Waves that clear waste from the brain during sleep are impaired in older adults

As people age, they may sleep for shorter periods and wake up more frequently during the night. In addition to those disruptions, MIT researchers have now discovered that a process responsible for flushing waste from the brain during sleep is commonly impaired in older people.

The researchers found that waves of cerebrospinal fluid (CSF), which help remove debris from the brain, are significantly reduced in older adults. This impairment is linked to a decrease in the slow brain waves that normally occur during deep sleep, the researchers found, and is not just a result of getting less sleep.

The researchers now hope to explore whether the disruption of CSF flow increases the risk of neurodegenerative disorders, and if any of the sleep patterns they observed might offer a way to predict an individual’s risk.

“We know that people age really differently. Some people age very successfully with great cognitive outcomes, and other people develop cognitive decline. If we can figure out who has great trajectories of aging and what are the factors that are associated with that, we might be able to use that to predict future outcomes,” says Laura Lewis, the Athinoula A. Martinos Associate Professor of Electrical Engineering and Computer Science, a member of MIT’s Institute for Medical Engineering and Science and the Research Laboratory of Electronics, and an associate member of the Picower Institute for Learning and Memory.

Her lab also plans to investigate whether there might be any interventions that could improve CSF flow in older adults.

Lewis is the senior author of the study, which appears today in Nature Aging. Sydney Bailes, a postdoc in the MIT Laboratory for Information and Decision Systems, and Stephanie Williams, a former visiting student at MIT who earned her PhD from Boston University in 2026, are the lead authors of the paper.

Smaller waves

During the deepest stages of sleep, also called non-REM sleep, high-amplitude, low-frequency waves contribute to physical recovery, memory consolidation, and many other functions. In a 2019 study, Lewis found that these slow waves are also critical for generating waves of CSF that help flush waste from the brain.

That finding led her to wonder whether reductions in slow waves in older people might also reduce CSF flow. Previous studies of aging have yielded conflicting evidence; some found that CSF flow goes down in older adults, while others found it goes up, but these studies did not monitor flow during sleep.

In the latest study, the researchers recruited volunteers in two age groups: 18 to 39 and 60 to 84. The tests lasted for about two hours and began late at night, so that participants would be sleepy. 

The researchers used traditional fMRI to measure participants’ blood flow in the brain, along with a modified MRI technique, developed by Lewis, that measures CSF flow. While the subjects were in the scanner, the researchers also measured their brain waves using EEG. 

Many of the participants fell asleep in the scanner, allowing the researchers to compare their brain activity while awake and while sleeping. 

As they had seen in earlier work, the researchers found that when the younger adults fell into non-REM sleep, their EEG slow waves became stronger, and their CSF waves became larger. However, in the older group, the average CSF waves were significantly smaller than those seen in the younger group. 

Impaired flow

In additional experiments done while subjects were fully awake, the researchers tried to figure out what was causing the age-related declines in CSF flow.

First, they explored the possibility that changes in a phenomenon known as neurovascular coupling might play a role. When localized brain activity increases the demand for oxygen, nearby blood vessels dilate and deliver more oxygen and nutrients. That increase in blood flow can be measured with fMRI.

The researchers tested each subject’s neurovascular coupling by measuring blood flow while the subjects looked at a flickering checkerboard image — a task that typically drives an increase in neural activity, along with an increase in blood volume.

In many of the older subjects, while their neural activity did go up, the blood flow didn’t increase much, suggesting that the blood vessels’ ability to respond to increased oxygen demand has diminished.

During another task, the researchers tested whether the brain’s vascular system might be impaired in the older subjects. They asked the subjects to hold their breath, then measured what happened inside the brain. Normally, this leads to an increase in carbon dioxide levels in the blood, which triggers an increase in cerebral blood flow. However, in the older participants, the average blood flow increase was smaller than the increase seen in the younger subjects.

These two factors — reduced neurovascular coupling and diminished blood flow — accounted for much of the decline in CSF flow but could not explain all of it, according to a statistical model developed by the researchers. This suggests that there may be other age-related factors that contribute to the reduction in CSF flow, they say.

“Aging is so complex, and so many different things change, that it’s really hard to disentangle the individual factors,” Bailes says. “We try to account for some of the mechanisms, but there definitely are a lot of other factors related to vascular health and lifestyle. If you were able to measure all of them, you might be able to break it apart into different explanations for why an individual’s CSF flow might be more or less impaired.” 

Enhancing CSF flow

Previous studies have found that sleep disturbances are linked to a greater risk of neurodegenerative disorders such as Alzheimer’s and other forms of dementia.

None of the subjects in this study had been diagnosed with a neurodegenerative disorder, but the researchers found that some of the older adults showed signs of gray matter atrophy, which was correlated with a greater decrease in CSF flow. They hope to continue studying some of the participants in hopes of discovering whether changes to CSF flow are associated with a higher risk of developing a neurodegenerative disorder.

They also hope to test possible interventions that might help increase CSF flow. In a recent study, Lewis’ lab found that short bursts of “pink noise” during sleep could increase CSF flow in younger adults, but they have not yet tested that intervention in older adults. 

In other work, they plan to study why some older adults don’t experience a dropoff in CSF flow. In this study, the researchers found that while as a group, older adults showed weaker CSF waves during sleep, some individuals in the group maintained normal CSF waves.

“It’s really intriguing that for some of these older adults, their CSF fell right in the range of the healthy young adults,” Lewis says. “We are really interested in what’s different in those people and how that might be related to future health outcomes.”

The research was funded by the National Institutes of Health, the Simons Collaboration on Plasticity in the Aging Brain, a McKnight Scholar Award, a Pew Biomedical Scholar Award, an MIT EECS Transformative Research Award, the Picower Institute Neurological Disorder Discovery Research Fund, a Sloan Fellowship, and the Department of Defense National Defense Science and Engineering Graduate Fellowship Program.

School of Engineering welcomes new faculty in 2026

The MIT School of Engineering welcomes new faculty members across six of its academic units in 2026. This new cohort of faculty members, who have recently started their roles at MIT, conduct research across a diverse range of disciplines.

