuning, ensuring these machines support emotional well-being in a meaningful way.
Ellen added: “The next challenge is designing robots that can interpret human emotions and respond dynamically — just as horses do. This requires advances in emotional sensing, movement dynamics, and machine learning.
“We must also consider the ethical implications of replacing sentient animals with machines. Could a robot ever offer the same therapeutic value as a living horse? And if so, how do we ensure these interactions remain ethical, effective, and emotionally authentic?” More

Imagine cells navigating through a complex maze, guided by chemical signals and the physical landscape of their environment. A team of researchers at the University of Maryland, Baltimore County (UMBC) has contributed an important discovery about how cells move, or migrate, through this maze of bodily tissues, using the fruit fly egg chamber as a model system. Potential implications include better understanding of diseases like cancer and advancing medical treatments.
Published iniScience, the team’s study combines biological experiments and mathematics to reveal new insights into cell migration. By integrating mathematical modeling with advanced imaging, the team discovered that the physical shape of the egg chamber, combined with chemical signals called chemoattractants, significantly influences how cells move.
“This paper takes an interdisciplinary focus with tight collaboration between a mathematical framework and experimental design,” says UMBC mathematician and co-author Brad Peercy. “The results promote the idea that complex distribution of chemical attractants can explain specific variations in migratory movement.”
Peercy’s enthusiasm highlights the study’s innovative approach, which merges precise mathematical models with real-world biological experiments to uncover patterns that were previously invisible.
Following the breadcrumbs
The team’s work focuses on border cells, a type of cell in fruit fly egg chambers, which are a model system for studying cell migration because of their similarities to processes in human development and disease. The team found that the border cells’ movement wasn’t only driven by continuously increasing chemical concentrations from one end of the egg chamber to the other, as earlier models suggested. Instead, the physical structure of the tissue — narrow tubes alternating with wider gaps — played a critical role.
“This was the first time that we characterized that there were these patterns of migration behavior that ended up correlating to aspects of the tissue geometry,” explains biologist Alex George, a co-author who completed his Ph.D. at UMBC in 2024 and will begin a postdoctoral fellowship at the Geisel School of Medicine at Dartmouth in a few weeks. He likens the migration process to Hansel and Gretel following breadcrumbs through a forest: On a flat plain, the trail is clear, but in a landscape with ravines and valleys, the breadcrumbs pool in unexpected ways, complicating the path.

To understand this, co-author Naghmeh Akhavan, who completed her Ph.D. in mathematics at UMBC this spring, developed mathematical models that simulate how cells respond to both chemical signals and tissue geometry together. “Alex’s experiments showed that the speed is not exactly the way previous models showed it,” she says. Her models revealed that cells speed up in narrow tubes and slow down in larger gaps, a pattern confirmed by George’s imaging.
Both approaches — wet-lab experiments and modeling — bring unique strengths to the work. Putting them together “is like unveiling the invisible from two different perspectives,” George says. “My experiments would refine her model, and her model would refine my experiments.”
And then, “When our model shows exactly what Alex found in his experiments, we love that,” Akhavan adds.
New strategies, new discoveries
The study’s broader impact lies in its potential to inform fields beyond developmental biology. Cell migration is critical in processes like wound healing, immune responses, and cancer metastasis.
“Most research on how cells navigate the world has focused only on chemical signals or only on structural ones, so this is one of the first studies to consider how those two things impact each other, which is likely to be relevant in many cases,” explains UMBC biologist and co-author Michelle Starz-Gaiano. By showing how tissue geometry and chemical signals interact, the research could guide new strategies for controlling cell movement via medical treatments.
The team’s work continues to evolve, including recent experiments at the Advanced Imaging Center at the Janelia Research Campus in Virginia, where George used specialized microscopes to capture previously unseen dynamics of the relevant chemoattractants. These findings will further refine the team’s models, opening new avenues for research.
“We are developing new experimental strategies both on the biology and the math side of things,” Starz-Gaiano says, “so it will be exciting to see where this will take us next.” More

