A one-size-fits-all approach likely isn’t the best strategy when designing artificial intelligence systems that assist users in disease diagnosis.

A new study by researchers at MIT and elsewhere found that, while AI assistance generally improved the accuracy of non-experts and clinicians in diagnosing skin diseases, AI explainability methods had different impacts depending on the users’ knowledge level. 

Explainable AI methods help users know when to trust a model’s predictions by describing or validating the model’s decision-making. For instance, a model might use a heat map to highlight image regions that were most important in its diagnosis or a large language model (LLM) to explain the prediction in plain language.

In this study, researchers tested non-experts and primary care providers in skin disease diagnosis, with and without the help of different explainable AI systems. 

They found that non-experts’ diagnostic accuracy improved, but it was largely due to deference to the AI system. Non-experts trusted LLM-based explanations whether they were right or wrong, and found explanations more convincing when they were vague or generic.

By contrast, clinicians were not tripped up by incorrect AI assistance and performed best when given only a model’s prediction, with no accompanying explanation. 

“Good AI systems can improve performance in some health settings, but this has to be balanced carefully with algorithmic deference that can lead to more error. We know that both AI and explainability methods can engage automation bias in humans, and this anchoring effect is something that must be accounted for when we design AI systems,” says Marzyeh Ghassemi, an associate professor in MIT’s Department of Electrical Engineering and Computer Science (EECS), a member of the Institute for Medical Engineering and Science, and a principal investigator at the Laboratory for Information and Decision Systems and the Abdul Latif Jameel Clinic for Machine Learning in Health.

“These findings are important as patients increasingly turn to AI to help with their health care. Our findings show that those with the least medical knowledge are most likely to be led astray when explainable AI models give an erroneous output,” says Roxana Daneshjou, a co-author and assistant professor of biomedical data science and dermatology at Stanford University.

These results underscore the importance of building AI systems with users in mind and of developing explainability methods that encourage critical thinking rather than overreliance on the model, the researchers say.

“It’s getting obvious that we cannot just assume a good AI will solve all problems. We need to pay careful attention to the users who will be using the AI system, because the same explanation can help an expert and mislead a beginner. Often the people who could benefit most from AI are the ones most likely to be led astray by it, so how we present a recommendation matters as much as whether it’s correct,” says lead author Orson Xu, an assistant professor in the Department of Biomedical Informatics at Columbia University.

Ghassemi, Xu, and Daneshjou are joined on the paper by many authors, including MIT graduate student Haoran Zhang, undergraduate Reina Wang, and Luis Soenksen PhD ’20, a research affiliate at the Jameel Clinic, along with clinicians and researchers. A description of the work appears today in Nature Medicine.

Exploring explanations

Several FDA-approved AI interfaces are being used to help clinicians identify skin conditions in medical images, as a way to streamline early diagnosis. In addition to providing a prediction of whether disease is present in the image, these tools often use one of several methods that explain the model’s decision-making.

At the same time, non-experts can perform digital diagnosis on their own using AI-powered search engines that predict skin diseases based on user prompts. These systems often use LLMs to explain the model’s prediction in simpler terms.

The researchers explored the effects and potential benefits of these explainable AI tools on primary care physicians and non-experts in dermatological disease detection. They tested users by showing them medical images plus an AI prediction of skin disease, employing different explainable AI approaches. 

These approaches included: an AI prediction and confidence level with no explanation, a method that provides similar images to reinforce its prediction, a heat map-based approach that highlights important image regions, and an LLM that explains the model’s reasoning in plain language.

Non-experts were tasked with deciding whether an image of a skin mole was cancerous, with and without the help of explainable AI. Clinicians were given the more challenging task of providing a differential diagnosis of dermatological disease.

The researchers found that all explainable AI approaches improved the accuracy of non-experts, mostly because the tools helped users diagnose non-cancerous moles. 

In addition, when they employed a fairness-constrained model designed to combat bias against darker skin tones, the system significantly improved accuracy and reduced diagnostic disparities based on skin tone.

“But the reason non-expert users are better is because they are more reliant on the models. When the model is wrong, it hurts performance more than it helps performance when the model is right. We were just able to train very good AI models for this setting,” Ghassemi says.

This deference effect is largest with LLM explanations, and users were more confident about their wrong answers when aided by an LLM.

On the other hand, clinicians were resilient to incorrect AI explanations and, of all the explainability methods, LLMs boost their accuracy the least.

“It really comes down to how each group uses the explanation. A clinician already has a diagnosis in mind and checks the AI against their own training, so a bad explanation gets caught. Meanwhile, a non-expert can use that exact same explanation to form an opinion in the first place, so a plausible, confident-sounding rationale can pull them toward the wrong answer. The same tool ends up being an asset for one user and a liability for another,” Xu says.

Overcoming the deference effect

When the researchers dug deeper, they found that users who were most deferential to AI assistance were the worst performers on the task without the help of AI. 

They also found that the time at which users were presented with AI explanations influenced their behavior. If an explanation is given first, before the user can perform the diagnosis on their own, they tend to become more deferential to the model.

In addition, AI systems outperformed humans when the presentation of disease was subtle, but humans performed much better if there are atypical symptoms or unrelated features in an image.

Taken together, these results indicate that explainable AI can cause overreliance on models and lead users to blindly follow AI recommendations even when they are wrong. 

Rather than using LLMs to generate more detailed explanations, it might be more effective to force users to give a diagnostic hypothesis first, then provide an AI-based suggestion to highlight other possible conditions for consideration. 

“We really want AI to improve creativity and either upskill or fill in gaps where users are missing subtle presentations. Otherwise, we risk engaging automation bias and then, when the model is wrong, users can’t recover,” Ghassemi says. 

This research was funded, in part, by the National Science Foundation, Schmidt Sciences, the National Bureau of Economic Research, and Columbia University.

Alexander “Sasha” Rakhlin PhD ’06, the Distinguished Professor in Data, Systems, and Society at the MIT Institute for Data, Systems, and Society (IDSS); and a professor of brain and cognitive sciences at MIT, has been named the next director of the MIT Statistics and Data Science Center (SDSC). 

Rakhlin succeeds Ankur Moitra, the Norbert Wiener Professor of Mathematics, associate director of the IDSS, and a faculty member in the MIT Department of Electrical Engineering and Computer Science (EECS) who has been SDSC director since 2021. Philippe Rigollet, the Cecil and Ida Green Distinguished Professor of Mathematics and a core faculty member in IDSS, also served as interim director in 2024-25.

“Sasha is one of the sharpest theoretical minds working in statistics and machine learning today, and also one of the most devoted mentors I know,” says Fotini Christia, the Ford International Professor of the Social Sciences and director of IDSS, which houses SDSC. “He has helped train an entire generation of interdisciplinary scholars through the Interdisciplinary Doctoral Program in Statistics (IDPS), while his own research keeps pushing the boundaries. The SDSC could not ask for a more fitting leader.”

Rakhlin is the inaugural holder of the Distinguished Professorship in Data, Systems, and Society, an endowed chair created in 2025 by the generosity and vision of IDSS professor Richard “Dick” Larson, an “MIT lifer” and pioneer in operations research, queueing theory, and system optimization.

“I am honored to take on this role,” says Rakhlin. “The strength of the Statistics and Data Science Center has always been its people — students, postdocs, and faculty from across MIT who bring sharply different perspectives to the most interesting problems of the day in statistics, machine learning, and AI. My goal is to support that community as it takes on the constantly evolving questions reshaping the field.”

Rakhlin has been connected to the Statistics and Data Science Center as a visiting professor since 2016, before formally joining MIT in 2018 in the Department of Brain and Cognitive Sciences and IDSS. As the initial chair of the Interdisciplinary PhD in Statistics program at the SDSC, Rakhlin has seen the successful defense of over 75 IDPS PhD students across a variety of departments at MIT, including IDSS’ own Social and Engineering Systems program.

“I have been fascinated by machine learning since my PhD work more than 20 years ago, drawn by its beautiful connections to statistics, probability, algorithms, optimization, and game theory,” says Rakhlin. “At the Statistics and Data Science Center, I work alongside colleagues who share this fascination and pursue these connections in many directions. The recent revolution in AI is extending this web into the sciences; it promises to accelerate discovery, and it raises new questions for statistics. Answering them demands a rigorous science of the tools themselves. As AI enters medicine, energy, and public life, its safety and security are, at their core, statistical and mathematical questions: quantifying uncertainty, providing guarantees, understanding failure, and resisting manipulation.”

As Rakhlin puts it, the SDSC is built for this moment. “Statistics is a shared language across MIT,” he adds. “Through the Interdisciplinary Doctoral Program in Statistics, the center connects students and faculty from economics and political science to physics and engineering. Collaborations in areas from biology to nuclear fusion have shown how statistical thinking accelerates science itself.” 

As director, one of his goals is to deepen these interdisciplinary connections. He hopes to help make SDSC the Institute’s home for the rigorous foundations of data science and AI, and a bridge to the scientific and societal questions where those foundations are most needed.

Rakhlin received his bachelor’s degrees in mathematics and computer science from Cornell University, and doctoral degree from MIT. He was a postdoc at the University of California at Berkeley in EECS before joining the University of Pennsylvania, where he was an associate professor in the Department of Statistics and co-director of the Penn Research in Machine Learning center.

The 2026 Northeast Microelectronics Internship Program (NMIP), organized by the MIT Microsystems Technology Laboratories, brought together 30 exceptional students from leading universities across the Northeast for an immersive week exploring the rapidly evolving world of semiconductor technology and microelectronics. Held July 13-17, the externship provided undergraduate students with an opportunity to experience the complete microelectronics innovation ecosystem, from academic research laboratories to advanced manufacturing facilities.

Throughout the week, students visited several of the region's premier institutions, including MIT.nano, IBM Research, GlobalFoundries, Rensselaer Polytechnic Institute (RPI), and NY CREATES, where they engaged with researchers, engineers, faculty, graduate students, and industry leaders working at the forefront of semiconductor innovation.

The program began at MIT.nano with an inspiring overview of the microelectronics landscape led by Vladimir Bulović, director of MIT.nano, and Farhad Varzhegoo, director of strategic initiatives and partnerships at the Northeast Microelectronics Coalition Hub. Their presentations challenged students to think beyond today's technologies and consider the broader societal impact of tomorrow's innovations.

"What will the next innovation in microelectronics look like, and what should the world of tomorrow focus on?" they asked, encouraging participants to view engineering not only as a technical discipline, but also as a means to solve meaningful real-world challenges.

Following the opening session, Farnaz Niroui, the Emmanuel E. Landsman Career Development Chair and assistant professor of electrical engineering and computer science at MIT, organized a series of graduate student research presentations showcasing the breadth of microelectronics research taking place across MIT. The presentations explored topics spanning integrated circuits, nanoelectronics, photonics, quantum technologies, and advanced materials.

After the student research presentations, participants attended an industry panel exploring the transition from academia to careers in microelectronics. Organized and moderated by Susan Feindt, fellow emeritus at Analog Devices and visiting research scientist at MIT, the panel featured professionals from Rage Systems, Cadence Design Systems, Analog Devices, and RTX (Raytheon), who shared their career journeys, discussed the differences between research and industry, and offered advice on navigating career opportunities in the semiconductor sector.

"What stood out to me most about our day at MIT was the opportunity to engage deeply with PhD students in this field and understand the kind of opportunities available by pursuing a doctoral program," says Shanti Visurakapalli, a current undergraduate student at MIT. "I think this experience, complemented with the industry panel, gave many of us in the program the perspective we needed to weigh future graduate and professional options."

Throughout the week, participants connected classroom concepts with real-world applications through behind-the-scenes access to some of the nation's most advanced research and manufacturing environments. Students explored MIT's interdisciplinary laboratories, observed High-NA EUV lithography and quantum hardware development at IBM Research, toured GlobalFoundries' state-of-the-art 300mm semiconductor fabrication facility, learned how groundbreaking academic research transitions into commercial manufacturing at RPI, and gained insight into next-generation semiconductor fabrication at NY CREATES.

"The externship gave me a behind-the-scenes look at the advanced technologies driving the microelectronics industry while allowing me to connect one-on-one with researchers and industry professionals," says Sean Kim, a student at Princeton University. "Learning about emerging research and receiving career advice broadened my perspective on the field and inspired me to pursue a career in microelectronics."

Beyond the technical experiences, the externship emphasized professional development and networking. Students engaged in meaningful conversations with engineers, scientists, faculty members, and graduate researchers who described their career paths, offered advice, and discussed the many pathways available within the semiconductor industry. These interactions provided participants with valuable perspectives on careers in research, manufacturing, design, and emerging technologies.