“I’m pleased to welcome these talented colleagues to our engineering faculty,” says Paula T. Hammond, dean of engineering and an Institute Professor. “They bring fresh perspectives that will open new directions for research and create opportunities for our students to learn through discovery. I look forward to supporting them as they pursue ambitious ideas, build collaborations across MIT, and help our students grow, both in the classroom and in the lab.”

Faculty with appointments in the Department of Electrical Engineering and Computer Science (EECS) and the Institute for Data, Systems, and Society (IDSS) report into both the School of Engineering and the MIT Stephen A. Schwarzman College of Computing.

The new engineering faculty include:

Andrew Akerson joined the Department of Mechanical Engineering as an assistant professor last January. His research interests include optimal design, computational solid mechanics, and material characterization. He previously worked as a research scientist at Meta’s Reality Labs Research. Akerson received his BS and MS in aerospace engineering and mechanics from the University of Minnesota and his PhD in mechanical engineering from Caltech.

Jake Hecla ’17, the Charles Stark Draper Career Development Professor, joined the departments of Aeronautics and Astronautics and Nuclear Science and Engineering as an assistant professor in July 2026. His current research interests focus on aerospace nuclear power and propulsion, heat rejection in space, and long-standoff characterization techniques for space nuclear systems. Previously, Hecla was a Stanton Postdoctoral Fellow in the MIT Laboratory for Nuclear Security and Policy. He holds a BS in nuclear science and engineering from MIT and an MS and PhD in nuclear engineering from the University of California at Berkeley.

Zachary Manchester, the Boeing Career Development Professor, joined the Department of Aeronautics and Astronautics as an associate professor in January. He is interested in building autonomous systems that match, and eventually surpass, the levels of performance, agility, and reliability of humans. He previously taught in the Robotics Institute at Carnegie Mellon University. Manchester received his BS in applied physics, his MEng in mechanical engineering, and his PhD in aerospace engineering from Cornell University.

Eric Sun joined the Department of Biological Engineering as an assistant professor in January. He develops new computational and machine learning models to study complex biology across multiple scales with applications in aging, rejuvenation, and immune interactions. Sun is also a member of the Ragon Institute of Mass General Brigham, MIT, and Harvard. He obtained his BS in chemistry and physics and his MS in applied mathematics from Harvard University and his PhD in biomedical informatics from Stanford University.

Haoning Tang joined the Department of Mechanical Engineering as an assistant professor in January. She conducts interdisciplinary research at the intersection of nanophotonics, optical instrumentation, and device engineering, with a focus on metamaterials, quantum materials, and MEMS-enabled platforms. She was previously an adjunct professor in the electrical engineering and computer science department at the University of California at Berkeley. Tang received her BS in electrical and electronics engineering from Hong Kong University of Science and Technology and her PhD in applied physics from Harvard University.

Stylianos Varchanis, the Raymond A. (1921) and Helen E. St. Laurent Career Development Professor, joined the Department of Chemical Engineering as an assistant professor in July. He develops theoretical and computational models to understand fundamental transport processes in complex fluids, soft matter, and living systems, with relevance to materials engineering, energy, and biotechnology. Before joining MIT, Varchanis was a Flatiron Research Fellow at the Flatiron Institute of the Simons Foundation in New York. He received his BS, MS, and PhD in chemical engineering from the University of Patras in Greece.

Hengyun Zhou ’17, the Thomas D. and Virginia W. Cabot Professor, joined the Department of Electrical Engineering and Computer Science as an assistant professor in March. He is developing full-stack architectures for resource-efficient, fault-tolerant quantum computing. His work spans quantum error correction, algorithms and compilation, and hardware architecture. Zhou previously worked as a research scientist for QuEra Computing Inc. He holds a BS in physics and mathematics from MIT and a PhD in physics from Harvard University.

The School of Engineering is also welcoming Edward “Jas” Bordeau Brooks, the KDD Career Development Professor in Communications and Technology, who joined the Department of EECS as an assistant professor in September 2026. For more information about Brooks, read the Schwarzman College of Computing’s recent article. 

The School of Engineering also remembers Thomas Rose, MIT’s POSCO Professor of Materials Science and a member of MIT’s Center for Materials Research in Archaeology, who joined the Department of Materials Science and Engineering’s faculty in January. Rose died on Sept. 2 following a bicycle accident. His research focused on ancient metallurgy and pyrotechnology that shaped early human societies, including how copper and its alloys were produced, transformed, and circulated. The work integrated geochemistry, mineralogy, experimental archaeology, and materials science.

MIT Schwarzman College of Computing welcomes seven new faculty for 2026

The MIT Schwarzman College of Computing welcomes seven new faculty members in core computing and shared positions to the MIT community in 2026. They’re inventing and using artificial intelligence and computing technologies in new ways to transform areas such as environmental science, quantum computing, human-AI interaction, and macroeconomics and finance.

“It’s a pleasure to welcome these outstanding faculty members to MIT, whose innovative research is leading advances in AI and computing and its impact on society,” says Dan Huttenlocher, dean of the MIT Schwarzman College of Computing and the Panasonic Professor of Electrical Engineering and Computer Science.

These hires include core faculty with appointments in the Department of Electrical Engineering and Computer Science (EECS), which reports jointly to the MIT Schwarzman College of Computing and School of Engineering, as well as shared faculty positions between the college and other departments, including Urban Studies and Planning, Mathematics, Economics; and, at MIT Sloan School of Management, the Finance Group and the Work and Organization Studies Group.

The new shared faculty include:

David Bruns-Smith, an assistant professor at the MIT Sloan School of Management, who holds an MIT Schwarzman College of Computing shared position with EECS, joined MIT in September. He is also a member of the Laboratory for Information and Decision Systems. His recent research has focused on debiased machine learning with applications in macroeconomics and household finance. Previously, he was a postdoc at Stanford Data Science, and he completed his PhD in computer science at the University of California at Berkeley.

Zana Buçinca, an assistant professor at the MIT Sloan School of Management, who holds an MIT Schwarzman College of Computing shared position with EECS, joined MIT in September. She is also a member of the Laboratory for Information and Decision Systems. Her research interests lie at the intersection of human-AI interaction and the future of work, and involve designing human-AI interaction techniques that complement people and amplify their values in AI-assisted work. She was a postdoc at Microsoft in the Office of Applied Research, and received her PhD in computer science from Harvard University.

