AI model decodes the language cells use to communicate

As an embryo develops from a small cluster of stem cells, those once “blank slate” cells begin to take on more specialized roles like brain, liver, or muscle cells, and organize themselves into three-dimensional structures such as tissues and organs.

The fate of each cell — what type of specialized cell it will become — depends on which genes are turned on or off in the cell. These patterns of gene activity shape the cell’s structure and function, enabling it to take on a specific role in the body.

But this decision isn’t up to individual cells. They constantly send and receive chemical signals to and from neighboring cells, which help them understand where they are, what stage of development they are in, and what they should become. These messages spread through multi-step sequences called signaling pathways, which translate external signals into specific changes in gene activity in the cell.

For researchers, being able to retrace the sequence of instructions a cell has received would offer a powerful way to understand how tissues develop and how these processes go awry in disease. 

But this has been difficult to achieve because scientists have long assumed that the effects of signaling pathways vary widely across cell types, meaning they would need to map each pathway separately in each cell type — an arduous and painstaking process.

Now, in a new study led by MIT assistant professor of biology and Whitehead Institute for Biomedical Research member Pulin Li and graduate student Nicholas Hutchins, researchers have discovered that each signaling pathway leaves behind a unique “fingerprint” — a distinctive pattern of gene activity that reflects the particular signals the cell has encountered.

Importantly, these fingerprints are consistent across different cell types for the same signaling pathway, which means that instead of mapping each cell type separately, scientists can reconstruct signaling histories across many cell types using these pathway-specific fingerprints.

This discovery was made possible through a machine learning model called IRIS. This model can detect fingerprints of different signaling pathways and pinpoint which signals a cell received at different stages of development inside an embryo, even for cell types it hasn’t encountered before. 

This artificial intelligence-driven approach marks a major advance over traditional methods, which require researchers to experimentally test every pathway in every possible cell type, and opens up the possibility to comprehensively map the signaling histories of every cell inside a mouse or human embryo at an unprecedented scale.

“Think of voice recognition systems like Siri, which are trained mainly in English, but then use that training to help them recognize other languages,” says Li. “This is called transfer learning, and this is why IRIS can work across many different cell types.”

The researchers’ detailed open-access findings, published in the journal Nature Methods on Sept. 8, could accelerate stem cell engineering for regenerative medicine and improve the creation of organoids — miniature, 3D models that mimic real organs — for studying disease mechanisms and testing new drugs. This is because once researchers learn the pattern of signals that drives a stem cell to become a specific cell type, they can recreate those signals to control the fate of stem cells in a lab or medical setting.

IRIS is a neural network-based model, which, in essence, is an AI-driven system designed to recognize patterns in complex data, similar to how our brains spot patterns. It examines a cell’s overall gene activity and estimates which signaling pathways were likely “on” at specific times during development.

Li and Hutchins trained IRIS on a large experimental dataset that measured how thousands of human embryonic stem cells responded to dozens of combinations of six major signaling pathways at multiple stages of development. This created a comprehensive map, or atlas, of how signaling combinations influence cell behavior. 

The team then tested IRIS using single cells from mouse embryos during gastrulation, a stage when cells are rapidly branching into different fates. IRIS could accurately predict when and where specific signaling pathways would activate in cells destined to become part of heart, gut, muscle, and spinal cord tissue. This variation in signaling molecules is what guides cells to form the right structures in the right places.

With this approach, scientists can not only begin to understand the fundamentals of cell-to-cell communication, but also gain a practical roadmap for guiding stem cells into specific, functional cell types in the lab.

When the researchers employed IRIS to identify signals needed to create a cell type critical for lung development, the model predicted that activating a specific signaling pathway would encourage lung-specific development. Experiments in mouse embryos confirmed the model’s prediction. By identifying the precise combinations of signals that drive lung cell development, they can more reliably generate accurate, lab-grown models of lung tissue.