"One of the most rewarding aspects of the externship is seeing students from different universities come together around a shared passion for innovation," says Preetha Kingsview, NMIP program administrator. "The friendships they build, the conversations they have with researchers and industry leaders, and the excitement they bring to every visit create an experience that extends far beyond the technical program."

For many students, the experience proved both transformative and inspiring. The opportunity to witness cutting-edge research firsthand while building connections with leaders across academia and industry deepened their understanding of the semiconductor ecosystem and reinforced the critical role microelectronics plays in addressing global challenges.

"For more than half a century, microelectronics has transformed the world, but I believe its most exciting chapter is only just beginning," says Tomás Palacios, the Clarence J. LeBel Professor of Electrical Engineering and Computer Science at MIT and faculty director of the NMIP Program. "From AI and quantum computing to sustainable energy and advanced manufacturing, nearly every technological revolution of the coming decades will be built on advances in semiconductor technology. Today's undergraduate students will become tomorrow's innovators, entrepreneurs, and industry leaders, and programs like the NMIP Externship help inspire and prepare them to shape that future."

By bringing together leading universities, research institutions, and industry partners, the program provides students with a comprehensive view of the semiconductor ecosystem while helping build the highly skilled workforce needed to sustain U.S. leadership in microelectronics.

The 2026 externship demonstrated the power of connecting education, research, and industry. Through a week of laboratory tours and technical presentations, it gave students a firsthand view of how scientific discovery becomes technological innovation — and inspired many to become part of the future of microelectronics themselves.

The NMIP Externship was made possible by the Microelectronics Commons Northeast Microelectronics Coalition Hub and the Microelectronics Commons Northeast Regional Defense Technology Hub (NordTech). Additional support was provided by the MIT Microsystems Technology Laboratories, the MIT Institute for Soldier Nanotechnologies, and the Semiconductor University Research Program for Superior Energy-Efficient Materials and Devices (SUPREME) Center, part of the SRC JUMP 2.0 program.

Molecules are among the smallest building blocks available for making next-generation devices. Their unique, customizable properties enable promising applications in emerging computing, sensing, optical, and quantum technologies.

But integrating molecules into functional devices at scale remains a challenge. Traditional semiconductor manufacturing processes can damage small and fragile molecular materials. Now, MIT researchers have developed a scalable fabrication technique that incorporates delicate molecular materials into electronic devices on a chip without causing damage.

Their method extends the capabilities of standard semiconductor manufacturing processes to accommodate molecules. The researchers first prefabricate the device components using traditional processes. Then, they introduce the molecules and harness nanoscale surface forces to mechanically transform the fabricated device, which self-assembles without damaging the molecules. 

The team demonstrated the robustness and scalability of their technique by fabricating more than 1,000 devices using sub-nanometer molecular layers. 

“Our platform combines the scalability of conventional semiconductor manufacturing with the precision and control of self-assembly. This establishes a new fabrication framework for the scalable, high-throughput integration of emerging nanoscale and quantum materials, including molecules, into functional devices with architectures and capabilities that were previously infeasible,” says Farnaz Niroui, an associate professor of electrical engineering and computer science (EECS), a member of the Research Laboratory of Electronics (RLE), and senior author of a new paper describing the work.

She is joined on the paper by co-lead authors Sarah Spector and Peter Satterthwaite, EECS graduate students; Jeremiah A. Johnson, the A. Thomas Guertin Professor of Chemistry at MIT; and others at MIT. The research appears today in Nature Nanotechnology.

Building with molecules

Molecules are small clusters of atoms with structures and chemistries that can be precisely designed. This allows their properties to be engineered across a wide design space.

Once integrated into device architectures, these molecules could enable next-generation electronics and computing platforms that are smaller, faster, and more adaptable, as well as higher-performance photonic devices and emerging quantum technologies.

To build a functional system, molecular building blocks need to be integrated with other device layers. In electronic systems, a critical step is making electrical contacts to the molecules by interfacing them with metallic surfaces. However, the harsh chemicals and processes needed for traditional chip manufacturing damages these fragile molecular materials, reducing reliability and performance.

To leverage the scalability of standard fabrication techniques while achieving the precision needed for handling molecules, the MIT researchers developed a decoupled, two-step approach.

They first fabricate all the device components using standard semiconductor manufacturing, then incorporate the molecular material after-the-fact to finish building the device.

“By bringing the delicate materials into the process only after we have fabricated the main device elements, it allows us to use conventional processes that are normally not compatible with these nanomaterials,” Satterthwaite says.

In their demonstration, the researchers fabricated a scaffold with two metal electrodes separated by a precisely sized gap. Then, they deposited the molecular layer on the electrode surfaces. 

Finally, the researchers leverage nanoscale forces to gently pull the top electrode onto the molecules, forming the final device in a nondestructive way. This creates a self-aligned, damage-free electrical contact to the molecules. 

Using the forces

While gravity is a dominant physical force that holds our world together, different forces dominate at the nanoscale. One, called the capillary force, causes liquid to get sucked into small spaces. (Plants rely on capillary forces to draw water into their stems.) 

By carefully engineering the stiffness of the electrodes, when the solution containing the molecules evaporates, capillary forces gently pull the two metal surfaces together with the molecules sandwiched in between.

Once the two electrodes are in place, the researchers must hold them in a stable state. To do so, they rely on another nanoscale force known as the van der Waals force. 

Van der Waals forces cause surfaces to attract one another. By controlling the device surface area and molecules properties, the researchers ensure these forces will be strong enough to hold the electrodes in a stable structure without damaging the molecules. 

“Nanoscale forces play a critical role in our approach. Instead of fabricating exactly the structures we ultimately want, we make something mechanically mobile and use forces to transform it into an architecture that would otherwise be impossible to fabricate,” Spector explains.

They used this technique to fabricate more than 1,000 devices with molecular layers less than 1 nanometer thick. Even at this tiny scale, the fabricated chips comprised a high yield of working devices, 96 percent on average. The robust devices also endured tens of thousands of electrical cycles without showing any sign of degradation.

“The stability really stands out. This is a critical feature for moving molecular devices toward practical applications, but it has been a persistent challenge in the field,” Satterthwaite says. 

Importantly, this versatile technique allows circuit- and system-level integration of molecular devices, pushing the field beyond the study of isolated devices, the researchers say. They demonstrated this by building an interconnected array of molecular memory devices which could have applications in next-generation computing platforms. 

Their technique can also be extended to other materials and device architectures. 

In the future, the researchers want to build on this platform to investigate and develop new classes of multifunctional computing and sensing devices and systems. 

“By enabling the pristine integration of emerging molecular materials and other atomic-scale matter into functional devices at scale, our platform accelerates discovery and design of these materials with tailored functionalities and their deployment in emerging technologies,” Niroui adds.

This research was funded, in part, by the U.S. Defense Advanced Research Projects Agency (DARPA), the Semiconductor Research Corporation, the U.S. National Science Foundation (NSF), the MathWorks Fellowship, and the Netherlands Organization for Scientific Research. Device fabrication was carried out, in part, using MIT.nano facilities.

You probably think you are rational. Day to day, you try to “make the best possible use of the information available to you,” as MIT philosopher Brian Hedden PhD ’12 writes in his 2015 book, “Reasons without Persons.” 

If you think you are rational because of how you plan for the future, and change your beliefs over time, however, Hedden will be skeptical. Circumstances change, and when using reason, he thinks, all that matters is the present. Thus, he also writes: “The requirements of rationality should be impersonal, avoiding reference to the relation of personal identity over time.”

This is what Hedden calls “time-slice rationality,” the idea that in applying reason, we exist in little slivers of time and knowledge. There are not permanently reasonable people, just reasonable decisions. 

“We are temporally extended people with hopefully long lifespans, but we’re made up of lots of different time slices,” Hedden says. “So there’s me now, me yesterday, me next year. We should think of the locus of rationality as the time slice, not the temporally extended person.” 

Those past versions of you, Hedden thinks, are like teammates in sports: You are connected to them, but not quite the same person. The payoff from viewing things this way, he contends, is a more streamlined and realistic picture of our thinking. 

“Time-slice rationality” helped launch Hedden’s career. Today he is a professor in MIT’s Department of Linguistics and Philosophy and associate dean for MIT’s Social and Ethical Responsibilities of Computing (SERC) initiative. In a nod to his work, let’s examine some time slices from Hedden’s career.

Falling for philosophy

Hedden grew up in Virginia and attended Princeton University as an undergraduate, where he expected to study electrical engineering. But early on in college, he took some philosophy classes — an introductory course in logic, a history of early modern philosophy — and liked them. A lot. 

“Engineering is still near and dear to my heart, but I really got into philosophy,” Hedden says. 

Almost before he knew it, Hedden had found his primary intellectual interest. Upon graduating from Princeton, Hedden spent a year working for an education nonprofit in Nicaragua, but he had already decided on his next stop: graduate school in philosophy.

That led Hedden to MIT. He wanted to study the philosophy of language, and at MIT, the famous linguistics program is part of the same department as philosophy. Hedden applied to the Institute, was accepted, and arrived on campus eager to pursue his chosen field. 

Lounge life

A funny thing happened to Hedden after he arrived at MIT, however: He stopped studying the philosophy of language. Blame Frank Gehry, the architect. 

MIT’s Department of Linguistics and Philosophy is in the Stata Center, which opened in 2004 and was designed by Gehry to have all kinds of common spaces, including double-height lounges. After Hedden started graduate school at MIT, he got drawn into the philosophical discussions happening in these common areas. Before long, he had changed his intellectual focus. 

“This was largely due to conversations that were happening in the lounge,” Hedden recalls. “The Stata Center has spaces that really encourage collaboration. A lot of people there were talking about puzzles involving probability, epistemology, and decision theory, and I found those things really stimulating, and wound up specializing in those areas. Things wouldn’t be the same if the building were differently arranged.”

Advised by MIT professors Caspar Hare, Robert Stalnaker, and Roger White, Hedden wound up writing his doctoral thesis — three papers — on rationality and decision-making. That formed the basis of “Reasons without Persons,” whose title alludes to a famous work by philosopher Derek Parfit.

While it might seem unusual to draw to draw a distinction between our past, present, and future selves, Hedden thinks we actually do that frequently, in everyday life and cultural work. In Greek mythology, Odysseus rationally chains himself to the mast of his ship, anticipating that his future self will be irrationally unable to resist the call of the sirens.

“That’s a dramatic example, but we do this all the time, like when we buy a gym membership and hope that our future selves will irrationally care about sunk costs and go to the gym to avoid having wasted the money,” Hedden says. 

Heading down under

After earning his MIT PhD, Hedden spent two years as a junior research fellow at Oxford University, then landed his first faculty job in academia — at the University of Sydney, in Australia, in 2015. Five years later, he moved to Australian National University, in Canberra, leaving when he returned to MIT in 2025. 

“I love Australia; I think the quality of life is amazing, the culture is great, the natural world is unbelievable,” Hedden says. The country also has, he observes, “a great philosophy scene,” fed in part by decades of interaction with American scholars. 

Hedden’s work kept evolving during his decade in Australia. He started examining specific, applied topics more often, including many questions about legal processes and evidence. For instance: Should juries even deliberate? In one 2017 paper, Hedden suggested they should not, because, among other things, “deliberation destroys the independence of jurors’ judgments” in ways that can be counterproductive. 

Or: Is there such a thing as “higher-order” evidence, which is evidence about what conclusions your evidence supports? In a 2021 paper, Hedden and now-MIT colleague Kevin Dorst concluded that virtually all evidence fits this billing. 

Hindsight bias: Not bias

Or take another Hedden paper in this vein, from 2019, casting new light on the familiar topic of “hindsight bias.” We often alter our views about things after they happen, which can seem like gratuitous second-guessing. 

Is it, though? Suppose you are investigating a railroad crash and find evidence that a crash was more likely than people imagined. That might simply be useful new knowledge. Suppose your favorite basketball team loses a game, and you reexamine why you thought they would win; perhaps a star player’s injury was more serious than you imagined. Are you changing your basic views, or just conducting a realistic reassessment?

“This is perfectly rational and what we should expect,” Hedden says. “Some people say, ‘Oh, that’s hindsight bias.’ I think it’s just a reasonable conclusion to draw.”

To be sure, in the paper itself, Hedden engages with theoretical philosophical work about evidence and view formation; much of his work bridges academic theory and practical everyday applications. Like many of his papers, this one also evinces the fun of reworking conventional wisdom. 