Alexandra Cabral, an assistant professor in the Department of Urban Studies and Planning, who holds an MIT Schwarzman College of Computing shared position with the Institute for Data, Systems, and Society (IDSS), joined MIT in September. Her research focus includes low-cost sensor networks, smart cities, environmental informatics, and low-resource computing. Previously, she was a postdoc in the Ka Moamoa Lab at Georgia Tech and served as a 2025-26 NSF ICICLE Educational Fellow. She earned her PhD in computer science from Harvard University.

Alexandru Damian, an assistant professor in the Department of Mathematics, who holds an MIT Schwarzman College of Computing shared position with EECS, joined MIT in September. He is also a member of the Laboratory for Information and Decision Systems. His work is focused on developing a predictive, and ultimately prescriptive, theory of deep learning. He holds a BS in mathematics from Duke University and a PhD in applied and computational mathematics from Princeton University.

Lindsey Raymond PhD ’24, an assistant professor in the Department of Economics, who holds an MIT Schwarzman College of Computing shared position with EECS, joined MIT in July. She is also a member of LIDS. Her research involves the effects of digital technologies and artificial intelligence on labor markets, firms, and market competition. She was a postdoc in economics and computation at Microsoft, and earned her PhD from MIT Sloan.

The new core faculty are:

Jas Brooks, the KDD Career Development Professor in Communications and Technology, joined EECS as an assistant professor in September. They are also a member of the Computer Science and Artificial Intelligence Laboratory. Their work focuses on perceptual engineering: the design of computational interfaces that sense and modulate human sensory perception. Previously, they were a Rising Star in EECS, a Siebel Scholar, and a National Science Foundation Graduate Research Fellow. They hold a BS and PhD in computer science from the University of Chicago. 

Hengyun (Harry) Zhou ’16 joined EECS as an assistant professor in March. He is also a member of the Research Laboratory of Electronics. His research focuses on accelerating the development of large-scale, fault-tolerant quantum computers through full-stack architectural innovation. Before coming to MIT, he previously served as the Quantum Error Correction Architecture Lead at QuEra Computing. He holds a BS in physics and mathematics from MIT and a PhD in physics from Harvard University.

Helping robots fit into the human workplace

Despite fears about robots replacing humans at work, there are still plenty of things humans can do far better than machines. That’s fine with Robust AI, a startup building robots designed to complement what humans do best.

The company’s mobile robot, named Carter, resembles a utility cart on wheels and is designed to fit seamlessly alongside humans in fast-changing warehouse and factory environments. Carter features a force-sensitive handlebar for easy guidance, an AI-powered vision system, and holonomic wheels for instant movement in any direction.

As a result, humans can cut Carter off, push and pull it in any direction with one hand, and direct it with gestures or the onboard touch screen.

The company’s robots are currently deployed in logistics warehouses and on factory floors, where they help humans while picking or stocking shelves and moving items across facilities autonomously.

“In these environments, what really determines success is how the overall system performs,” explains Anthony Jules ’92, SM ’92, who co-founded the company with MIT Professor Emeritus Rodney Brooks. “The robot does some of the work, the people do some of the work, and the backend systems do some of the work. It’s that combination that determines how well your system performs. We believe it’s really important that people have an appropriate level of agency and input so that they’re always moving the system forward.”

Collaborative robots

Brooks was a faculty member at MIT from 1984 until 2010 and served as the director of the Computer Science and Artificial Intelligence Laboratory (CSAIL) from 1997 until 2007. While at MIT, he co-founded iRobot, which invented the popular Roomba vacuum and went public in 2005. Later, he co-founded the company Rethink Robotics, which also designed collaborative robots.

Jules met Brooks as a graduate student at CSAIL in the 1990s. The two never worked together, but their labs were on the same floor. The pair kept in touch after Jules graduated and co-founded the company Sapient, which went public before being acquired by Publicis in 2015.

In 2019, Brooks bumped into Jules at a robotics conference. He told Jules he was thinking of starting a robotics software company, which he compared to a video game engine for training robots. Robust AI was founded later that year.

The Covid-19 pandemic caused the founders to shift to creating robots that could autonomously disinfect surfaces, but when it was discovered that Covid-19 was not transmitted via surfaces, they had to pivot again.

“That forced us to decide how to use this technology we had been building, which was really about helping robots understand the world more fully,” Jules recalls. “Long story short, I said to Rodney, ‘I have an idea. You’re going to love it and you’re going to hate it. You’re going to love it because it’s solving problems for customers that we’ve been hearing about, and you’re going to hate it because we have to build hardware.”

The founders had been interviewing potential customers for years through Robust AI’s early pivots. They finally decided on logistics and manufacturing as the industries to target early on.

“When people think about automation in factories, they think about pictures of the 5 percent that are fully automated with $50 to $200 million dollar systems,” says Jules, who now serves as Robust AI’s chief innovation officer. “They don’t think about the vast majority of factories that aren’t automated at all. These factories don’t even have a conveyor belt. They are not going to tear everything out and replace it with a $200 million system.”

Robust AI’s robots can be phased into operations as needed and offer several workflows for customers. In one workflow, the robot can guide human workers to specific spaces to retrieve items. In another it can follow humans to important spots for loading. The primary workflows involve Carter robots moving things from one place to another, often in large warehouses or factories, where they say human workers can walk on average 30,000 steps a day.

“These are huge spaces, so a person might load the robot with things and then need to bring them to another station, but that could be 400 feet away,” says, Brooks, Robust AI’s chief technology officer. “Now the person can just say ‘done’ and the robot brings it instead. How the robots are integrated into the workspace is entirely up to the customer.”

Getting work done

Leila Takayama, Robust AI’s vice president of human robot interaction, recently discovered a peer-reviewed study of robots in hospitals, which were used to do things like gather dishes, move sheets, and transport other equipment. The researchers discovered that, in many hospitals, a large number of robots were turned off and against the walls. The nurses said they broke.

“What really happened was the nurses were rushing with a patient on a gurney, and the robot sees them in the hallway and stops dead, and they couldn’t get the gurney around them,” Brooks says. “The robots were getting in the way of nurses’ primary tasks.”