Specialized robots that can both fly and drive typically touch down on land before attempting to transform and drive away. But when the landing terrain is rough, these robots sometimes get stuck and are unable to continue operating. Now a team of Caltech engineers has developed a real-life Transformer that has the “brains” to morph in midair, allowing the dronelike robot to smoothly roll away and begin its ground operations without pause. The increased agility and robustness of such robots could be particularly useful for commercial delivery systems and robotic explorers.
The new robot, dubbed ATMO (aerially transforming morphobot), uses four thrusters to fly, but the shrouds that protect them become the system’s wheels in an alternative driving configuration. The whole transformation relies on a single motor to move a central joint that lifts ATMO’s thrusters up into drone mode or down into drive mode.
The researchers describe the robot and the sophisticated control system that drives it in a paper recently published in the journal Communications Engineering.
“We designed and built a new robotic system that is inspired by nature — by the way that animals can use their bodies in different ways to achieve different types of locomotion,” says Ioannis Mandralis (MS ’22), a graduate student in aerospace at Caltech and lead author of the new paper. For example, he says, birds fly and then change their body morphology to slow themselves down and avoid obstacles. “Having the ability to transform in the air unlocks a lot of possibilities for improved autonomy and robustness,” Mandralis says.
But midair transformation also poses challenges. Complex aerodynamic forces come into play both because the robot is close to the ground and because it is changing its shape as it morphs.
“Even though it seems simple when you watch a bird land and then run, in reality this is a problem that the aerospace industry has been struggling to deal with for probably more than 50 years,” says Mory Gharib (PhD ’83), the Hans W. Liepmann Professor of Aeronautics and Medical Engineering, director and Booth-Kresa Leadership Chair of Caltech’s Center for Autonomous Systems and Technologies (CAST), and director of the Graduate Aerospace Laboratories of the California Institute of Technology (GALCIT). All flying vehicles experience complicated forces close to the ground. Think of a helicopter, as an example. As it comes in for a landing, its thrusters push lots of air downward. When that air hits the ground, some portion of it bounces back up; if the helicopter comes in too quickly, it can get sucked into a vortex formed by that reflected air, causing the vehicle to lose its lift.
In ATMO’s case, the level of difficulty is even greater. Not only does the robot have to contend with complex near-ground forces, but it also has four jets that are constantly altering the extent to which they are shooting toward each other, creating additional turbulence and instability.
To better understand these complex aerodynamic forces, the researchers ran tests in CAST’s drone lab. They used what are called load cell experiments to see how changing the robot’s configuration as it came in for landing affected its thrust force. They also conducted smoke visualization experiments to reveal the underlying phenomena that lead to such changes in the dynamics.
The researchers then fed those insights into the algorithm behind a new control system they created for ATMO. The system uses an advanced control method called model predictive control, which works by continuously predicting how the system will behave in the near future and adjusting its actions to stay on course.
“The control algorithm is the biggest innovation in this paper,” Mandralis says. “Quadrotors use particular controllers because of how their thrusters are placed and how they fly. Here we introduce a dynamic system that hasn’t been studied before. As soon as the robot starts morphing, you get different dynamic couplings — different forces interacting with one another. And the control system has to be able to respond quickly to all of that.” More

Open heart surgery is a hard thing to practice in the real world, and airplane pilots cannot learn from their mistakes midair. These are some scenarios where virtual reality solves really hard problems, but the technology has limits. That’s the upshot of a review of experimental research on VR, published in the journal Nature Human Behavior.
“Virtual reality is not for everything,” said Jeremy Bailenson, lead author and director of Stanford’s Virtual Human Interaction Lab. “What we’ve long showed in the lab is that VR is great when used sparingly and thoughtfully. Otherwise, the pros typically don’t outweigh the cons.”
As a medium, VR is very intense, added Bailenson, who is the Thomas More Storke Professor and professor of communication in the School of Humanities and Sciences. It blocks out the real world. VR can make users uncomfortable or even experience “simulator sickness,” a type of motion sickness.
While companies like Meta and Apple have invested heavily in the tech, betting on wide consumer adoption, the review’s findings show it’s better used in short doses — minutes not hours — and only for certain things. The researchers recommend saving VR for “DICE” experiences, those which if done in the real world would be dangerous, impossible, counterproductive, or expensive.
Some examples of what to do (and not do) in VR, from the paper’s five core findings:
1. Travel to awesome or personally challenging places. Don’t go to VR for run-of-the-mill meetings.
Stroll through the ruins of Pompeii or visit the Grand Canyon. The visceral nature of VR lends itself well to experiences where “being there” matters. In fact, some psychologists are using VR in exposure therapy, allowing people to face something they fear while they are physically safe. One study found that people treated for fear of flying with VR had no return of their symptoms three years later.