“I do think that these contrarian views are right,” Hedden says. “But I also find a certain joy in going my own way, or having a skeptical take to get people to rethink views they’re falling into without fully realizing it.” 

After an odyssey, back at MIT

After nearly a decade in Australia, Hedden received an offer to return to one of his intellectual homes: MIT offered him a place on the faculty. Arriving back at the Institute in 2025, Hedden found many things had changed — new buildings on campus, new programs — while some were recognizably the same. 

“The philosophy department looks very similar in terms of the healthy culture,” Hedden says. “It’s always been known as a collaborative, high-energy place with a fantastic graduate program. And it’s really welcoming. That lounge discussion culture is still there. Sometimes cultures can be fragile. There could have been people who let it lapse. But it’s still there, and that’s great.”

Meanwhile, Hedden has added to his intellectual portfolio by becoming associate dean at SERC, a burgeoning initiative at MIT examining a wide range of civic issues around computing. SERC has supported 40 postdocs around MIT since 2022, in all five MIT schools plus the MIT Schwarzman College of Computing. It has also awarded 30 seed grants for faculty research in the last three years.

“There’s been really broad interest from faculty and students,” Hedden says. “I’m interacting a lot with computer scientists especially, but people across the Institute everywhere. The College of Computing is a unifying force.” 

For that matter, the SERC Scholars program had 75 students accepted last fall, from first-year undergraduates to doctoral candidates, working on projects including surveillance, artificial intelligence, the energy impact of the sector, and more. Hedden is also making a point to develop more courses across SERC topics.

“It’s often the younger people, undergraduates and graduate students, the postdocs, the junior faculty, that gives us a constant infusion of new ideas and energy,” Hedden says. “We’re hoping to keep the momentum going.”

In this slice of time, that sounds pretty reasonable. 

For Hugh Smith, the challenge of building an energy-secure future isn’t about creating the world’s “best” battery. It’s about designing the right battery for the right job.

As a fifth-year PhD candidate in MIT’s Department of Materials Science and Engineering, Smith studies sodium-ion batteries, an emerging alternative to the lithium-ion batteries that power everything from smartphones to electric vehicles. By replacing expensive critical minerals like lithium, nickel, and cobalt with more readily available elements like sodium, iron, and manganese, his research aims to make energy storage both more affordable and more sustainable.

“I’ve believed for a very long time that the biggest engineering problem humanity faces is the transition to clean energy,” Smith says. “Batteries are a critical bottleneck in that transition.”

Growing up in Albany, New York, Smith was drawn to materials science because it combined two of his favorite subjects: chemistry and math. What kept him interested, however, was the field’s ability to touch nearly every aspect of everyday life.

“Anytime you interact with a solid material, there are people who intentionally designed that material for a specific purpose,” he says.

That idea of designing materials with a real-world purpose eventually led him to batteries. After earning his undergraduate degree in materials science from Case Western Reserve University, Smith came to MIT to explore how new battery chemistries could reduce costs without sacrificing performance.

Consumers often want batteries that charge quickly, last for years, store large amounts of energy, and remain inexpensive. But in reality, improving one characteristic of this technology usually means compromising another. A smartphone battery, for example, prioritizes energy density and long lifespan, while a battery storing electricity for the power grid doesn’t need to be lightweight or compact. Instead, cost and reliability become the most important considerations.

Rather than chasing an all-encompassing solution, Smith focuses on finding the right balance for specific applications, often juggling competing priorities. Instead of strengthening a singular characteristic, Smith works to maximize as many components of the battery as possible, including cost, performance, sustainability, and reliability, depending on how it will be used.

“It’s trying to balance everything,” he says. “It’s not catering extremely to some properties and then abandoning others.” 

The sodium-ion batteries Smith studies could eventually provide lower-cost options for electrical grids or more affordable electric vehicles. Because sodium-ion batteries can largely be manufactured using the same infrastructure already developed for lithium-ion batteries, they also offer a potentially smoother path toward commercialization than many emerging battery technologies.

Smith’s graduate school journey has been defined as much by the process of learning how to do research as by the science itself. He joined a brand-new research group at MIT as its first graduate student, and helped establish the lab run by Professor Iwnetim Abate. Without senior graduate students or postdocs to turn to for day-to-day guidance, he often had to teach himself new techniques and how to troubleshoot when things went wrong.

“I learned not to be fearful of new things,” Smith says. “Just because I didn’t know how to do something didn’t mean I couldn’t figure it out.”

He says the experience transformed him into a more independent researcher and someone who is willing to dive headfirst into unfamiliar problems.

Before beginning graduate school, Smith spent seven months at the Battery Innovation Center in Newberry, Indiana, an experience that broadened his understanding of how scientific discoveries become real technologies. Working alongside materials scientists, chemists, mechanical engineers, and chemical engineers showed him that no single discipline can solve the challenges of battery development alone.

“It requires a huge team effort,” Smith says. “It requires a lot of different types of knowledge.”

He says the experience also helped him better understand where his own expertise could make the greatest impact and when collaboration across disciplines is essential.

Outside the lab, Smith makes time to stay active through MIT’s intramural sports program, where he plays soccer, ultimate frisbee, football, and volleyball on teams with fellow graduate students. The games offer a chance to unwind after long days of research while strengthening the friendships he’s built throughout graduate school. He also enjoys fishing around the Boston area with friends and exploring New England’s coastal towns, museums, and historic sites.

As he prepares to graduate in the winter and pursue a career in battery research and development, Smith hopes to continue designing technologies that support the transition to clean energy. 

“Lots of smart people have already made wind and solar very cheap,” Smith says. “The issue is reliability, and batteries can help solve that problem. I hope the work I’m doing helps to affordably unlock the transition to an electric grid powered by reliable clean energy, and an electrified transportation network.”

Daniela Rus, director of MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) and the Panasonic Professor of Computer Science, has received the 2026 High-Tech Prize of the Bavarian Minister-President for her contributions to robotics, artificial intelligence, and autonomous systems. 

Awarded jointly by the Bavarian State Government and the Bavarian Academy of Sciences and Humanities, it is the most highly endowed award for technology and engineering in Germany. Rus accepted the prize on July 23 at the Herkulessaal of the Munich Residence.

The selection committee cited four strands of her work: self-organizing robot collectives, soft robotics, autonomous mobility, and brain-inspired artificial intelligence. Together they describe a 30-year effort to build machines that hold up outside the lab, in conditions no one scripted in advance.

That effort has arrived at a moment when physical AI has become a preoccupation for industry leaders and policymakers alike. When human-robot collaboration comes up, the examples tend to be household chores or the factory floor. Rus is working several orders of magnitude wider than that, developing algorithms and systems that put autonomous robots into transportation, agriculture, medicine, the home, and environmental monitoring. She is also a pioneer of soft robotics, where compliant machines manipulate the world more safely and adapt to it more readily than rigid ones can.

Her emphasis throughout has been on giving robots the intelligence to reason and adapt in the real world, through algorithms whose behavior can be explained.

"AI gives machines the ability to do work that humans don't want to do," she says. "It's not a battle between humans and machines. Both form a system that solves problems that neither humans nor machines can solve alone."

At CSAIL she leads the Distributed Robotics Laboratory, where that principle has produced some unusual solutions to durable problems. Her group helped build an ingestible origami robot capable of retrieving swallowed button batteries from a child's digestive tract, and a fleet of small autonomous boats that assemble themselves into bridges and platforms, turning a city's waterways into infrastructure that can be reconfigured on demand.

She also helped invent liquid neural networks, an architecture inspired by the compact nervous system of a millimeter-long worm. The networks can steer a vehicle through an unfamiliar environment using as few as 19 control neurons, a level of efficiency that conventional architectures cannot approach. The research led Rus and former CSAIL affiliates Ramin Hasani, Alexander Amini, and Mathias Lechner to found Liquid AI out of MIT CSAIL, building models designed from the start for the hardware constraints of the devices they run on.

"Daniela Rus is a pioneer in soft robotics and physical AI," noted Lorenzo Masia, professor of intelligent bio-robotic systems at the Technical University of Munich, in a press release. "The prize will help to bring this science to the forefront."

Rus' previous honors include the 2025 IEEE Edison Medal and the 2024 John Scott Award. She is a member of the 2002 class of MacArthur Fellows and has been elected to the French National Academy of Medicine, the National Academy of Engineering, and the American Academy of Arts and Sciences. She is a fellow of the Association for Computing Machinery, the Institute of Electrical and Electronics Engineers, and the Association for the Advancement of Artificial Intelligence.

This spring, 25 MIT students and postdocs traveled to Washington to meet with congressional staffers and advocate for sustained federal investment in scientific research. 

With recent cuts to National Science Foundation programs and continued uncertainty surrounding the federal research budget, these conversations were especially timely. Over the course of just two days, participants met with 62 congressional offices representing 32 states to discuss the importance of federal support for scientific research, higher education, and other policy concerns related to their individual research areas.

Each spring, the MIT Science Policy Initiative (SPI) organizes Congressional Visit Days (CVD), a program that introduces graduate students and postdocs to the federal policymaking process while demonstrating the many ways scientists can engage in policy advocacy. In addition to meeting with congressional offices, participants connect with Washington-based MIT alumni and members of the MIT Washington Office to learn about careers at the intersection of science and public policy.

This year's CVD was co-organized by Audrey Parker, a PhD student in civil and environmental engineering at MIT, and Ian Robertson, a PhD student in physical oceanography at MIT and the Woods Hole Oceanographic Institution (WHOI). Robertson reflects on the experience:

"Having attended the trip as a participant last year, stepping into the role of a co-leader this year was a big commitment that was well worth the payoff. It was rewarding to build on the work of past leaders, strengthening the CVD experience for participants by training and encouraging them to discuss not just general science funding advocacy in their meetings, but also specific policies tied to their research. I look forward to the future success of the CVD program in continuing to provide students and postdocs a template for science policy conversations with Congress and helping them realize the various avenues in which they can connect research to policy throughout their careers."

To prepare for the trip, participants attended three training sessions led by SPI in collaboration with the MIT Washington Office and the MIT Policy Lab. These sessions provided background on the federal appropriations process, the role of congressional staff in shaping legislation, and practical strategies for communicating scientific expertise to policymakers. The training sessions also gave participants a chance to practice sharing their research and policy pitches with each other in mock "Hill meetings."

While on the Hill, students advocated for both general science funding for the upcoming fiscal year as well as specific policies tied to their research in artificial intelligence, environmental science and engineering, energy, space, and health. They encouraged offices to edit language in bills, support bills already on the floor, or sponsor new bills. Staffers on both sides of the aisle were particularly interested in discussing AI privacy and security across disciplines. They also expressed strong interest in hot environmental topics, such as deep-sea mining, and were eager to learn more about its associated environmental consequences. In many cases, conversations about participants' research and science-based policy priorities reinforced the need for continued federal funding of science, enabling staffers to connect abstract funding decisions with the researchers and projects they support.

The experience proved valuable for both the MIT delegation and the congressional offices they visited. Rodrigo Zuniga, a first-year PhD student participant, highlights:

"Going to Washington, D.C., offered a whole new perspective of the role of science in politics for me. In today's news and social media landscape, it's really easy to see Washington as irreversibly polarized, but in talking to staffers from both the majority and minority parties, I saw a general desire for bipartisanship and widespread support for science. What's most clear to me after this experience with CVD is that there is a lot of room and pressing need for humans with scientific training and expertise to participate proactively in local, state, and federal government."

As scientific and technological issues continue to shape public policy, opportunities for researchers to engage with policymakers have never been more important. Programs like Congressional Visit Days help equip the next generation of scientists with the knowledge and confidence to communicate the value of research beyond the laboratory, strengthening connections between the scientific community and the policymakers whose decisions shape its future.

Nitrogen gas is abundant in Earth’s atmosphere, but most living organisms can’t readily use this nitrogen. Only a subset of microbes that have enzymes known as nitrogenases can break nitrogen gas apart and convert it into ammonia.

There are three different classes of nitrogenases found in nitrogen-fixing microbes, which vary based on the types of metal that they contain. Nitrogenases that contain the metal molybdenum are the most efficient, and two new studies from MIT offer an explanation for why that is.

The findings could help guide the design of engineered enzymes or synthetic catalysts that can convert nitrogen gas to ammonia, the researchers say.

The team found that while molybdenum doesn’t directly bind to nitrogen, it helps nearby iron atoms bind to nitrogen more strongly. This is a critical first step in breaking the bond between the two nitrogen atoms that form nitrogen gas.