Robust AI’s approach to technology development is to understand workers’ primary tasks deeply and build features that accelerate those tasks, whether a human or a robot is doing them.

“We let humans do what humans are good at in the physical world, which we’ve had millions of years of evolution to get good at, and have the digital system respond to that and provide us with all the things that it is good at,” Brooks explains.

MIT-led team receives funding to develop next-generation alternatives to modern weed killers

MIT researchers will lead an up to $53 million effort, awarded by the U.S. Advanced Research Projects Agency for Health (ARPA-H) as part of its STREAM (Systems for Tracking and Resilience in Efficient Agricultural-input Management) program, to develop next-generation weed killers that help farmers and protect human health. 

More than 1 billion pounds of herbicide are used in the United States every year. Herbicides help farmers by preventing significant crop loss due to weeds, and they enable regenerative farming practices that protect the soil, but some weeds are becoming resistant to the most common herbicides. 

Exposure to herbicides may contribute to increased risk of cancer, metabolic and endocrine disorders, and other adverse health outcomes. Farm workers can be exposed from frequent herbicide use, but many U.S. residents are also exposed to these chemicals in small amounts through food and drinking water. 

The MIT-led research team will develop new generative artificial intelligence tools to help them discover herbicide alternatives that are more effective and developed with human health in mind. Such alternatives could also help to reduce U.S. dependency on imported phosphorus, which is a critical ingredient in a major herbicide, glyphosate.

“With herbicides like glyphosate and atrazine, there’s a real crisis because of resistance mutations that are coming,” says Christopher Voigt, head of MIT’s Department of Biological Engineering. “And if you combine the resistance mutations with issues around national security of the supply chain of the precursors to herbicides, we could find ourselves without these core agrochemicals that are absolutely essential for getting the high crop yields and economical prices that the United States now sees.”

Voigt is the principal investigator for the project, known as Precision Learning for On-target Weed control (PLOW). The team is also expected to include researchers from the Broad Institute of MIT and Harvard, the University of North Carolina at Chapel Hill, the University of Tennessee at Knoxville, the University of Illinois at Urbana-Champaign, and Robigo.

The researchers are grouped into three teams that will tackle different aspects of the project, aligned with the STREAM Technical Areas. The generative AI group will develop models that can predict molecules — either chemical compounds or small interfering RNA (siRNA) — that disrupt the same plant enzymes targeted by glyphosate and other herbicides.

Glyphosate kills weeds by interfering with an enzyme called EPSP synthase, which plants need to produce several critical amino acids. After many years of herbicide use, some weeds have begun to acquire genetic mutations that allow them to become resistant to those herbicides, which makes them less effective and poses a significant threat to crop yields.

“There’s a real need for new herbicides because of resistance mutations that have been arising,” Voigt says. “There are also concerns about herbicides’ possible impacts on the environment and on humans, so it’s important to identify alternative ways of making herbicides, which as a general class of molecule are really critical to agriculture.”

The MIT-led team plans to design generative AI models that will be able to predict molecules that can help prevent resistance from arising. The models will also optimize for other features such as low cost of synthesis, low risk of inducing resistance, and low risk of harming humans or causing ecological damage. 

The group developing these AI models is led by Connor Coley, the Warren K. Lewis Associate Professor of Chemical Engineering at MIT; Regina Barzilay, the School of Engineering Distinguished Professor for AI and Health at MIT; and Alexander Tropsha, the K.H. Lee Distinguished Professor in the UNC Eshelman School of Pharmacy.

The models are expected to generate billions of possible candidates, which will then be tested by a second research team. This high-throughput screening group will be led by Angela Koehler, the Charles W. and Jennifer C. Johnson Professor of Biological Engineering at MIT, and Jaime Cheah, the associate director of scientific outreach and collaborative screening in the Center for the Development of Therapeutics at the Broad Institute.

“Those libraries will be screened both to identify those molecules that hit their target, but also to then provide all of that screening data back to the computational approach in order to improve it,” Voigt says.

Once the candidate pools are narrowed down, researchers on a third team will begin greenhouse studies to evaluate how each molecule performs under different agricultural conditions. That team will be led by Neal Stewart and Scott Lenaghan, co-directors of the Center for Agricultural Synthetic Biology at UTK; Patrick Tranel, the Tom and Peggy Ainsworth Professor of Crop Sciences at the University of Illinois; Tim Willson, the Harold Kohn Distinguished Professor at the UNC Eshelman School of Pharmacy; and Andee Wallace PhD ’20, CEO and founder of Robigo. 

The first greenhouse tests of initial RNAi candidates are expected to be complete by the end of the first year of the program.

As part of this project, the research team will also develop living microbial systems that degrade agrochemicals before they reach food, soil, or waterways. Engineered bacteria, fungi, and srRNA will convert chemicals and formulation ingredients into biomass, providing a scalable strategy to reduce environmental and human exposure.

This research is funded by the Advanced Research Projects Agency for Health (ARPA-H) under agreement number 140D042690030. The views and conclusions contained in this article are those of the authors and should not be interpreted as representing the official policies, either expressed or implied, of the United States government.

A single AI model simulates the behavior of many substances, from sliding sand to folding proteins

Today, a scientist who wants to simulate a physical process must pick the right tool for that one job, such as a solver built for flowing water or another designed to model the way a protein folds. Each one is hand-coded around the specific equations that dictate how that process behaves.

Researchers at MIT, the University of British Columbia (UBC), and elsewhere have now replaced that patchwork with a single deep-learning system, called WorldParticle, that could help engineers model different physical phenomena, from sliding sand to folding proteins, with one unified artificial intelligence architecture.

Scientists and engineers could use this system to model processes that typically require separate simulators, such as a fluid sloshing against a deformable solid. This could offer a faster, more accurate way to simulate how sloshing oil might impact an elastic valve in a storage tank, for instance. It could also help engineers study where and how granular materials, like sand or soil, could move when a soft robot touches them.

WorldParticle has been extensively validated across diverse phenomena, from microscopic protein motion to macroscopic physical simulations. Pictured are animations showing molecular, fluid, and granular simulations.