“In these ways, IRIS is helping us decode the language cells use to talk to each other at a much faster rate than we could realistically achieve through experiments,” Hutchins says.

These improved models would allow researchers to study diseases like asthma, lung cancer, and pulmonary fibrosis, in which lung tissue becomes scarred, often without a known cause. They can then use these models to test potential therapies, and ultimately design regenerative treatments that can repair the damaged tissue.

de MIT News https://ift.tt/AGyvdKW

Satellite data reveal how Antarctica’s ice cliffs behave over time

With rising temperatures, the buttressing ice shelves at the edges the Antarctic ice sheet are melting and calving off at record rates, leaving behind sheer cliffs of solid ice. The stability of these exposed ice cliffs could influence how Antarctica further breaks apart and contributes to sea level rise around the world. 

Scientists have assumed that an ice cliff’s stability depends on its height: Ice cliffs that are taller than 100 meters — about the height of the Statue of Liberty — should collapse under their own weight. And once a cliff sloughs off, it should expose even taller, inland cliffs that collapse in turn. At least, that’s been the idea.

But researchers at MIT and the Woods Hole Oceanographic Institution (WHOI) have found that the stability of an ice cliff depends on more than its height. And even when a cliff topples, it doesn’t necessarily set off a runaway collapse. 

In a study appearing today in Science, the team presents an analysis of over 15,000 ice cliffs in Antarctica that were monitored by satellites over a four-year period. They observed how the ice cliffs and the larger ice sheet evolved over time, and found that those ice cliffs that collapsed did so at a much faster rate than scientists had predicted. But these collapsing cliffs also quickly stabilized, instead of continuing to erode the larger ice sheet. 

“We found that when the cliffs were retreating, they were doing so faster than what has been modeled, which would mean more ice loss and higher contributions of sea level rise,” says study author Caroline Needell, a graduate student in the MIT-WHOI Joint Program. “But because we found that the retreat was short-lived, and all the glaciers eventually stabilized, that would suggest that things won’t be so bad. We don’t see this runaway effect.”

The study’s co-authors are Catherine Walker, assistant scientist at WHOI, and Jeremy Bassis, associate professor at the University of Michigan.

Ice height

In glacial science, ice cliff collapse is described by a process known as marine ice cliff instability (MICI), which assumes that if an ice cliff exceeds a critical height, it will collapse. If then a taller cliff is exposed, it too should collapse, triggering a cascade of instabilities that shrink and erode the ice sheet over time. 

In previous studies, scientists have characterized MICI mainly by modeling the physical properties of ice, scaling these relationships up to the size of an ice cliff and combining these predictions with a few observations of actual ice cliffs in Antarctica. Through these efforts, scientists had determined that ice cliffs become unstable and collapse when they are taller about 100 meters.  

“Predictive ice-sheet models that have been used so far to represent ice cliff instability just take into account cliff height,” Needell notes. “That’s the only variable that matters. And once the cliff height reaches the critical threshold, usually 100 meters, it should initiate this runaway effect of ice sheet retreat.”

But she and her colleagues wondered: How tall do ice cliffs really get? And when they collapse, do they really set off a runaway retreat? 

To answer these questions, they looked to data collected by NASA’s Ice, Cloud, and Land Elevation Satellite-2 (ICESat-2) — a satellite that passes over the Earth’s polar regions and uses a laser altimeter to measure the height of ice sheets, glaciers, sea ice, and the level of the ocean in and around Antarctica. 

The team collected the satellite’s data from 2019 to 2023, which included measurements every three months of over 15,000 ice cliffs. From this dataset, they observed that the average ice cliff was about 30 meters tall — much shorter than the proposed critical height. But they did see a few rare ice cliffs that exceeded 100 meters. 

They also looked for other features that might influence a cliff’s stability. Interestingly, they found that ice cliffs that more or less stayed in place over time also had thicker mélange — a term for the slushy mix of broken icebergs and sea ice that gathers at the front of an ice cliff. 