“It’s that initial binding step that’s really the hard part. Once you’ve started to break the nitrogen-nitrogen triple bond and make some new nitrogen-hydrogen bonds, it’s pretty easy to get the rest of the way,” says Daniel Suess, the Arthur Amos Noyes Associate Professor of Chemistry at MIT and a senior author of both papers.

MIT postdoc Tong Wu and former postdoc Madeleine Ehweiner are the lead authors of one of the papers, and Alexandra Brown PhD ’23 is the lead author of the other. Kyle Lancaster, a professor of chemistry at Cornell University, is a senior author of the latter paper, along with Suess. Both papers appear today in the journal Chem.

Efficient enzymes

Before microbes evolved the ability to fix nitrogen around 3 billion years ago, the strong triple bond between atoms of N2 could only be split with high-energy events such as a lightning strike.

“Once an enzyme came along that could convert dinitrogen to ammonia, that changed the game because now cells could harvest nitrogen from the air for biomass,” Suess says.

Within the active site of nitrogenase is a catalytic cofactor that typically consists of a cluster of iron, sulfur, carbon, and in some cases another metal. Nitrogenases whose cofactors contain molybdenum are the most efficient, followed by those containing the metal vanadium. Nitrogenases that don’t have any metal other than iron are the least efficient.

Why the molybdenum-containing enzyme is more efficient has been a puzzle, especially because it’s thought that molybdenum itself doesn’t bind directly to nitrogen gas.

“In all cases, iron is thought to interact with N2, so it’s a bit of a mystery,” Suess says. “If all the chemistry is happening at iron, why is it that this molybdenum is affecting catalysis?”

To answer that question, Suess’s lab has developed simpler versions of iron-sulfur clusters that they can use to model the naturally occurring cofactors. These can be modified by adding different metal atoms, allowing the researchers to study how those metals change the cofactors’ properties. 

In the first paper, led by Wu and Ehweiner, the researchers swapped in different metal atoms and then measured the ability of the iron in the cofactor to bind to nitrogen. They found that only cofactors with a large metal atom, such as molybdenum or tungsten, were able to strongly bind N2. With vanadium,  chromium, or iron, which are smaller, the cofactors did not bind N2 and performed other reactions instead.

“That paper essentially recapitulates what you see in biology, which is that the iron-sulfur clusters that have molybdenum in them seem to be better at binding dinitrogen than those with lighter metals,” Suess says.

Sharing electrons

In the second paper, led by Brown, the researchers uncovered a possible mechanism that explains that phenomenon. 

In that paper, the researchers studied how cofactors containing different metals interact with compounds called N-heterocyclic carbenes. These molecules behave similarly to N2 in some ways, making them a good model for this type of study. Like N2, they are resistant to accepting any electrons from another molecule, which is an essential step to breaking chemical bonds. 

The researchers found that when molybdenum was included in the cluster, it became easier for iron to donate some of its electrons to the N-heterocyclic carbenes, in a process known as back-bonding. This occurs because molybdenum, a large atom, has large orbitals that can overlap with the orbitals of the nearby iron atom. That alters iron’s electron density in ways that make it easier for iron to pass electrons to N2.

“Without these direct metal-metal interactions, the iron has to do all the work, but adding the molybdenum allows for this electronic cooperativity,” Suess says.

Once N2 is bound to an iron atom, the rest of the reaction can proceed. A proton can come in from water or another source to create an N-H bond, which then makes it much easier for the remaining N-N bonds to be broken and bind to protons, forming NH3. 

The findings could help guide scientists who are working on designing enzymes that could be engineered into organisms that help them generate their own NH3, eliminating or reducing the need for fertilizer. The results could also help chemists to design synthetic catalysts that could produce ammonia industrially, using less energy than the Haber-Bosch process. 

“The general principle is that you can make an iron site in any context behave differently when you have these metal-metal interactions than when you don’t have these interactions,” Suess says. “The primary result of these findings is to teach us about the natural world and how nature accomplishes this really important and miraculous reaction. And, maybe that can be translated into new processes.”

The research was funded primarily by the U.S. Department of Energy, the National Science Foundation, and the National Institute of General Medical Sciences.

Researchers in the lab of Sam Peng, the Pfizer Inc. - Gerald Laubach Career Development Assistant Professor of Chemistry at MIT and a core institute member of the Broad Institute of MIT and Harvard, have developed a groundbreaking super-resolution imaging technology that allows scientists to visualize molecular structures with sub-angstrom-level localization precision — three orders of magnitude beyond the nanometer limits of standard fluorescent dyes — while drastically simplifying the imaging process. 

Unlike traditional dyes that fade rapidly under illumination and limit data collection, the platform, called U-STORM (Upconversion enabled Stochastic Optical Reconstruction Microscopy) utilizes a new class of compositionally engineered upconverting nanoparticles (UCNPs) that blink spontaneously and indefinitely. 

This work represents a fundamental shift in both optical materials and biological imaging. An open-access description of the study was published July 27 in Nature Nanotechnology.

Overturning a decades-old paradigm

For decades, the scientific community widely considered upconverting nanoparticles to be completely photostable and non-blinking. Because localization-based super-resolution microscopy techniques like STORM rely entirely on the stochastic “blinking” (switching between “on” and “off” states) of light emitters to distinguish closely packed molecules, UCNPs were historically deemed unsuitable for this type of imaging.

“Our laboratory has long been interested in overcoming these limitations,” says Peng. “Our work began with a question: Can we develop a super-resolution imaging platform that is simultaneously long-term, multicolor, simple to operate, and capable of achieving extremely high localization precision without using imaging buffers or additional optical control?”

By meticulously controlling nanoparticle composition, the MIT and Broad Institute team discovered that these small (~10nm) core-shell particles could actually be coaxed into spontaneous blinking under continuous near-infrared excitation. Remarkably, this blinking behavior continues indefinitely without the need for complex imaging buffers, oxygen scavengers, or external optical modulation.

U-STORM’s key breakthroughs

An angstrom is a tiny unit of measurement used by chemists to measure size and distances at the atomic level. U-STORM’s ability to blink indefinitely has afforded researchers the opportunity to collect over 88,000 localization events from the same particle, sharpening the localization precision down to an unprecedented 0.6 Å.

Unlike conventional multicolor super-resolution imaging, which requires multiple expensive lasers and meticulous optical alignment, U-STORM can operate with just one near-infared laser, which works to simultaneously excite nanoparticles emitting different colors. This results in a drastic reduction of an experiment’s complexity.

To obtain images with multiple colors, rather than capturing images sequentially over multiple rounds, U-STORM captures multiple colors simultaneously. Researchers have successfully demonstrated this by mapping epidermal growth factor receptor dimers and multimers in biological samples under physiological conditions without any specialized imaging buffers.

Broader impact

Beyond expanding the boundaries of microscopy, this research establishes an entirely new design principle for lanthanide nanomaterials. The team is already working to expand the color palette, make the particles even smaller and brighter, and deploy U-STORM to investigate complex nanoscale protein organizations and cellular signaling pathways.

Ultimately, U-STORM promises to provide laboratories worldwide with an accessible, easy-to-implement, yet incredibly powerful route toward high-precision molecular imaging.

Before the advancement of scientific data storage and collaboration via the cloud, medical investigators seeking health research breakthroughs had to overcome significant obstacles to collaboration and key clinical data gathering. 

Data were siloed and difficult to distribute, so those looking to undertake research had no option but to gather them themselves. This not only made research more expensive, but it was challenging to compare findings across datasets. 

In 1975, researchers studying arrhythmias at MIT and Boston’s Beth Israel Hospital envisioned another way: the team began collecting and digitizing electrocardiogram recordings with the intention of not only studying them, but of also making them available to the wider research community.

The team built their own computers for the process, painstakingly duplicated tapes one by one, and created more than 100,000 annotations for the recordings. The process took years, but by summer 1980, the tapes were finally ready. The team initially thought their tool would reach fewer than a dozen academic and industry groups. But interest kept pouring in. Over the next decade, they went on to mail about 100 copies. 

The data eventually became the first database of the global platform PhysioNet — founded in 1999 at the Harvard-MIT program in Health Sciences and Technology — as a clinical data repository for complex physiological signals. 

At the time, that type of data-sharing, which may seem like the default today, was a near-revolutionary idea. PhysioNet’s “founding was incredibly visionary,” says Thomas Heldt, Richard J. Cohen (1976) Professor in Medicine and Biomedical Physics, associate director of MIT’s Institute for Medical Engineering and Science, and the senior author of a recent paper in Nature Health examining the platform’s impact. 

Eventually, those magnetic tapes sent through the mail became burned CD-ROMs, which then evolved into FTP servers hosted on the newly minted internet. Today, as PhysioNet looks back at over 25 years of operation, the platform hosts hundreds of databases, and has become one of the most comprehensive biomedical and clinical data repositories in existence. Last year, more than 15,000 scientific publications cited PhysioNet, and users from more than 180 countries have registered on the platform. It is widely used by researchers, manufacturers, and clinical decision-makers.

“The research impact is truly significant,” says Heldt, who is also a professor in the MIT Department of Electrical Engineering and Computer Science and a principal investigator at the Research Laboratory of Electronics, “and quite humbling.” 

“It is really beautiful to see that such a vision has proven right and so enabling for so many people.”

Setting a standard 

Around 2009, a PhD student named Tom Pollard was conducting research on critically ill patients at one of London’s leading hospital systems. Although the hospital generated large volumes of valuable clinical data, the infrastructure and processes needed to curate and support their wider research use were still developing. 

“Hospital data were collected primarily to support immediate patient care, with less attention given to how they might be curated and reused for research,” says Pollard, now a research scientist at MIT’s Laboratory for Computational Physiology (LCP), technical director of PhysioNet, and the lead author on the Nature Health paper. 

The problem was not simply privacy. Hospital information systems were built primarily to support patient care and administration, not research. Data were fragmented across systems and rarely curated with future reuse in mind, making it difficult and expensive to turn them into coherent research resources.

But Pollard needed data to complete his dissertation. After poking around on the internet, he eventually discovered the Medical Information Mart for Intensive Care (MIMIC), a database of de-identified electronic health records hosted by PhysioNet. Recognizing its potential, his clinical supervisor, Kevin Fong, organized a visit to Boston. Soon afterward, Fong and Pollard were sitting across the table from Roger Mark, discussing how their teams might collaborate.

Academic incentives have long favored publications and exclusive analyses over the less-visible work involved in preparing data for others to use. That tension persists today. PhysioNet’s founders embraced a different model, believing that sharing research resources could accelerate discovery and ultimately improve human health, he says. MIMIC became central to Pollard’s dissertation, and after completing his PhD, he came to MIT to help build the next generation of the database. 

In the years since PhysioNet was established, the value of sharing research data has gained much wider recognition. The late Roger Mark, MIT’s distinguished professor of health sciences and technology emeritus and one of PhysioNet’s founders, described its purpose as building an “accessible multinational community around data” to “positively impact global health.”

Earlier this year, Mark and the late George Moody, PhysioNet’s co-founder, jointly received the prestigious IEEE Biomedical Engineering Award for their contributions to PhysioNet and biomedical signal processing. IEEE cited their “leadership in ECG signal processing and global dissemination of curated biomedical and clinical databases, thereby accelerating biomedical research worldwide.”

The source code for the platform, like much of its data, is public. According to the Nature piece: “As the platform evolved, PhysioNet’s community broadened substantially beyond its origins in signal processing and cardiovascular health to encompass clinical informatics, critical care and machine learning for health.” People have used that to build their own PhysioNet-esque infrastructure, says Heldt. Pollard points to similar platforms like Health Data Nexus as examples of PhysioNet’s legacy. 

Although there are now more resources out there hosting similar electronic health data, according to Google DeepMind researcher Vivek Natarajan, both PhysioNet and MIMIC “set the standard,” he says, “and it’s still the standard right now.”

That standard, according to those who use the platform, changed how research is conducted. Access to data should not be the determinant for which ideas are possible, according to Ziad Obermeyer, an associate professor at the University of California at Berkeley School of Public Health and the College of Computing, Data Science, and Society. 

“PhysioNet changed how I think about the bottleneck in research. It is often not ideas or talent. It is friction. When access to data is slow, expensive, and hard, the ideas that die first are the high-risk ones, the things that probably will not work, but would be transformative if they did. That is exactly the wrong model if you want real progress,” he says. “PhysioNet lowers the fixed cost of trying ambitious ideas, and that changes what science becomes possible.”