Image credits: courtesy of the researchers

The system represents every material it simulates, from a dune of sand to a protein molecule, as tiny particles and predicts where those particles will move one step at a time. Because it learns the underlying dynamics rather than being programmed with an equation, the model can adapt to materials, forces, and environments it never encountered during training.  

In the long term, the researchers hope this work helps scientists discover new materials and explore physical processes they would be unable to effectively study with existing simulation methods.

“We want to have a GPT moment for simulation,” says Peter Yichen Chen, a former MIT postdoc who is now an assistant professor at UBC, and is a co-corresponding author of an open-access paper on this technique. “Every time people want to build a simulator for a particular physical phenomenon, they need to code it up from scratch. Our vision is to have a foundation model that works out of the box for any scenario you want to simulate. This is a first step in that direction.”

Chen is joined by co-corresponding and co-lead author Minghao Guo, an MIT postdoc; co-lead authors Caoliwen Wang and Siyuan Chen, UBC graduate students; Wojciech Matusik, an MIT professor of electrical engineering and computer science who leads the Computational Design and Fabrication Group within the Computer Science and Artificial Intelligence Laboratory (CSAIL); and others at UBC, Georgia Tech, Inria, Meta, the University of Pennsylvania, the University of Utah, the University of California at Los Angeles, and the University of Hong Kong. The research will be presented at the ACM SIGGRAPH Conference and Exhibition on Computer Graphics and Interactive Techniques in Asia.

Unified simulation

For this work, the researchers drew inspiration from world models. These powerful AI systems learn the rules that govern the physical world so they can predict how objects or environments will change over time in response to certain actions or conditions.

The researchers applied these concepts to develop a unified AI architecture that can simulate multiple types of physical phenomena. 

“If you study deformable objects and fluids, you would need two simulators — one for each physical regime. But if you unify them into one model, maybe you could couple the physics of these two processes and gain insights into new problems?” Guo says.

WorldParticle breaks an object into thousands or millions of tiny, three-dimensional dots, or particles, using what is known as a Lagrangian representation. It tracks the movement and velocity of the particles over space and time to simulate the material’s dynamics.

Typically, when scientists model materials and physical processes using Lagrangian particles, they use a governing equation that defines the fundamental laws which control how that material behaves over time and space. They must convert that complex equation into a formula a numerical solver can handle, repeating these steps for each material and process they want to model.

Instead, the researchers designed a neural network that learns to solve governing equations in general, eliminating the need to design a new model and solver to handle each one. 

Once WorldParticle converts an object into particles, a neural network takes the 3D positions and physical properties of the particles as inputs. It outputs their positions and properties for the next frame of the simulation, predicting the material’s motion over time and space in one-second increments. 

“It is a similar architecture to ChatGPT. But instead of predicting one word at a time to write a piece of text, we are predicting how these particles will move, one step at a time,” Chen says.

Before each simulation step, WorldParticle separates the external forces that affect the physical system, like impulse forces that affect an ocean wave when it slams into a seawall. 

After predicting the initial states of the particles, WorldParticle integrates the external forces back into the simulation, correcting its particle predictions based on how those forces affect the object’s behavior. In this way, the system can generalize to unseen materials, changing environmental conditions, and varied forces.

A general, scalable framework

But because a complex object could be broken into millions of particles, the problem quickly becomes too computationally intensive for the model to handle. 

To streamline processing, the researchers employed a merging technique to combine and compress particles while still allowing WorldParticle to capture fine-grained details of their behavior.

“We needed to do a lot of detailed engineering and employ many tweaks to the architecture for it to be truly generalizable and scalable,” Guo says.

They demonstrated WorldParticle by modeling six different physical phenomena, including deformable objects, granular materials, and proteins, with higher accuracy than numerical simulation methods. 

For instance, WorldParticle effectively captured how a granular object would collapse into a pile while some baseline methods left sand particles floating in the air. Their method also modeled how protein structures change over time about 100 times faster than numerical simulators, Guo adds. 

The general-purpose framework could be applied to many materials and processes, including soft robot manipulation tasks, RNA evolution, and astrophysical phenomena like the merging of galaxies, the researchers say.

Moving forward, they want to adapt WorldParticle to simulate these and other physical systems. They also want to build on this work toward the development of a full, general-purpose simulation pipeline for scientific discovery. 

 “We see the unification of computation and science giving us new insights and tools to discover physical phenomena or explain some unexplainable physical phenomena,” Guo says. “With a unified model, we could unlock many discoveries that we can’t even imagine.”

New system gives users more control over their online experience

A user who is frustrated with changes a social media company made to their favorite messaging platform, like the introduction of an annoying AI chatbot, can’t do much to address the situation, short of leaving the platform.

Now, MIT researchers are offering an alternative path.

They’ve created Social.Wiki, a system that allows users to easily edit any social application made within Social.Wiki, and choose which of the edits made by others they want to see. 

For example, a user could change a messaging site on Social.Wiki to remove an AI chatbot and instead add a restaurant voting feature so their group chat can finally decide where they are going out for dinner.

“The infrastructure that we are all dependent on online is very rigid. It is based on these one-size-fits-all applications that we don’t have a lot of power to change. This project is about trying to imagine what a different web would look like where anyone can participate,” says Theia Henderson, an electrical engineering and computer science (EECS) graduate student, and lead author of a new paper on Social.Wiki.

The researchers envision a world in which, if a user is unhappy with their social app, they can edit it to change it however they like — without getting cut off from their social connections.

The platform enables users with no coding expertise to create a range of websites, including social sites like photo sharing networks, discussion forums, and dating apps. It handles all back-end work, eliminating the need for creators to set up servers or arrange databases.

Social.Wiki employs a decentralized infrastructure that allows users to retain ownership of their data as they create, edit, and navigate social websites within the platform. It also includes a “data guard” that protects user data from malicious or buggy applications. 

Through case studies and user tests, the researchers demonstrated the breadth of applications people can create with Social.Wiki, including a rideshare app where riders propose the prices they are willing to pay and a collaborative canvas for creating art together.

While Social.Wiki is still in its early days, the researchers plan to continue refining the platform as they deploy it to a wider audience. 