“We measured mélange height and found taller cliffs have thicker mélange, which could be a built-in stabilizing effect,” Needell explains.

The satellite data also showed that the height of the surrounding water can stabilize a cliff: The deeper the water, the more it can buoy and push back against the cliff, keeping it from collapsing. 

Glacial speed bump

“After we did that global analysis, we thought, ok, let’s zoom in on the tallest cliffs and see how they’re behaving,” Needell says.

The team focused in on four regions around the Antarctic ice sheet where the satellite captured the tallest ice cliffs, each about 100 meters tall, with one that was 127 meters tall. They compiled measurements of these four cliffs and how they evolved every three months, over four years. 

From these measurements, they saw that indeed, the tallest cliffs collapsed, and the ice sheet in those regions retreated, at a surprisingly faster rate than MICI models had predicted. 

“But that retreat was short-lived,” Needell notes. 

After that, the ice spread back out toward the ocean.

“It’s this rapid retreat, punctuated by periods of stabilization,” Needell says. “That was puzzling to us. Then, we thought: We need to look at all the things that affect cliff height.”

The first thing they considered was the topography of the land underneath the ice. If that topography is relatively flat, then the slope of the overlying ice sheet reflects that actual height of the ice itself. But if there’s a bump or rise in the land, the ice sitting directly over that bump is in fact shorter than it appears. The topography then could act as a literal speed bump, thinning the overlying ice, and making it less likely to collapse. 

When the researchers checked topographic measurements of Antarctic, they found that indeed, ice cliffs stabilized around regions where the land happened to rise up. 

They also found that the taller a cliff is, the faster they “thin out” — as they collapse, once tall cliffs settle into shorter, more stable cliffs that prevents further collapse. 

“Now we know what to look for,” Needell says. “We really need to know the bed topography, and we need to know the ice thinning rate, in order to get ice cliff instability right, because this value really matters in the context of projecting sea level rise.” 

This research was supported, in part, by NASA, the U.S. National Science Foundation, and the DOMINOS project, a component of the International Thwaites Glacier Collaboration.

de MIT News https://ift.tt/uLxRSkY

Building trust in governments through digital platforms

Around the world, many people report distrusting their government. In some places, such as Latin America, North America, and Africa, trust in government has declined in recent decades. How can governments rebuild this trust?

New research from the MIT Governance Lab and the behavioral science research organization Busara examines how government partnerships with trusted private organizations can shape public trust in government. In Kenya, county governments have worked with a telecommunications company to develop an online public service portal. A survey of people in and around the capital city of Nairobi found that highlighting the role of the company increased people’s willingness to use the portal, but hurt their trust in government. 

The results of this work, published in the September issue of the journal Government Information Quarterly, can help researchers understand how government digitization projects and relationships with technology companies affect public trust.

Trust in government and online public services

Studies have suggested that without public trust in government, people are less likely to follow rules and laws, and governments have a harder time solving problems that require public cooperation. Research has linked low levels of trust to people not paying taxes, lower voter turnout, lower Covid-19 vaccination rates, and higher Covid-19 infection rates.

In Kenya, a 2023 survey found that only 43 percent of people trusted the government. Corruption, poor service delivery, and flawed elections have likely hurt public trust. The Kenyan government is also a pioneer in digital technology for governance. Mobile money is widely used, and a centralized service delivery platform called eCitizen has made thousands of digitized services available to citizens. Such digital services can allow governments to serve people more efficiently and, if executed well, can build trust in government competency.

Does who gets credit for making a digital tool impact people’s trust?

Recently, Kenya’s 47 county governments partnered with Safaricom, a highly trusted telecommunications company, to create an online portal called My County Hub. The portal aims to allow people to request government emergency services, pay for business permits and parking, receive information from the government, and more.