The AI boom

PhysioNet, once a repository mainly for those working in biomedical signal processing and the health-care fields, has evolved in its 25 years. Originally, the holdings consisted solely of cardiovascular ECG data. Now PhysioNet is a largely a source for electronic health records, imaging data, and software and AI models.  

Particularly as artificial intelligence approaches took off, “the community shifted,” Heldt explains. Those in need of signal processing data still use PhysioNet databases, but the pool of users has expanded to encompass staff at large tech companies, teachers, and practitioners in all areas of medicine, as well as researchers in health-related machine learning and AI. Today, that latter group “dominates the user community,” says Heldt. 

The platform hosts the highest-quality datasets available for health-care AI research, according to Natarajan, whose research involves AI, science, and medicine and who has published several papers that used its datasets. 

“It has been an important cornerstone that has catalyzed all the progress in health-care AI over the last decade,” says Natarajan. In addition to using PhysioNet data, he and his colleagues have contributed data to the platform, helping create the self-sustaining ecosystem that typifies PhysioNet. 

Looking toward the coming decades, stewards of the platform like Heldt and Pollard envision continuing to expand its reach with an annual conference. The team is also preparing to pilot a new system that will allow users to annotate data and contribute their own expertise, enriching PhysioNet’s resources for the next phase of the platform.

“The kind of research that people want to do now needs to be interdisciplinary. Statisticians, computer scientists, clinicians, pharmacists, and nurses must all come together and contribute their knowledge to develop algorithms that are useful for people” says Pollard. “The community has broadened, and advances in AI have expanded both the questions researchers can address and what they believe is possible.” 

Robert E. Cohen, the Raymond A. (1921) and Helen E. St. Laurent Professor of Chemical Engineering, Emeritus, whose pioneering research helped shape the fields of polymers and soft matter while inspiring generations of students, passed away peacefully on July 9 following a long battle with Parkinson's disease. He was 79.

"Bob Cohen was an innovator in every sense of the word: in his research, his approach to mentorship, and in every aspect of our community at MIT," says Kristala Prather '94, the Arthur D. Little Professor and head of the Department of Chemical Engineering (ChemE). "Bob combined extraordinary intellect with remarkable humility. As a teacher, colleague, advisor, and friend, he had a gift for making people feel respected, valued, and heard. That generosity shaped every part of his work and inspired everyone fortunate enough to know him."

During more than four decades at MIT, Cohen continually reimagined how chemical engineering students should be educated. Recruited for his expertise in polymer science, he brought to MIT the polymer laboratory course he had developed during his postdoctoral work at the University of Oxford, establishing class 10.467 (Polymer Science Laboratory). The rigorous undergraduate course introduced generations of students to polymer synthesis, physical chemistry, and the evaluation of mechanical properties through hands-on experimentation.

In 1986, Cohen founded the Program in Polymer Science and Technology, now known as the Program in Polymers and Soft Matter (PPSM). Recognizing that advances in polymer science require expertise spanning chemistry, physics, engineering, and materials science, he created one of MIT's first truly interdisciplinary graduate programs. PPSM continues to prepare doctoral students to tackle complex challenges across the broad field of polymers and soft materials.

"Bob Cohen is the reason I returned to MIT as a graduate student," says Paula Hammond '84, PhD '93, Institute professor, dean of the School of Engineering, and a PPSM alumna. "His vision for multidisciplinary polymer education was unlike anything I had experienced. I benefited from him as a teacher in the classroom, as a member of my thesis committee, and a life-long mentor. As a department head, I saw firsthand the extraordinary impact he had on generations of students and on the field itself."

Cohen also conceived the unique PhD in chemical engineering practice (PhDCEP) degree, recognizing that future leaders in chemical engineering would benefit from combining advanced research with industrial experience and business education. The first and only program if its kind, the PhDCEP program integrates coursework, MIT's renowned David H. Koch School of Chemical Engineering Practice, doctoral research, and study at the MIT Sloan School of Management. 

Cohen also founded and directed the DuPont/MIT Alliance from 2000 to 2012, creating a highly successful partnership that brought together researchers from MIT and DuPont to develop innovative materials and manufacturing technologies. The collaboration advanced research in bioelectronics, biomimetic materials, alternative energy, and metabolic engineering, while fostering lasting collaborations across disciplines.

Cohen was a prolific collaborator whose pioneering research established him as one of the world's leading chemical engineers. His contributions include omniphobic surfaces, block copolymer nanoreactors for inorganic cluster synthesis, tough-stiff nanocomposites, chain folding in confined geometries, and layer-by-layer assemblies at the biotic-abiotic interface. Yet when asked about his proudest accomplishments, he rarely pointed to his scientific discoveries. Instead, he spoke about his students, and took immense pride in watching many former PhD students go on to become faculty members and leaders at top institutions around the world.

Raised in Oil City, Pennsylvania, Cohen developed an early appreciation for chemical engineering. After earning his master's and doctoral degrees from Caltech and completing a postdoctoral fellowship at the University of Oxford, he joined the MIT faculty in 1973. Over the next four decades, he became internationally recognized as a groundbreaking researcher, educator, entrepreneur, and mentor. 

Cohen was a member of the National Academy of Engineering and the American Academy of Arts and Sciences, as well as a fellow of the American Institute of Chemical Engineers, the Polymer Division of the American Chemical Society, the American Physical Society and the Materials Research Society. Cohen co-founded MatTek Corp., helping translate advances in biomaterials into practical applications.

Although Cohen received many prestigious honors throughout his career, he often said the award that meant the most to him was the inaugural Paul J. Flory Polymer Education Award, presented by the American Chemical Society in 2012. The honor recognized his leadership in building the interdisciplinary PPSM program and transforming undergraduate polymer education. Fittingly, it celebrated what he valued most: helping students discover their potential.

Cohen is survived by his beloved wife, Jane; his son, Eliot Cohen, his wife Jacqueline Aldred Cohen, and their children Ada, Brennan, and Callan; his daughter, Genevieve Cohen, and her daughter Emma; his sister, Nancy Stein, and her husband Herb; sisters-in-law Lee Woodman and Betsy Woodman; brother-in-law Wally Coleman; and many beloved nieces and nephews.

A memorial service is scheduled for Oct. 17 at the MIT Chapel. In lieu of flowers, donations may be made in Cohen’s memory to the Michael J. Fox Foundation.

MIT alumni Yu Deng ’11 and Hong Wang PhD ’19 were among the four young mathematicians awarded Fields Medals on July 23 at the 2026 International Congress of Mathematicians (ICM). The other two honorees were John Pardon and Jacob Tsimerman.

The Fields Medal is awarded once every four years at the ICM, and is regarded as one of the highest honors a mathematician can receive. 

Yu Deng received his BS in mathematics at MIT in 2011, and was a Putnam Fellow in 2010. He earned his Fields Medal for his work in partial differential equations (PDE), including the rigorous derivation of the Boltzmann equation from hard-sphere dynamics for rarefied gases, the derivation of wave kinetic equations from nonlinear dispersive systems, and probabilistic approaches to nonlinear Schrödinger dynamics. His first published paper (in Analysis & PDE) was on the latter topic, and stemmed from summer research conducted at MIT on a problem suggested by Gigliola Staffilani. Deng is currently a professor at the University of Chicago.

Hong Wang received her PhD at MIT in 2019 under the supervision of Larry Guth PhD ’05. She is awarded the Fields Medal for her work in harmonic analysis and geometric measure theory, including applications of multiscale and decoupling techniques to the local smoothing conjecture for the planar wave equation, and other major advances such as the solution of the Kakeya problem in three dimensions (with Joshua Zahl). As a student in the department, she was a graduate mentor in the Summer Program in Undergraduate Research (SPUR) and, alongside her mentee, was awarded the Hartley Rogers Jr. SPUR Prize, presented to the best student-mentor team. Wang, a Silver Professor of Mathematics at New York University and a professor at the Institut des Hautes Études Scientifiques in Paris, is the third woman ever to win a Fields Medal.

“The achievements of Yu Deng and Hong Wang are truly monumental, and we are all elated that they were awarded Fields Medals,” department head and RSA Professor of Mathematics Michel Goemans says. “Their success is a testimony of the amazing mathematical talent we have at all levels at MIT, and the top-quality education, mentorship, and research opportunities we provide to both our large pool of math majors and our PhD students, during their lifelong mathematical journey.” 

Goemans adds, “MIT is a unique and exciting place to learn mathematics, and I am sure we have more future Fields medalists among our students and junior members of the department.”

A new method developed by MIT researchers makes robots better at thinking ahead while they are acting, leading to smoother motions and quicker reactions.

This technique enables the artificial intelligence model that plans a robot’s motion to forecast its future position. The model uses this prediction to seamlessly transition current movements into the next actions.

Many existing methods cause a robot to stop and think about what it needs to do next, leading to slow and jerky motions. By basing its calculations on the future state of the robot, rather than its current position, the MIT method helps robots operate much faster.

Importantly, the technique does not add any computational overhead to the planning process and can be applied to varied robotic hardware.

This new method doubled the speed of robots performing activities like pick-and-place tasks, while significantly reducing lag time between motions. It also boosted the performance of robotic arms in highly dynamic activities, such as playing table tennis and Whack-a-Mole.

The system could be especially useful for robots that perform fast and agile maneuvers in challenging real-world environments, like emergency response or search-and-rescue. It could also allow robots to react more quickly when recovering from mistakes.

“This work sets up a good foundation for efficient, fast, accelerated, and low-cost robotics applications. We look forward to expanding our work into the latest world action models, so it has even stronger capabilities as we keep pushing to make physical AI faster,” says Song Han, an associate professor in the MIT Department of Electrical Engineering and Computer Science (EECS), member of the Research Laboratory of Electronics, and lead author of a paper on this method.

Han is joined on the paper by co-lead authors Jiaming Tang, an MIT EECS graduate student, and Yufei Sun, a student at Tsinghua University; as well as others at Nvidia, the University of California at Berkeley, the University of California at San Diego, and Caltech. The research will be presented at the Intelligent Robots and Systems Conference. 

Forecasting the future 

In state-of-the-art robotics applications, generative AI systems called vision-language-action (VLA) models act as the brain of a robot, planning its next moves and executing those actions. 

A VLA model takes environmental observations from the robot’s camera and instructions about its task, outputs the next few motions as one chunk of actions, then executes those actions on the robotic hardware.

But VLA inference — the real-time procedure during which the model processes visual inputs, reasons about the task, and outputs actions — is computationally demanding, so the robot can experience substantial pauses while planning its next actions. These pauses disrupt the fluidity of its motions and make it slower to react to changes in the environment. 

“Our motivation was to overlap the thinking process with the execution process to make the reaction speed faster,” Tang says.

The MIT researchers developed a new system called VLASH that enables a VLA to predict the future state of the robot and its environment. It uses this information to plan the next set of motions while the robot is completing the current action chunk.

This solves a major hurdle faced by many other methods, which use the current state of the robot to predict its next moves.

“Since the environment will change after the robot moves, if we plan based on stale observations of the current environment, there will be a misalignment that causes very unstable control,” Tang explains.

VLASH avoids this misalignment due to a key insight by the researchers. Although the model doesn’t know exactly what the environment will look like in the future, it does know the robot’s current position and how it will move to perform the actions it is about to take. 

The framework uses this information to predict the state of the robot after it completes its current chunk of actions. It uses that estimation to plan the next motions.

“In this way, we give the robot awareness of its future state,” Tang says.

Augmenting acceleration

On its own, this technique speeds up the robot’s motions by eliminating lag time that usually occurs between action chunks, accelerating reaction speeds more than 30-fold. 

But to make their approach even faster, the MIT researchers generate coarser chunks of actions, so the robot executes a few larger steps that follow the same trajectory. This technique is called action quantization.

While action quantization led to a slight dip in accuracy, it enables a robot to complete the overall task two to three times faster.

However, the researchers found that simply feeding future robot states to the VLA during deployment is not enough to enable accurate and stable control of the robot. 

They developed a training-augmentation method that groups training data in such a way that the VLA learns to use future state information instead of current observations. 

By reusing some training data, this fine-tuning method accelerated training fivefold with no additional computational overhead. 

“Even though there is a very large model working in the background, VLASH lets the robot react and execute its actions very fast, much more like a human would. This could help to make robots for all sorts of dynamic tasks more effective,” Tang says.

When compared with baseline methods in simulation, VLASH consistently performed faster while maintaining the accuracy of robotic maneuvers. The system also outpaced these methods on real hardware in pick-and-place, stacking, and sorting tasks.

For instance, VLASH placed cubes in a box while sorting them by color twice as fast as these methods, while achieving the same 90 percent accuracy as the best baseline. The system can also perform highly dynamic tasks like playing ping-pong and whack-a-mole.