“We want to empower the creation of far more social applications that meet the needs of unusual small groups. Each group is unusual, but everyone is in unusual groups. Everybody has their own needs for social applications that we want to be able to address,” says David Karger, professor of EECS, a member of the Computer Science and Artificial Intelligence Laboratory (CSAIL), and senior author of the paper.

Henderson and Karger are joined on the paper by EECS graduate students Carmel Schare and Ana Dodik; Clemens Klokmose, a professor of computer science at Aarhus University in Denmark; Ziv Epstein, a postdoc in the MIT Media Lab; and David D. Clark, an MIT research scientist. The paper will be presented at the ACM Symposium on User Interface Software and Technology in November.

Letting users choose

Social.Wiki is a Wikipedia-like system in which users can navigate to different web applications or create their own web application within the Social.Wiki ecosystem. Each website has an “edit” button that allows any user to change the content of the site, from the text color on a message board to the pricing algorithm in a ride-sharing app.

Clicking the “edit” button in Social.Wiki opens a page that lets users view and edit the source code manually or prompt an AI assistant to make changes to the site. 

This platform builds on the researchers’ prior work on Graffiti, a web framework with a decentralized infrastructure that makes building personalized applications easier. 

With Social.Wiki, the researchers sought a collaboratively editable system that would allow anyone to revise web applications to be ideally suited for their needs and the needs of groups they belong to. 

To enable this structure, Social.Wiki relies on a strategy the researchers call “plural governance.” Instead of a central moderator deciding which content users will see, each Social.Wiki user can choose the rules that determine which edits they will see on Social.Wiki webpages.

Users can also identify “trusted editors,” so they will see edits from individuals or organizations they consider reliable.

For instance, a user could customize their rules, or “governance policy,” so they only see edits to a site that have been endorsed by at least two of their friends. When the Social.Wiki browser loads a site for a user, it applies the governance policy that user selected to decide which version of the site to present.

For example, based on each user’s choices and edits, the system might show one user a ride-sharing app with a uniform pricing algorithm while another user might see the same app with a dynamic pricing algorithm.

“No website is tied to a particular machine. This means we have no centralized ownership over the websites or the underlying infrastructure,” Henderson explains.

Security and creativity

In this decentralized infrastructure, user data are stored separately from the websites that access it, which helps keep data secure.

Social.Wiki also employs a data guard, which prompts a user to review each request from a website that is attempting to access their data.

“In Social.Wiki, these websites are held in common and anybody can edit them, but the data guard makes sure that each user stays in control of what happens to their data,” Karger adds.

The researchers evaluated Social.Wiki through several case studies. For one, they developed a messaging platform that allows participants to modify the chat’s styling and interface without affecting the experience of others. In another, they built a collaborative ride-sharing app that enables riders to adjust the fare they want to pay using a transparent “living wage” calculator.

They also sponsored a hackathon in which users built Social.Wiki applications.

Participants developed a range of creative sites, including a housing co-op communication tool that organizes chats on a map of the co-op, as well as a garden-themed microblogging site where posts are “planted” as flowers.

“Even the little bit that has been done so far with Social.Wiki demonstrates that there is so much more that could be done with social apps if users could specialize them to meet a particular group’s needs,” Karger says.

Moving forward, the researchers are working to deploy Social.Wiki to a broader audience and gather feedback from more diverse user groups. They also want to make technical upgrades to enable end-to-end encryption on private message data and provide better developer tools.

This work was funded, in part, by Schmidt Sciences and the National Science Foundation (NSF). 

Regina Barzilay and Jim Collins win Pritzker Prize for AI and Science Research Excellence for work in AI-enabled antibiotic discovery

Regina Barzilay, MIT’s School of Engineering Distinguished Professor of AI and Health, and James Collins, MIT’s Termeer Professor of Medical Engineering and Science, are the recipients of the 2026 Pritzker Prize for AI in Science Research Excellence. 

The award recognizes their joint work in “pioneering AI-enabled antibiotic discovery, integrating deep learning, bioengineering, and experimental validation to identify novel antibacterial candidates against drug-resistant pathogens.” 

Established in 2024 by the University of Chicago in partnership with Caltech, the Pritzker Prize for AI in Science Research Excellence recognizes contributions that jointly advance AI and the natural sciences or engineering.

Barzilay and Collins, who are faculty co-leads of the Abdul Latif Jameel Clinic for Machine Learning in Health (Jameel Clinic) and affiliates of the MIT Schwarzman College of Computing and the MIT School of Engineering, published their first breakthrough paper together in 2020 with then-MIT postdoc and first author Jonathan Stokes (who worked in the Collins lab from 2017 to 2021) in Cell. The paper described the first artificial intelligence-discovered antibiotic, called halicin, named for the fictional AI system HAL 9000 in Stanley Kubrick’s 1968 film 2001: A Space Odyssey.

A broad-spectrum antibiotic that was originally studied as a potential diabetes drug, halicin was identified by a deep learning model trained on 2,500 molecules. Upon testing, the research team found that halicin was able to kill antibiotic-resistant bacteria (i.e., “superbugs”) such as Clostridium difficile, Acinetobacter baumannii, and Mycobacterium tuberculosis. A discovery process that once took years was completed in just a few weeks. Amidst the growing alarm around the global antimicrobial resistance crisis, the results demonstrated that AI could prove to be an effective tool for rapidly identifying new and effective antibiotics.

Following halicin, Barzilay and Collins worked with Stokes (now an assistant professor at McMaster University) to use machine learning to identify abaucin in 2023, a narrow-spectrum antibiotic demonstrating effectiveness against A. baumannii, an antibiotic-resistant bacteria commonly found on hospital doorknobs and in the wounds of U.S. soldiers stationed in Iraq and Afghanistan. Notably, abaucin’s “narrow spectrum” efficacy against a single species of treatment-resistant bacteria is its main advantage, minimizing the likelihood of bacteria developing resistance and harming beneficial bacteria in the gut.

“Historically, antibiotic discovery has depended on laborious, trial-and-error strategies,” says Collins. “This process is slow, costly, and unpredictable, often yielding few viable antibiotic candidates. With generative AI, we’re now seeing a shift from AI being used as a tool for discovery to AI being used as a design tool. This shift opens new frontiers in antibiotic discovery that could bring us the next generation of lifesaving drugs.”