In the new study, researchers told the more than 11,000 online survey participants how they could access My County Hub and some of the services it would offer. Some people were told the county government and Safaricom were both responsible for the portal, while others were told the government alone or the company alone was responsible. The survey also asked people about their trust in the county government before and after they received this information. The researchers hypothesized that people who were told the partnership was a public-private collaboration would have more trust at the end of the survey, with people’s trust in Safaricom transferring to the government partner.

Overall, people expressed more trust in government after being told about My County Hub, suggesting that digitization itself could be tied to increasing trust. However, trust did not seem to transfer from Safaricom to the government: people who were told about the partnership did not come to trust the government more than those who were told that the government alone was responsible. Trust sometimes transfers in settings where people don’t have strong beliefs about the untrusted group. But people tend to have strong beliefs about government, which could make it harder to sway their opinions.

Crediting the company could hurt people’s relationship with government

People who were told Safaricom alone was responsible for the portal expressed a greater willingness to use the portal and more interest in learning more about it. But relative to people who were told the government alone made the portal, the people in the Safaricom group had lower levels of trust in government afterwards. This could be because people viewed Safaricom as doing the work of the government, which could lower their faith in the government’s competency. 

The results suggest that while emphasizing the role companies play in developing digital services could encourage people to use these services in the short term, it could also erode people’s confidence in the government in the long term. People with higher trust in government were also more willing to use My County Hub in the future, so long-term trust in government could be important for service provision.

The researchers also noticed that people told about the partnership did see their trust in government increase if they had high levels of trust in the government and Safaricom at the start of the survey. And the decreased trust in government in the Safaricom group was driven by people who trusted Safaricom but distrusted the government at the start. 

These findings are in line with prior research showing that preexisting trust levels influence how trust changes during digitization efforts. In the future, research on people who distrust the government could look at other strategies for building trust through digital services.

The study’s authors include Lily L. Tsai, Ford Professor of Political Science and the director and founder of the MIT Governance Lab, and MIT Governance Lab Research Director Lula Chen.

de MIT News https://ift.tt/5vUTMh4

Q&A: The “collision” between AI, your brain, and society

As artificial intelligence use becomes more widespread, scholars are working to understand its effects on the people who use it. MIT Professor Eric So offers one of the most complete syntheses yet, in his new book, “The Collision: What AI Does to Us,” published this month by W.W. Norton. Recognizing that AI offers opportunities to complete work faster, and in some cases has been turned into useful domain-specific tools, So nonetheless looks at the costs of relying on it too broadly. He contends that heavy use of AI can lead to a kind of cognitive atrophy, in which people fail to develop and retain essential thinking skills. In general, he says, evidence shows that our ability to learn gets dramatically better when we wrestle with material, work through initial failures, keep iterating, and in the process acquire better thinking skills while retaining more knowledge. 

These conclusions are drawn from recent research by So and others, in academia and industry, including analysis from AI firms themselves, such as an Anthropic study showing that about half of queries from college students were low-effort queries for simple answers. So also weaves personal reflections into his narrative — having long studied financial markets and behavior, he is familiar with analyzing how new tools and information affect our decisions. MIT News talked to So, who is the Sloan Distinguished Professor of Global Economics and Behavioral Science in the MIT Sloan School of Management, about the book. 

Q: What prompted you to write “The Collision”?   

A: I wrote this book in response to patterns I was seeing in myself and other individuals. I open the book with this story where I was instinctively reaching for AI to help me think through problems I normally would have handled myself. A colleague had reached out for assistance and I had essentially taken her question and written half a sentence, then said, “ChatGPT, help me finish this professionally.” I made a few light edits and sent it off. And I sat back and really felt incredibly concerned about how I had removed my own thinking from the process.

I have found students are outsourcing their thinking to AI, rather than thinking through things themselves, turning to it to answer homework that would normally help them build mastery in a subject. We need to start understanding what this technology is doing to us and how it interacts with human cognition and our psychology, and give people some practical insights. 