In the future, the researchers want to combine VLASH with more powerful generative AI systems called world models that can predict the robot’s actual environmental observations, in an effort to boost performance and open new applications.

This work is supported, in part, by the MIT-IBM Computing Research Lab, Amazon, the National Science Foundation, and Nvidia.

After a closet cleanout, what options are there for recyling our old threads? Not many. Apart from bringing used clothes to a donation center, there is no process for recycling textiles like there is for bottles and cans. And, the average American throws out around 81 pounds of clothing each year. That amounts to more than 11 million tons of textiles that end up in the landfill or incinerator. 

But MIT engineers hope to cut down on the growing mountain of textile waste, with a new, recyclable yarn. 

The team has designed a yarn made from a form of plastic that is commonly used in milk bottles and grocery bags. The new yarn, which has a feel similar to traditional sewing thread, can be woven into stretchy, lightweight clothing. The researchers say that at the end of its use, a yarn-spun garment could be melted down and redrawn into new yarn, and then woven into new clothing or even cast into buttons, belt buckles, and other plastic accessories.

To demonstrate the yarn’s recyclability, the researchers spun a spool of yarn, melted the yarn down, and respun it into new yarn, multiple times. They found that even after 10 cycles, the yarn was as strong and flexible as conventional thread. 

They envision the new yarn could be an alternative to elastic spandex-polyester or spandex-nylon yarns, which are spun from a combination of fibers that cannot be recycled together. The team’s new yarn, in contrast, is made from a specific combination of plastic materials that mimics the tough and stretchy properties of spandex yarns, while also being easily recycled. 

“Eighty percent of textiles on the U.S. market currently contain some amount of spandex, which makes them nonrecyclable,” says Svetlana Boriskina, a research scientist in MIT’s Department of Mechanical Engineering. “There’s no widely adopted technology now that recycles textiles into textiles. With our new yarn, we hope to change that.”

Boriskina and her colleagues have published the details of the new yarn in a study published in the journal ACS Materials Letters. MIT co-authors include first author SeongHyeon Kim, Duo Xu, Volodymyr Korolovych, Domingo Flores-Hernandez, Kaniz Moriam, and Daniel Braconnier.

The core of the problem

Spandex is a polyurethane-based synthetic fiber that is springy but not very strong. A thread of an elastic yarn is made from two parts: a spandex-based core, surrounded by a sheath of tough polyester or nylon. The combination of these materials gives elastic yarns their unique stretch and strength. 

But this same material mixture makes elastic yarns nearly impossible to recycle. Yarns would first have to be chemically treated to separate the polyester sheath from the spandex core. The polyester-based sheath material could then be melted down and reused. But there is no way to recycle the yarn as a whole, without chemical separation.

“Even though chemical separation technologies exist, they add extra cost and complexity, and usually require toxic chemicals that are harmful to the environment,” Boriskina says. “That’s why most stretchy garments go to the dump.”

In 2021, Boriskina’s group developed a new type of yarn made from polyethylene. Polyethylene is the most common type of plastic in the world, used to make everything from grocery bags, water bottles, trash bins, and toys to industrial pipes and plastic sheeting. Polyethylene is a thermoplastic, meaning that it can be melted down and remade, and thus recycled. 

And yet, polyethylene had never really been considered as a textile. In their previous work, Boriskina and her colleagues showed they could spin yarn out of polyethylene, which they then wove into various garments. In those experiments, they focused on the yarn’s moisture wicking, stain-resisting, and cooling properties. 

Spaghetti yarn

In their new study, the group aimed to tailor polyethylene yarn to mimic the strength and flexibility of spandex; they also sought to demonstrate the yarn’s recyclability. 

They first looked for formulations of stretchy, polyethylene-based copolymers that resemble a spandex elastic core. Separately, they engineered polyethylene yarns that can act as the sturdier sheath. Looking through the scientific literature and combing through industrial reports, the team evaluated many chemical variations of polyethylene.

“The chemical structure of polyethylene is like Christmas garland — a backbone of carbon, carbon, carbon, and also these dangling ‘decorations’ of hydrogen atoms or short branches with the same structure as a backbone,” Boriskina explains. “How these chains are arranged can change the properties of the whole structure.”

“Polyethylene can give us a wide range of properties, depending on how you make it,” adds first author SeongHyeon Kim.

For the yarn’s core, the team used one polyethylene-based resin that results in a more stretchy fiber. They chose a second, stiffer resin as the basis for the yarn’s sheath. The researchers obtained pellets of each resin from a chemical manufacturer, and then put each type of pellet through a process of fiber fabrication, first pouring them into a hopper, then heating the pellets to about 350 degrees Fahrenheit, past their melting temperature. The melted polyethylene was then drawn through small extruders to make hair-thin fibers. 

“You just melt it in a barrel with a heater, and then you extrude and spin it into fibers,” Kim says. “It’s like a spaghetti machine.”

The team used an industrial yarn spinner to wind the sheath fibers around a core fiber to make the final, elastic yarn. 

Because both the yarn’s core and sheath come from the same chemical family of polyethylene, Boriskina says the materials do not have to be separated before recycling, in contrast to spandex-based elastic yarns. The new yarn can be melted as is, and reformed into new yarn or other plastic products.

“Because they are exactly the same chemistry, they play nicely together,” she says. “That’s what makes this yarn very recyclable.”

As a demonstration, the team twisted an elastic core-sheath yarn, then melted it down and re-spun it, 10 times. Each time, they tested the yarn’s mechanical properties by precisely stretching a thread and measuring the pulling force at which the thread eventually broke. From these tests, they found that the yarn’s recycled versions were just as strong as the original sheath yarn. These recycled yarns can now be used to make new stetchy yarns by twisting them around a newly spun elastic core.

“Now we have something that can be knitted and woven,” Boriskina says. “That is the next stage.”

The team says their new recipe for polyethylene yarn can be scaled up into industrial-sized spools. Just like conventional spandex fibers, it would take kilometers of yarn to weave a single textile. But once woven and used, the team envisions that a polyethylene garment could conceivably be dropped in a recycling bin and sent to a facility to be melted down and respun, enabling a more sustainable, circular fashion and textile economy. 

“Hopefully it will prevent the need for making more and more textile materials, because you can keep recycling a large portion of it,” Boriskina says. 

This work was supported in part by the DEVCOM Soldier Center through the U.S. Army Research Office, the Office of Naval Research Global via Tecnologico de Monterrey, and the MIT Portugal Program.

In Professor Anna-Christina Eilers’ research group, mentorship happens through small, meaningful gestures: thoughtful feedback on a draft, a check-in after a rough week, and a readiness to help when things get tough. For her students, these everyday moments have become a defining feature of her approach.

An observational astrophysicist, Eilers studies how the universe evolved from its earliest beginnings. Her research investigates the formation and growth of black holes across cosmic time, particularly during the “cosmic dawn,” when the first stars, galaxies, and quasars illuminated the young universe.

Working alongside her in this field, graduate students describe a mentor who pairs high expectations with genuine attentiveness, encouraging both scientific independence and a strong sense of community. This approach has earned Eilers recognition through MITs Committed to Caring initiative — a student-driven program honoring exemplary mentorship within the graduate community.

Showing up in the everyday moments

Students say one of Eilers’ defining qualities is her consistency. No matter how busy her schedule, they know they can count on thoughtful feedback, productive meetings, and regular conversations about both research and broader career development. While those practices may sound routine, her mentees emphasize that they are anything but guaranteed within many academic spaces.

“As Christina's advisees,” two students wrote in their joint nomination, “we are both extremely grateful for the professional and emotional support we constantly receive. She always keeps an eye out for us.”

Eilers’ support takes many forms. Students describe an advisor who carefully reads every draft, provides timely and detailed feedback, and creates space for conversations that extend beyond immediate research questions. 

Students also reflect on the manner in which Eilers celebrates their wins alongside them. “She brings our favorite desserts to group meetings when we publish a paper,” shared one nominator. 

Her attentiveness becomes especially meaningful when challenges arise. Students note that she regularly checks in on them and does not hesitate to step in when research collaborations become difficult or obstacles threaten to slow their progress. Rather than leaving them to navigate those situations alone, she helps identify solutions before small problems become larger ones.

For Eilers, building a successful research group means cultivating connections among its members as well as producing strong science.

One of the group’s traditions takes place whenever a member returns from a conference or research visit. The traveler brings back a small treat — cookies, chocolates, or another local specialty — to share during the next group meeting. Along with the snacks comes a conversation about the talks they attended, the researchers they met, and the ideas they brought home.

The tradition transforms an individual trip into a shared opportunity for learning, with new perspectives becoming part of the group’s collective conversation. These exchanges work to not only reinforce a sense of community, but also to expose students to research and ideas beyond their own projects.

Through moments like these, students develop both as researchers and as colleagues who celebrate one another’s successes and learn from one another's discoveries. 

Remembering the person behind the researcher

One of Eilers’ most consistent pieces of advice has little to do with coursework or research.

“I always recommend to incoming graduate students to find a hobby outside of work that they enjoy, and ideally where they interact with people they don’t work with,” she says.

She believes maintaining interests beyond the lab helps students sustain both their curiosity and their perspective. “Graduate school can be all-consuming,” she says, reflecting on her own experiences. “It's easy to let your research become your entire identity.”

This same philosophy shapes her mentorship: successful researchers are also people with lives, relationships, and interests beyond their work. Making space for those parts of life helps students build careers that are both ambitious and sustainable.

Eilers traces her approach to the advisors who shaped her own career.

“I was very fortunate to have had — and continue to have — several mentors who have challenged me scientifically and supported me along the way," she says. “They modeled how to pursue excellent research without losing sight of the importance of personal connection and integrity.”

Her students see these values reflected within the group environment. They are encouraged to tackle ambitious questions while developing the confidence to think independently, but they know that guidance is available when they need it. 

In their nominations of Eilers, students describe an advisor who is present in both the ordinary and the difficult moments — someone who notices when support is needed, advocates for her students, celebrates their successes, and builds a community where students consistently feel seen. 

Through this steady commitment, Eilers demonstrates that care is not separate from academic excellence. Rather, it creates the conditions that allow excellence to flourish.

MIT researchers are set to contribute to the U.S. Department of Energy’s (DOE) Genesis Mission, with 15 collaborative projects among those selected for funding under Genesis Phase I, DOE announced Wednesday.

The Genesis Mission, a national initiative, intends to build “the world’s most powerful integrated science discovery platform” by incentivizing cross-sector collaborations that leverage AI, supercomputing, quantum systems, and advanced scientific instruments to accelerate breakthroughs in energy, scientific discovery, and national security.

“MIT researchers are proud to be leading and contributing to projects under the Genesis Mission, in vital areas of research that support national priorities,” says Ian A. Waitz, MIT’s vice president for research. “The Genesis Mission represents a fantastic opportunity to catalyze the power of universities, industry, and the U.S. national laboratories to advance science, technology, and innovation for the benefit of the nation and the world.”

The DOE announced the initial projects during its Genesis Summit in Washington on Wednesday. The research funding to MIT is pending completion of negotiations toward an award agreement for each project. In phase I, funded project teams will work to demonstrate research workflows that integrate AI with scientific investigation, and to rigorously evaluate the scientific merit of their approach.

Projects under the Genesis Mission are collaborative by design; teams must draw on the expertise of researchers from academia, industry, and/or the national laboratories. Among the selected phase I projects with MIT involvement are those that aim to develop powerful quantum sensors to help explain fundamental questions about the universe; advance knowledge of chemical-free methods to extract rare earth elements; model the behavior of plasma in fusion tokamaks and future fusion reactors; develop digital twins for fusion magnet systems; exploit the self-assembly of biomolecules to design materials with targeted properties; generatively design rotating blades for machinery systems; and more. Phase I projects that identify promising pathways toward transformative capabilities at scale may be considered by DOE for further Genesis Mission funding.