In Barzilay’s view, “despite significant progress in molecular modeling, we are still in early stages of drug discovery. With AI, we can only automate parts of the pipeline, rather than totally transforming the discovery process. To make it happen, we need to work really hard to build realistic AI models of human biology.”

MIT-led team receives funding to develop a wearable monitor for early disease detection

MIT researchers will lead a four-year, up to $38 million project, funded by the U.S. Advanced Research Projects Agency for Health (ARPA-H), aimed at developing a noninvasive, wearable patch that continually monitors biomarkers of heart disease and sepsis.

The project is called PYTHIA, which stands for “predictive platform for physiological and therapeutic health intervention and adaptive management.”

According to the project plan, the patch, which will capture clinical-grade health information in real time, will be powered by heat from the patient’s body. It could provide essential information to help clinicians diagnose heart failure and sepsis early, before the diseases have caused irreversible damage. 

The researchers believe the minimally invasive device could also track the progress of a patient’s recovery after they are discharged from the hospital, and assist clinicians in developing more effective therapeutic approaches.

Catching common killers

Heart disease is the number one cause of death in the U.S., while sepsis kills at least 350,000 U.S. adults each year and is the leading cause of death in hospitals, according to the Centers for Disease Control and Prevention.

“We targeted these diseases because they are very lethal, but we also see opportunities for detecting biomarkers related to heart disease and sepsis to offer a solution that could really help improve diagnosis and treatment,” says principal investigator Ahmad Bahai, a professor of the practice in the Department of Electrical Engineering and Computer Science (EECS) and a member of the Research Laboratory of Electronics at MIT.

Bahai will be joined on the project by MIT researchers Joel Voldman, the William R. Brody Professor of EECS; Tim Swager, the John D. MacArthur Professor of Chemistry; Hadley Sikes, the Willard Henry Dow Professor in Chemical Engineering; and Collin Stultz, the Nina T. and Robert H. Rubin Professor of EECS; with additional clinical and industry collaborators.

“We have created an ecosystem of R&D, with MIT at the center of it, to facilitate not only leading-edge research, but also a path toward translation into a device that can be used in the clinic,” Bahai says. “Our goal with this project is to impact the future of personalized health care.”

PYTHIA grew out of the MIT Program for Health Sciences and Semiconductors (HS+S), which Bahai co-founded when he joined MIT three years ago. The program focuses on leveraging advances in semiconductor technology and nanotechnology to invent new, transformative approaches to diagnostics, therapeutics, and drug discovery.

Inspired by his previous successful work on continuous glucose monitors, Bahai conducts research on a wide range of technologies including electrochemical devices, optics, magnetics, and ultrasonics in his “Waves, Bites, and Molecules” lab. His group’s goal is to offer clinical-grade wearable devices that track biomarkers continuously to enable real-time intervention techniques.

The glucose monitor research inspired Bahai and his colleagues to consider other biomarkers they could monitor in real-time using semiconductor technology, which led to the launch of PYTHIA.

The chip-based, wearable patch will monitor heart failure and sepsis biomarkers in interstitial fluid, which is the fluid under the skin. The biomarkers, molecules in blood plasma that are related to the onset of these diseases, are hard to detect because they exist in very low concentrations.

To measure the biomarkers in detail, the system will execute a chemical reaction that binds certain chemicals with the biomarker molecules in the interstitial fluid, making them easier to identify. The advanced microelectronics in the patch will conduct measurements, capturing biomarker data with enough sensitivity for clinical applications. 

Since the device will conduct measurements continuously, amassing a huge amount of data, PYTHIA will incorporate on-chip AI models to translate raw data into useful information for clinicians.

“This is kind of like having the patient give a blood test every 15 minutes or so, but it happens automatically. That is the power of this technology,” Bahai says.

From here, the team will start developing and testing electronic components for the tiny, low-power chip. Then they plan to advance the wearable device through pre-clinical and clinical trials, to the point that it can be commercially scaled and manufactured, Bahai says.

“This project has real potential to improve health care because, in addition to leveraging MIT’s leading-edge science and technology, we are also teaming up with clinical and industrial collaborators with direct experience of the practical constraints, the regulatory environment, and the clinical need,” he says. “The impact of this research could be huge.”

PYTHIA is funded through the ARPA-H Delphi Program led by Program Manager Leonard Tender. Delphi research teams apply a microelectronics design strategy to biosensing in an effort to develop devices that can safely, securely, and continuously monitor the body’s internal health signals. 

The project described here is supported by ARPA-H under agreement number AY2AX000243. The content of this announcement does not necessarily represent the official views of the Advanced Research Projects Agency for Health.

Making machines that can listen

The crack of a bat accompanied by a roar from the crowd can capture a baseball fan’s attention, even if they miss seeing the home run. Humans may be visual creatures, but we also perceive the world around us by listening to it. 

To improve artificial perception, MIT Professor Paris Smaragdis SM ’97, PhD ’01 seeks to make machines that can hear and understand sound even better than humans.

A machine with super-human hearing could rapidly pinpoint the location of a traffic accident by listening for squealing tires and crunching metal, or identify when an industrial drill might break while it is boring through dense bedrock to dig a geothermal well.

“I like to say that I do the same thing the computer vision folks do, but instead I do it for hearing,” says Smaragdis, a professor in the Music and Theatre Arts program of the School of Humanities, Arts, and Social Sciences holding an MIT Schwarzman College of Computing shared position with the Department of Electrical Engineering and Computer Science.

His research group explores areas like machine learning for sound source separation, smart devices that can augment the audio surroundings of humans, and better tools to help people understand, manipulate, analyze, and create music.

“Having had a musical background, I’ve been interested in examining these ideas. What does it take to make machines understand sound? There is a huge domain of applications for machines that can listen,” says Smaragdis, who joined MIT in 2025 and is also part of the Music Technology program and a member of the Research Laboratory of Electronics.

A musical prelude

Smaragdis, who grew up in Greece, began playing classical guitar at a young age. He enjoyed playing traditional pieces but reached a turning point in his musical journey when his father brought home a small synthesizer.

To improve artificial perception, Professor Paris Smaragdis seeks to make machines that can hear and understand sound even better than humans. Image: Jake Belcher

“That little synthesizer had a function that allowed you to make your own sounds, and I became completely captivated by the possibilities,” he says.