Q: Part of that means understanding the incentives people have to use AI, or how the allure of AI takes hold, even when there are long-term costs, in your view. 

A: There are converging factors drawing us to AI. For example we have a real biological instinct to conserve our mental energy, and socially, we face situations where AI is used by other people to race ahead, and it rachets up the pressure on everyone else to use AI. Combine that with the problems detecting AI use in a practical way, and all these factors create what I call “AI gravity,” this constant pull to use AI. It’s not simply a matter of telling people to think for themselves. At this moment we’re in, AI is colliding with society.

Q: So we need to look at these systemic factors, not just individual decisions, when it comes to AI use, like businesses urging employees to use it. 

A: Many organizations are not just suggesting use of AI: It is now the basis of your evaluation, meaning if you want a bonus or promotion, it is incumbent upon you to use these tools. I think a subtle process is taking hold in academia, where people say, “My future employer will want me to use AI, so I’d better start using it now.” But the problem is that as you start to use AI now, we’re losing touch with the practices and capabilities we’re seeking to build, and in turn that makes us more dependent on using AI in the future, and it becomes a self-fulfilling prophecy.

Q: What is AI doing to our minds? 

A: AI is not just changing what we can do, it’s changing who we are as thinkers. AI removes a lot of the friction necessary to build foundational mastery. It is causing many people to lose their capacity for pushing through problems. AI offers instant relief, and we become accustomed to that. Many people are turning to it as an access point for information, so it’s affecting our ability to remember. Also, AI is changing our interpersonal relationships. A key point in the book is that these models are simultaneously sycophantic, meaning they want to make us happy because that’s how they drive engagement and get positive feedback, but they are also really persuasive. We lose the ability to connect with other people because we’re so accustomed to this frictionless environment where the models always agree with us.

Q: Your book emphasizes the value of “productive struggle.” You write, “Struggle doesn’t mean you’re suffering.” Instead, we get a handle on things by struggling to learn them. 

A: We need to recast how we think about struggle, as something that is formative rather than a problem to be solved. I draw a lot of parallels in the book between some of the societal changes that occurred post-industrialization, when suddenly people no longer needed to be physically active to be economically productive. We saw more people suddenly develop physical exercise as something we do intentionally, that required reframing of what physical struggle meant. And I think we need that sort of shift about what we think about cognitive struggle. 

Q: Given all this, what can we do to maintain some reasonable level of cognitive effort today?

A: At the personal level, I emphasize four principles for navigating this age of AI. One is embracing the struggle, as we’ve talked about. The second rule is: Value who you are without AI. I’ve seen so many presentations over the last couple of years where someone has put together this very polished-looking presentation with Claude, but when they present it, it’s clear they didn’t put in the work necessary to understand the material. The formative struggle creates what I refer to as an “inseparable task bundle,” since making the slides allows me to engage in substantive discussion. In the absence of valuing who you are without AI, we get the product, but not the ability to capitalize on it.

The third rule is about reinvesting your cognitive surplus. There are a lot of tasks for which AI can save you time — rote activities — but it matters what you do with those time savings. If you’re just funneling the same energy into the same tasks, you’re missing out on the potential transformational benefits of AI. The most innovative changes are not the ones that simply allow us to automate, but stretch the capabilities of  humanity in a productive way and build expertise. Or, people can channel that freed-up time to focus on family, learn a new language, cultivate skills. 

The last rule is about using AI as your cognitive trainer. AI is this dual force, where it is very bad for thinking if you use it as an answer machine, but can be great for learning if you use it as a tutor. AI can be tailored to you. However, it requires that you seek out the friction of having a tutor, relative to having an answer machine. And that requires motivation. 

This is a very challenging moment, and the problem is one we need to take seriously. 

de MIT News https://ift.tt/fkJsFZC

MIT remembers Thomas Rose, a brilliant researcher and generous colleague

More than 100 people filled the MIT Chapel on Sept. 24 to pay tribute to Assistant Professor Thomas Rose, the MIT archaeometallurgist who died last month following a bicycle crash in Cambridge, Massachusetts.