Six of the selected projects are to be led by MIT principal investigators (PIs):

• AI-Driven Discovery of Electrochemical Separation Methods for Rare Earth Elements
  MIT lead: Martin Bazant (Department of Chemical Engineering, ChemE), Chevron Professor in Chemical Engineering and professor of mathematics
   
• AI for Learning Missing Constitutive Structure in Fracture Models
  MIT lead: Laurent Demanet (Department of Earth, Atmospheric and Planetary Sciences), professor of applied mathematics in the Department of Mathematics and co-director of the MIT Center for Computational Science and Engineering
   
• AI-Driven Quantum Sensing for Precision Tests of Fundamental Physics
  MIT lead: Ronald Garcia Ruiz (Laboratory for Nuclear Science, LNS), associate professor of physics and Thomas A. Frank (1977) Career Development Professor
   
• Multi-Agent Inverse Design of Block Polypeptoids Into Hierarchical Nanostructures
  MIT lead: Bradley Olsen (ChemE), Alexander and I. Michael Kasser (1960) Professor
   
• CATALYST: Core Accelerated Trajectories with Augmented Learning bY Sim-to-experiment Transfer
  MIT lead: Cristina Rea (Plasma Science and Fusion Center), principal research scientist and division head for data science
   
• Multi-Modal and Multi-Facility Application of the FM4NPP Foundation Model: Silicon Trackers and Electron Colliders
  MIT lead: Gunther Roland (LNS), professor of physics and division head for experimental nuclear and particle physics

MIT researchers are expected to participate in another nine selected projects led by other institutions, companies, and labs:

• Framework for Optimized Rotating Blade Design Using Generative Engineering (FORGE)
  Project lead: GE Vernova Advanced Research Center
  MIT lead: Faez Ahmed (Department of Mechanical Engineering), associate professor of mechanical engineering and the Esther and Harold E. Edgerton Career Development Professor
   
• Superconducting Polychronous Computation Near Criticality
  Project lead: Argonne National Laboratory
  MIT lead: Karl Berggren (Research Laboratory of Electronics), the Julius A. Stratton Professor in Electrical Engineering and Physics
   
• Scalable Agentic Digital Twins for Autonomous Precision Facilities
  Project lead: Texas A&M University
  MIT lead: Ronald Garcia Ruiz (LNS)
   
• Agentic AI for Real-Time Expedited Discovery from High-Complexity EIC Data Streams
  Project lead: Purdue University
  MIT lead: Philip Harris (LNS), associate professor of physics
   
• Self-Driving Discovery and Co-Design of MXene Memristors for 3D Compute-in-Memory Systems
  Project lead: Northeastern University
  MIT lead: Ju Li (Department of Nuclear Science and Engineering, NSE), the Carl Richard Soderberg Professor in Power Engineering and professor of materials science and engineering
   
• A-WILD: AI-driven Workflows for Intelligent Lab Discovery
  Project lead: Lawrence Berkeley National Laboratory (LBNL)
  MIT lead: Ju Li (NSE)
   
• A Foundational Generative AI Framework to Advance Water-Energy Security
  Project lead: LBNL
  MIT lead: Haruko Wainwright (NSE), Atlantic Richfield Career Development Professor in Energy Studies, assistant professor of nuclear science and engineering, and assistant professor of civil and environmental engineering
   
• An AI-Driven Platform for HLW Repository Design and Analysis with Digital Twins, GIS Data Integration, and Surrogate Models
  Project lead: LBNL
  MIT lead: Haruko Wainwright (NSE)
   
• Toward Physics-Informed Digital Twins for Fusion Magnet Systems
  Project lead: LBNL
  MIT lead: Holger Witte (LNS), associate director of MIT’s Bates Research and Engineering Center.

“The extraordinary response to this Genesis Mission application process demonstrates that America’s scientific community is ready to reimagine how discovery happens,” said DOE Under Secretary Darío Gil SM ’00 PhD ’03, in the DOE’s announcement. “Through the Genesis Mission, we are bringing together the nation’s leading researchers, institutions, and technology partners to build the next generation of scientific capability. We look forward to seeing these teams demonstrate new research workflows that accelerate discovery and reveal what is possible when AI and science advance together.”

A complete list of the first Genesis Mission projects selected for award negotiations is available from the U.S. Department of Energy.

As a Massachusetts Boys Soccer Player of the Year and an All-American in high school, Dilin Meloni had choices when it came to college. He came to MIT because he knew he wanted to be an Engineer (on the field) and an engineer (off) after he read a paper from the Plasma Science and Fusion Center as a junior at Needham High School. 

In mid-June, Meloni, a rising senior studying nuclear science and engineering and physics, took time from his summer internship at Commonwealth Fusion Systems to discuss his role as an attacking midfielder for the Engineers, his work with the Student Athlete Advisory Council, and more.  

Q: How does being an athlete affect your experience at MIT?

A: MIT has a huge social experience people don’t really understand from the outside. A lot of it is driven by our athletic communities. 

For example, MIT was very open and welcoming when I first got here. Especially for fall athletes; when you get here, it’s all there in the first two weeks. Athletics help you meet students who are very much like yourself: people who come to MIT for academics but who have also dedicated so much of their lives to a sport that it’s a major part of them. I think this common ground makes it a lot easier to meet people and find connections you might have a harder time finding if you're not an athlete.

And it still drives how I live. My fraternity is made up of pretty much the entire soccer team now, plus people from the football team, the swimming team, and some track athletes. We have some non-athletes as well. 

But the nice part is it doesn’t have to be everything you do — and it never is, because everybody has their academic work.

Q: Are these two parts of your life — athletic and academic — separate, or integrated?

A: Maybe it’s because of my major and what I’m interested in, but I just don’t find a lot of overlap. I’m a Course 22 and Course 8 double-major, and you don’t find a ton of varsity athletes in nuclear engineering or physics, unfortunately. But I’m trying to change that — the new players coming in, I try to convince them these are good paths to take.

Growing up, I spent a lot of time playing club soccer, and my academics were in a much smaller sphere. Here, I get access to every single part of the MIT ecosystem. I’m able to spend time in my lab with people who come from very diverse backgrounds, and they don’t really play sports. Or they don’t anymore. 

And at the end of the day, I can go back and hang out with my friends on the team. It’s a privilege to be a part of all these different groups. I’m not just siloed.

Q: Describe the Student Athletics Advisory Committee (SAAC). How do you represent the needs of all 33 varsity sports?

A: Thirty-three teams is a lot of teams. … Last summer we reached out to all the coaches and athletes before the preseason to ask for their ideas. They know where the pain points are, so we encouraged them to think about what changes they wanted. Then the co-president and I met with all of them in smaller groups and came up with a list. 

The main idea is that there’s overlap on the issues that pertain to all student-athletes, such as the quality of the food, preseason planning, or the new Sports Performance Facility — every varsity student athlete needs access to that space, so it matters how access and time are allocated. 

Or it might be something more specific. There are seven teams that play on the major turf fields. If you’re a swimmer, you’re probably not going to care much about the soccer fields, but you do share the pool with the water polo team. We also discuss things like NCAA [National Collegiate Athletic Association] legislation and how the school wants to approach it.

The nice part is, it’s pretty informal. I know a lot of student athletes. If I want to get in touch with someone from the track team, I don’t have to hunt them down. I could probably just find them in class. Or I text my friends and ask, “Hey, how was the food today?”

Q: How did you get involved? 

A: I started going my freshman year. My coach suggested that we send some younger players, so I went with the captain at the time. 

Student governance has always been important to me. When I was in high school, I was part of my town’s school committee and the larger school committees for Massachusetts. Students have lots of opinions, and there are lots of issues, even at a place like MIT. But if no one takes on the role, nothing’s going to happen. That’s just always how I viewed it — if there’s an area where you can make some change, why not go do it yourself?

Q: What’s the hardest part of the job?

A: The difficult part can be communicating realistic standards between administrators and student athletes, because students are going to ask for more than they’re going to get — which is not necessarily a bad thing. For example, student athletes are aware that we have to share fields. But all these teams always want to be out there. They want to have priority. They want continuous access to facilities so they can train and get better as a team.

In these meetings I’ve come to understand fully that there’s a cost associated with everything, and some things can be difficult, logistically. So finding compromises when it comes to field time, or food, or whatever it may be — I end up explaining a lot. This is how the school works. You can’t just demand everything. We try to meet in the middle as much as possible. Students get it.

But on the flip side, I need to make sure we are voicing student athletes’ needs and ideas for how to improve our experience, because if you’re enjoying your student-athlete experience, you're going to get more out of the school as well. 

Varsity athletes feel sometimes that they are an underrepresented or unloved group. If you’re trying to have a soccer practice and there are people running on the track, or if you kick a ball over the fence and someone steals it, or if you’re trying to book an off-season practice and a student club already booked the field — all these things can be frustrating.

A big part of SAAC is voicing complaints, but also understanding the compromises — and that we can work together to try and find the best middle ground.

Q: Do you think it’s working?

A: Definitely. DAPER and Student Life staff have been willing to talk, and listen, and overhaul how they do things. With such a large portion of the student body being athletes, this is really important to them. They’re now asking me and the co-president for input on other realms of student life, too. I think this year was really a tipping point.

Obviously, it’s a privilege to come to this school, but if you’re here you still need to be heard. At the end of the day, people are always going to find things to nitpick. Yes, there are things to improve on; it’s never going to be perfect. But as long as we have this communication channel open, I think we can make small changes that’ll lead to big changes eventually.

Dimitri Bertsekas PhD ’71, the Jerry McAfee (1940) Emeritus Professor in Engineering in the Department of Electrical Engineering and Computer Science (EECS), a principal investigator in the Laboratory for Information and Decision Systems (LIDS), and the Fulton Professor of Computational Decision Making at Arizona State University, died on June 3 at his home in Belmont, Massachusetts. He was 83 years old. 

Over the course of his career, Bertsekas’ research spanned, and had a definitive influence upon, several fields, including optimization, control, large-scale computation, reinforcement learning, and artificial intelligence. He served as a consultant to various private companies; an editor for several scientific journals; the founder of a publishing company, Athena Scientific; and chief scientific advisor of Bayforest Technologies, a London-based quantitative investment company. However, his most lasting impact may have come through his prolific authorship and co-authorship of over 20 highly influential books, monographs, and textbooks, and through his vast network of students, mentees, friends, and collaborators.

Bertsekas earned his undergraduate degree at the National Technical University of Athens, Greece, before obtaining his MS in electrical engineering at George Washington University in 1969, and his PhD in system science at MIT in 1971. He began his faculty career at Stanford University, where he spent three years, and the University of Illinois at Urbana-Champaign, where he spent five more before returning to MIT in 1979. He would stay with MIT’s Department of EECS until 2019, at which point he became a full-time faculty member at Arizona State University at Tempe. Along the way, Bertsekas taught, advised, and mentored students who would eventually become his colleagues at all four institutions. 

“Dimitri played a defining role in my career,” says Asu Ozdaglar, department head of EECS at MIT. “I decided to change my research focus after taking his nonlinear optimization class. The conceptual clarity and the mathematical rigor he has brought to every topic, combined with his ability to connect theory to important problems established a foundation that has continued to inform my scholarly work in the years to follow.” Another former MIT student, Jinane Abounadi, now executive director of the MIT Sandbox Innovation Fund Program, still remembers Bertsekas’ tutelage as a highlight of her time as a student at MIT: “I feel so fortunate to have had Dimitri as my professor and advisor. I had the opportunity to learn about optimization, dynamic programming, and neuro-dynamic programming from a true master.” 

A former student at the University of Illinois, Steven E. Shreve remembers being impressed by Bertsekas’ course on nonlinear optimization and asking if Bertsekas would consider becoming his PhD advisor. “Rather than answering my question directly, Dimitri gave me a preliminary draft of his manuscript, which eventually became his book 'Dynamic Programming and Stochastic Control,' and asked me to proofread it,” remembers Shreve, now Orion Hoch University Professor Emeritus in the Department of Mathematical Sciences at Carnegie Mellon University. “From this manuscript, I learned the theory of dynamic programming and mastered many important special cases. Talking with Dimitri as I read, I received one-on-one instruction. When the book finally appeared, Dimitri generously acknowledged my participation, as if I had done him a favor, rather than the other way around.” The gambit was typical of Bertsekas’ understated approach to mentorship; after the first successful collaboration, Bertsekas arranged a research fellowship for Shreve and challenged him to solve a fundamental question in dynamic programming. “I needed to learn a good deal of set theory to even think about the question he asked,” remembers Shreve, whose work on the problem was combined with Bertsekas’ notes to create their co-authored book “Stochastic Optimal Control: The Discrete Time Case.” 

“Working with Dimitri on [that book] is how I learned to write,” says Shreve. “I learned from Dimitri that if you want to be recognized for your research, you must present it so others want to read it, and I learned how to do that.” 