VR’s value fades if the environment is not dramatic. While there were hopes during the pandemic that people would turn to VR for more engaging meetings, that idea failed to catch on.
“If you are just sitting there staring and not moving your body, you probably can do that on a computer and save yourself some headset time,” Bailenson said.
2. Learn surgery or public speaking. Don’t solve basic math problems.
Educators had great hopes for virtual reality ever since simulators were first used to train pilots in 1929. As the technology developed, though, it became apparent that VR did not add much to abstract learning that can be taught well on a chalkboard.
Instead, virtual reality is best used with learning skills that are procedural, requiring one step and then another, as might be done in surgery or dissection situations. Spatial tasks where movement and immersion are helpful also work well in VR, such as practicing nonverbal behavior or performing in front of a crowd.
“The key with VR is to focus on learning scenarios that are jaw-droppingly special in that medium, as opposed to assuming that any media experience works better in a headset,” Bailenson said.

3. Try on a new identity in VR, but make sure it’s the right fit.
Self-perception changes how people behave, studies have shown, whether in a virtual world or the real one. For instance, if people choose more athletic avatars, they tend to move around more. Those with taller avatars tend to negotiate more aggressively. The opposite is also true, which means users should be careful when choosing an avatar, Bailenson advised.
“Understand that whatever avatar you’re going to use is going to change the way that you behave inside VR and for some time after you leave,” he said. “So be thoughtful and use platforms that allow you to choose an avatar that either matches your actual or ideal self.”
4. Take a VR fitness class. Don’t try to learn how to throw a baseball.
Athletic training is a great use for virtual reality, except when it comes to precision movements. Users have a hard time judging distance in the virtual world, the review found. It’s a persistent problem the technology has yet to overcome.
“High-level spatial activities are great in VR, but when you’re looking for down-to-the-centimeter accuracy, you should be wary of using commercial VR applications,” Bailenson said.
5. You can run in VR, but you can’t hide.
People are easily identified by how they move their bodies, research has shown, so users should be aware that even if their avatar itself masks their identity, the millions of movement data points automatically collected by the system can identify them. There is no true anonymity when wearing VR headsets.
“In VR you move your body, and the scene responds. That’s what makes the medium so special,” Bailenson said. “Natural body movements are so important that the medium literally can’t run if you turn off movement tracking.” More

Large language models (LLMs) — the advanced AI behind tools like ChatGPT — are increasingly integrated into daily life, assisting with tasks such as writing emails, answering questions, and even supporting healthcare decisions. But can these models collaborate with others in the same way humans do? Can they understand social situations, make compromises, or establish trust? A new study from researchers at Helmholtz Munich, the Max Planck Institute for Biological Cybernetics, and the University of Tübingen, reveals that while today’s AI is smart, it still has much to learn about social intelligence.
Playing Games to Understand AI Behavior
To find out how LLMs behave in social situations, researchers applied behavioral game theory — a method typically used to study how people cooperate, compete, and make decisions. The team had various AI models, including GPT-4, engage in a series of games designed to simulate social interactions and assess key factors such as fairness, trust, and cooperation.
The researchers discovered that GPT-4 excelled in games demanding logical reasoning — particularly when prioritizing its own interests. However, it struggled with tasks that required teamwork and coordination, often falling short in those areas.
“In some cases, the AI seemed almost too rational for its own good,” said Dr. Eric Schulz, lead author of the study. “It could spot a threat or a selfish move instantly and respond with retaliation, but it struggled to see the bigger picture of trust, cooperation, and compromise.”
Teaching AI to Think Socially
To encourage more socially aware behavior, the researchers implemented a straightforward approach: they prompted the AI to consider the other player’s perspective before making its own decision. This technique, called Social Chain-of-Thought (SCoT), resulted in significant improvements. With SCoT, the AI became more cooperative, more adaptable, and more effective at achieving mutually beneficial outcomes — even when interacting with real human players.

“Once we nudged the model to reason socially, it started acting in ways that felt much more human,” said Elif Akata, first author of the study. “And interestingly, human participants often couldn’t tell they were playing with an AI.”
Applications in Health and Patient Care
The implications of this study reach well beyond game theory. The findings lay the groundwork for developing more human-centered AI systems, particularly in healthcare settings where social cognition is essential. In areas like mental health, chronic disease management, and elderly care, effective support depends not only on accuracy and information delivery but also on the AI’s ability to build trust, interpret social cues, and foster cooperation. By modeling and refining these social dynamics, the study paves the way for more socially intelligent AI, with significant implications for health research and human-AI interaction.
“An AI that can encourage a patient to stay on their medication, support someone through anxiety, or guide a conversation about difficult choices,” said Elif Akata. “That’s where this kind of research is headed.” More