Synthesizers were taking a more prominent place in pop music during the 1980s, adding to his fascination with the technology.

As he advanced through school, Smaragdis wanted to study music made with synthesizers. But this was not an educational path offered by the more traditional European conservatories at the time.

Instead, he enrolled at Berklee College of Music in Boston, into a program where he could explore the use of computers and synthesizers to shape and manipulate sound.

“With computers and synthesizers, I could make new sounds that have never been heard before,” he says.

At the same time, Smaragdis became immersed in jazz fusion. He declared bass as his major at Berklee since it would allow him more opportunities to play with jazz ensembles.

“I found the bass to be sonically more interesting than the guitar. It was a more unexplored domain, and I’ve always gravitated toward unexplored domains,” he says.

That interest in the unexplored carried over into his coursework. He and mentor Richard Boulanger, a professor of electronic production and design, created new tools to help people use computers to write, produce, and enhance music. 

Through his studies at Berklee, Smaragdis embarked on a collaborative project with Professor Barry Vercoe (who passed away in 2025) at the MIT Media Lab to create software synthesizers for music creation. Building on that project, Smaragdis pursued graduate studies in Vercoe’s group, where he completed his master’s degree and PhD, followed by a postdoc.

“In the Media Lab, I got exposed more to the computer part of computer music. That’s where my attention shifted a bit. Now I really wanted to make computers that could understand sound,” he says.

Tuning in to unexplored domains

At that time, many students in the Media Lab were using signal processing techniques to help machines handle sound. Smaragdis chose a less explored approach — machine learning — to separate himself from the pack.

He found that training machine-learning models to understand sound offered advantages that couldn’t be attained with traditional signal processing techniques.

“Over time, it became really clear to me that we need to teach computers how to listen. We can’t just tell them how to listen. If anything, we don’t even know how wehumans listen,” he says. “We don’t have a vocabulary to explain what it means to listen, or how to tell the difference between a cat’s meow and the sound of a violin.”

After finishing his postdoc, Smaragdis began a career at Mitsubishi Electric Research Labs, where he worked on artificial intelligence systems that could scan video soundtracks and understand what is happening on screen. 

“What if you record an entire soccer game, but you only want to watch the highlights? In computer vision, it is very difficult to identify what a highlight is. But on the other hand, if you only listen to the sounds, then when the crowd goes crazy you would know that something interesting is happening,” he says.

That system was eventually incorporated into electronic devices, including a personal video recording system Mitsubishi was developing at the time, and into traffic monitoring systems and cell phones.

Smaragdis later joined Adobe, where he created the audio research team that built user-friendly tools to help artists, musicians, and producers manipulate sound without needing a lot of expertise.

“As we did that work, I kept coming up with deeper technical problems that needed to be resolved. I jumped into academia to address that itch,” he says.

Smaragdis joined the faculty at the University of Illinois at Urbana-Champaign, where he and his group incorporated the latest machine-learning techniques into audio-processing systems.

One of the technologies they developed utilized neural networks for separating individual sounds from audio mixtures. That project caught the attention of the team working on the 2021 documentary mini-series “The Beatles: Get Back.”

The production team enlisted Smaragdis to help clean up some archival Beatles recordings.

“That was a very different way of audio processing than what I was doing in academia. Now, I had to do something that sounded good but didn’t necessarily have to look good on paper. We didn’t have a formula we could optimize,” he says.

Every recording involved different challenges. For instance, in one clip John Lennon was playing music while Paul McCartney was horsing around, reading headlines from newspapers and making fun of George Harrison.

“We were trying to come up with a technological solution that could help us do everything, as opposed to addressing each problem on its own,” he says.

His contributions helped the team win an Emmy Award for sound editing for a nonfiction or reality program, in 2022.

A concerted effort

After 15 years at the University of Illinois, Smaragdis decided to join the faculty at MIT to tackle the thorny, interdisciplinary problems that were emerging in the field. 

His group now focuses on two main areas. For one, they develop systems that make computers more useful tools for musicians. 

One project in this area is enabling machines to understand how to listen to music, so they can provide constructive feedback to musicians.

“We want to enhance the toolbox a musician has. We don’t want to use computers to replace musicians,” he says.

On the other side, his group focuses on the fundamental tools scientists and artists alike can use to process, analyze, and manipulate audio.

Smaragdis, who grew up in Greece, began playing classical guitar at a young age. He enjoyed playing traditional pieces but reached a turning point in his musical journey when his father brought home a small synthesizer. Image: Jake Belcher

For instance, his team has recently been exploring algorithms that allow sophisticated signal processing where traditional frames of reference, such as sampling rates, frame sizes, or context-windows, are not fixed — or even defined. 

That implies different computational structures that are yet unexplored, despite their potential for more efficient processing, Smaragdis says.

“That requires rethinking a lot of the basics, but once you have that ability, you can start getting much more flexible and more powerful systems,” he adds.

They also explore ways to incorporate the latest techniques, like large language models, into speech recognition and audio enhancement.

“AI is an immensely powerful tool, but I don’t think we have found the ideal way to use it for sound and music yet,” he says.

He also enjoys mentoring students and helping them materialize their visions for research projects.

Since few of his students have experience using computation in sound, he emphasizes hands-on lessons so they see and hear how to transform equations into something tangible.

Smaragdis still feels a thrill from making music, whether with a computer or a more traditional instrument. Off campus, he tries to play jazz bass whenever he can find the time, and enjoys exploring the rich musical community of Boston with his wife and 15-year-old son.

Outside the musical realm, he plays and coaches underwater hockey — a sport where players wear diving masks and use sticks to push a heavy puck across the bottom of a pool toward their opponent’s goal. He coached the underwater hockey team at the University of Illinois, and is hoping to revive the MIT underwater hockey club soon and start hosting tournaments on campus.

While underwater hockey may be a lesser-known sport, it fits with Smaragdis’ lifelong interest in pursuing the underexplored.

“I’ve always like doing things that were different from what everyone else was doing. If you can create a new sound that no one has heard before, what does that mean about your ability to create new things? I still find these questions completely fascinating,” he says.