The clarity and elegance of Bertsekas’ explanatory style would become his educational hallmark. “Everyone recognized Dimitri’s great talents as a writer, but he went far beyond that, organizing entire subjects into something that was understandable and a well-organized totality,” says Robert Gallager, professor emeritus of electrical engineering at MIT, who co-authored a 1987 book with Bertsekas entitled “Data Networks.” “The field was changing rapidly then, with a factor-of-two decrease every two years in computation costs, and with optical fiber on the horizon for transmission. Dimitri and I each understood only parts of this field, with the rest a fast-moving learning experience. Dimitri was the ideal partner in this, able to quickly translate hard concepts into simple but accurate explanations and able to combine my knowledge with his into an understandable whole.” 

Bertsekas’ close colleague in LIDS, Munther Dahleh, remembers, “what always struck me was that, through his writing, one could almost hear Dimitri speaking directly to the reader. His intuition, clarity of thought, and distinctive perspective come through beautifully in his books. They reflect not only his profound technical contributions, but also his passion for teaching and his desire to help others understand the subject at a deep level. … In particular, his joint book with John Tsitsiklis on neuro-dynamic programming is a tour de force. It anticipated and helped define many of the ideas that later became central to reinforcement learning and approximate dynamic programming.” 

Tsitsiklis himself remembers the co-writing process with Bertsekas fondly: “For Dimitri, research was a creative form, combining craftsmanship and the creativity that we usually call art.” The definition of art and its practice was a subject of great fascination for Bertsekas, and one that he explored at length in his 2025 essay, “Academia, Art, and Life,” an attempt to meaningfully categorize creative work into three broadly descriptive roles — technician, craftsman, and artist — and to explore the overlaps between the three types of practice. Beyond his clear and lucid writing, Bertsekas was known for his strong graphic eye, a talent which he put to good use not only developing illustrations for all his textbooks, but in taking memorable and artistically inspired photographs of his worldwide travels. 

Longtime collaborator and friend David Castañón, now a professor of electrical and computer engineering at Boston University, remembers Bertsekas as a true Renaissance man who drew inspiration from countless sources: “Dimitri had an insatiable curiosity for algorithmic ideas, both theory and practice. Many of these ideas were inspired by new technologies (parallel computers, reinforcement learning, chess-playing algorithms) ... Whenever we met, Dimitri would introduce new concepts of interest; we would work out theoretical details, design and conduct numerical experiments, and generate results. Then, Dimitri’s artistic talents would take over: designing graphics to illustrate concepts, typesetting text and figures for the papers to be completed. He had a rare gift for generating concise explanations of complex concepts. These talents led to his publishing company Athena Scientific, where Dimitri and his coauthors generated elegant pedagogical volumes with broad appeal.” Tsitsiklis agrees, noting, “for Dimitri, [research] was about discovering meaning, to uncover the 'right' way to view a subject, enrich it, and convey it in a crystal-clear manner through his prolific writings.” 

Many of the 20-plus books either authored or co-authored by Bertsekas were adopted for use as textbooks at MIT in subjects including data networks, nonlinear programming, dynamic programming, network optimization, parallel and distributed computation, neuro-dynamic programming, convex analysis and optimization, probability, and reinforcement learning. Stephen Boyd, Samsung Professor in the School of Engineering at Stanford, testifies to the great impact of Bertsekas’ collected works: “generations of researchers in optimization, control, and many related areas learned these topics from Dimitri’s exquisitely clear and beautifully written text books. I was one of them; indeed, I went into these fields in no small part because of Dimitri’s books, and his influence has been with me the whole time.”

That influence can be measured by the sheer number of awards and honors Bertsekas accumulated over the course of his career, including the INFORMS 1997 Prize for Research Excellence in the Interface Between Operations Research and Computer Science for Neuro-Dynamic Programming, the 2001 ACC John R. Ragazzini Education Award, the 2009 INFORMS Expository Writing Award, the 2014 ACC Richard E. Bellman Control Heritage Award for “contributions to the foundations of deterministic and stochastic optimization-based methods in systems and control,” the 2014 Khachiyan Prize for Life-Time Accomplishments in Optimization, the SIAM/MOS 2015 George B. Dantzig Prize, and the 2022 IEEE Control Systems Award. Together with his coauthor John Tsitsiklis, he was awarded the 2018 INFORMS John von Neumann Theory Prize for the contributions of the research monographs “Parallel and Distributed Computation” and “Neuro-Dynamic Programming.” In 2001, Bertsekas was elected to the U.S. National Academy of Engineering for “pioneering contributions to fundamental research, practice and education of optimization/control theory.”

However, a more personal measure of Bertsekas’ impact can be taken by the warmth and affection with which his friends, co-workers, and former students uniformly remember him. Co-author John Tsitsiklis wrote, “I was most fortunate to be one of his apprentices, and to have lived his warmth and friendship.” His former student at MIT, Angelia Nedich, later became Bertsekas’ colleague at Arizona State University. She remembers: “Dimitri was an exceptional mind, a gifted soul that shed light for us seekers, but at the same time he was very humble as he enjoyed simple moments of life, a sip of good coffee, a bite of flavorful food, or a glass of spicy margarita on our road trips in Southwest. That is how I love to remember him.”

Former student Benjamin Van Roy, now a professor at Stanford, wrote about the transformation of Bertsekas from authority figure to friend (and the subject of friendly teasing). “I recall the intimidating comments of more senior PhD students as I began my own PhD journey in LIDS. Some referred to Dimitri as an “immortal.” Another comment I recall fondly — and often reminded Dimitri about — was: “Professor Bertsekas is a very handsome man!” Their bond continued long after Van Roy’s graduation. “Dimitri was a treasure to humanity: one of the great scholars of our time, a Renaissance man, and a phenomenal role model. I was privileged to be among the many he mentored, and even more privileged to count him as a longtime friend.” Yuchao Li, a postdoc mentored by Bertsekas at Arizona State University, remembers his mentor as an almost inexhaustible source of both inspiration and support: “For me, Professor Bertsekas was like a loving father, full of infinite wisdom. … He seemed to know everything, yet he remained deeply humble and open-minded. He was always eager to help, even at the slightest sign of difficulty in my life. He instilled in me a lasting faith in the very best qualities of human beings, and I will strive to carry that faith forward.”

Bertsekas was preceded in death by his son Costas. He is survived by his wife Joanna Bertsekas (née Palashas); his son Telis Bertsekas and his wife Wendy Bertsekas; and three grandchildren, Melina, Alexandros, and Leonidas. 

Stars and galaxies make up much of the universe’s ordinary, observable matter. But for decades, scientists have wrestled with a cosmic conflict: There should be much more. 

Physicists have good estimates of how much matter was present in the early universe. Shortly after the Big Bang, roughly 83 percent of all matter in the universe was composed of invisible dark matter, with ordinary matter making up the rest. And yet, these estimates exceed the amount of ordinary matter seen in stars and galaxies today. Where, then, did all the missing ordinary matter go? 

Now MIT scientists, as part of the CHIME/FRB Collaboration, are using far-off radio signals to reveal missing matter in the vast space between galaxies. The team has developed a new method to search out missing matter by combining locations of galaxies with detections of fast radio bursts. 

A fast radio burst, or FRB, is an ultrabright, millisecond flash of radio waves emitted by extremely energetic phenomena in the distant universe. As it travels through space, the signal from a fast radio burst gets stretched, or “smeared,” in time. The more missing matter that it passes through, the more smeared the signal becomes. 

The MIT-led team measured the degree of smearing experienced by thousands of FRB signals detected on Earth. Then they compared each FRB smear with locations of galaxies across the universe to determine how much of an FRB’s smearing was due to galaxy matter versus other, missing matter. 

The new method revealed not only whether missing matter was present, but also where. Specifically, the researchers discovered that it exists in very diffuse clouds surrounding groups of galaxies. These clouds extend out from the galaxies, to much further distances than scientists had predicted. 

“We find that, overall, where there are more galaxies, there tends to be more missing matter around them,” says Haochen Wang, a graduate student in MIT’s Kavli Institute for Astrophysics and Space Research.

The results, reported today in the journal Physical Review Letters, support the idea that matter is flung outside a galaxy through black hole jets, exploding stars, and other highly energetic processes within a galaxy. What’s more, the findings suggest that such processes are more energetic than scientists had thought. 

“We’re finding missing matter that is pushed out to larger scales,” says Kiyoshi Masui, associate professor of physics at MIT. “These measurements indicate that star activity, and activity from black holes, is stronger and much more violent than predicted.”

Masui and Wang are co-authors of the new study, which includes Shion Andrew, Adam Lanman, Kenzie Nimmo, and Ryan Raikman from MIT, and collaborators from multiple other institutions as part of the CHIME/FRB Collaboration. 

The shape of matter

The vast majority of ordinary, observable matter in the universe is built from baryons — a type of subatomic particle that includes protons and neutrons, and that makes up most of an atom’s mass. Scientists estimate that just 17 percent of the early universe was made from this “baryonic” matter, shortly after the Big Bang. 

Some of that early matter was forged into every substantial thing we see today, from planets, stars, and galaxies, to our own bodies. But as scientists have realized, this matter doesn’t quite add up. The total mass of all the stars, galaxies, and galactic clouds is about a tenth of the baryonic matter that existed in the early universe. There must be more matter, likely in the spaces between galaxies. But the universe is vast. Any leftover matter likely exists at extremely low densities, of around a single proton per cubic meter, making it extremely challenging to detect.  

Recently, however, Masui and others have found that such missing matter could be sussed out using fast radio bursts. FRBs were first discovered in 2007, and since then astronomers have detected several thousand of the mysterious, ultrashort signals from distant galaxies, billions of light years away. 

“What makes FRBs good to probe missing matter is that they have a special property,” Wang says. “They start out as a very quick flash, and as they pass through matter, they smear out in time. And we can measure that smearing very precisely, which is directly proportional to how much missing matter the FRB passed through.”

Researchers have previously taken advantage of this smearing property of FRBs to detect missing matter around galaxies. These efforts have confirmed that tenous clouds exist in the vast spaces between galaxies. Masui and Wang wanted to go a step further. 

“We’re not just probing if the gas is with the galaxy or not, but we are seeing the shape of the missing matter that’s around the galaxies,” Wang says. “By mapping the shape of missing matter, we can understand how galaxies form and how they interact with their environment.”

Galactic fountains

For their new study, the team mapped the shape of missing matter around galaxies by cross-correlating thousands of FRB measurements with locations of millions of galaxies. They used data from two sources: the Canadian Hydrogen Intensity Mapping Experiment (CHIME) and the Dark Energy Spectroscopic Instrument (DESI) survey. 

CHIME is a large radio telescope located in British Columbia, Canada, that is designed to scan the entire northern sky for incoming radio waves. The telescope is sensitive to ultrashort, ultrabright radio signals, and since it began observing, CHIME has detected about 4,000 fast radio bursts across the sky. 

DESI is an instrument that is mounted on the Mayall Telescope at Kitt Peak National Observatory, near Tucson, Arizona. The instrument makes detailed measurements of the light coming from over 30 million galaxies, to provide estimates of dark energy — the mysterious force that drives the expansion of the universe. 

From CHIME’s catalog of detections, members of the CHIME/FRB collaboration analyzed 2,870 FRB signals. Each signal is a burst of radio waves, at multiple wavelengths, from highest to lowest energy. The higher-energy “blue” waves typically are less affected by any missing matter they travel through, and therefore should arrive at a detector before lower-energy “red” wavelengths, which are more delayed, or “smeared,” in time. 

The team measured the smearing of each FRB’s various wavelengths, which they could then directly relate to the amount of matter that the FRB must have traveled through before reaching CHIME’s detectors. Masui and Wang then correlated these measurements with the locations of over 6 million galaxies provided by DESI data. In this way, they could look for an association between the missing matter and the galaxies, and measure where one is in relation to the other. 

Their analysis revealed a pattern: Missing baryonic matter tended to be found around galaxies and galaxy clusters. But rather than gathering close to galaxies in a dense ball, missing matter was scattered across a large radius, similar to a diffuse puff. 

“A galaxy is maybe a few 100,000 light years across, and we found missing matter out to about 4 million light years,” Masui says. “That’s further than the simulations predict, by quite a bit.”

“We are finding that the activity in galaxies is messier than we thought,” Wang says. “They’re more like fountains, and really push out gas to very large distances.”

The new results show that fast radio bursts can be a reliable method by which to search for missing matter. As CHIME continues to detect more FRBs, the team says its method can only improve.

“We got it to work for the first time, and will get it to work even more precisely as data gets better,” Masui says. 

CHIME and CHIME/FRB are supported by the Canada Foundation for Innovation, the Natural Sciences and Engineering Research Council of Canada and, the provinces of British Columbia, Québec, and Ontario. This study was supported in part by the U.S. National Science Foundation.