Faculty, students, and staff affiliated with the Department of Materials Science and Engineering (DMSE) and the Center for Materials Research in Archaeology and Ethnology (CMRAE), along with family and friends, gathered for a memorial service where colleagues spoke about the 37-year-old, who joined the DMSE faculty in January and had just started building a research program at MIT. They remembered him as a brilliant researcher and generous colleague with an extraordinary breadth of expertise and eclectic passions ranging from rowing to archery. 

A portrait of Rose sat on an easel next to the chapel’s flower-covered altar. A cellist played as guests entered and took their seats.

Chaplain to the Institute Thea Keith-Lucas opened the service by reflecting on the loss felt across MIT. 

“We honor the unique gifts that he brought to this world. We honor the ways that he was growing, and the possibilities that were cut short. We honor the small, unseen moments of courage and connection that create a life,” she said.

Finding a unicorn

To give a sense of what Rose meant to DMSE, department head Polina Anikeeva briefly explained archaeometallurgy — the study of how ancient societies produced and used metals — and its development at MIT, beginning in the 1960s. The field flourished after the establishment of CMRAE in the 1970s, but over the following decades it began to contract in the United States, and MIT had difficulty finding people with an increasingly rare set of skills. 

“We were looking for a unicorn. We were looking for two decades,” Anikeeva said. When Anikeeva and other DMSE faculty reviewed Rose’s resume in 2025, they were impressed by his unique training in geoscience, metallurgy, and archaeology and met him soon after. “We saw a brilliant, creative colleague — unassuming, intellectually fearless, adaptable, resourceful, and eagerly looking into the future.”

Mike Tarkanian, senior lecturer in DMSE and associate director for CMRAE, described the Zoom interview with Rose that kicked off the faculty search. Rose spoke fluently about his work, interests, and research vision — without the slides or notes that candidates typically rely on.

“I hung on every word — but it was so unusual and so risky that I wasn’t sure how it would be received,” Tarkanian said. “As soon as the call ended, my phone rang and it was a colleague who was also on the Zoom and he said, ‘That’s our guy. That has to be our guy.’”

Anikeeva also spoke of the enthusiasm he brought with him every day at MIT, noting his “almost childlike curiosity and his enormous positivity.”

“There were no obstacles, only opportunities and possibilities.”

To honor Rose and carry forward the work he had begun at MIT, Anikeeva said DMSE would launch the Thomas Rose Fellowship for Excellence in Archaeometallurgy, dedicated to supporting the most promising young scholars at the PhD and postdoctoral levels, “so that we can continue this field for the world and for the U.S. and for the memory of Thomas.”

A generous spirit

Speakers also remembered Rose’s kindness, generosity, and empathy toward everyone around him. Tarkanian stressed Rose’s commitment to community and helping others, noting his contributions to research tools used by archaeologists worldwide. He gave an example from Rose’s undergraduate years in his native Germany, when he gave tutorials for first-year archaeological science students who were struggling with chemistry and math. A former student told Tarkanian that an email Rose sent inviting students to the tutorial ended with, “I guess I’ll be more nervous than you folks tomorrow.”

“This was the Thomas we only started to get to know,” Tarkanian said, “someone who offered his expertise to others in need without question, while being thoughtful enough to worry out loud that he might be the nervous one and, by doing so, putting others at ease.”

DMSE Assistant Professor Tania Lopez-Silva, who started at MIT six months before Rose, recalled his patience and understanding. When she arranged a lunch with him, he said he would eat anything but mushrooms and eggplant.

“Somehow I misunderstood him, and I chose the place that probably has the most mushroom and eggplant options possible,” Lopez-Silva said. “He said that he still wanted to try the place, and he would find something to eat. That was Thomas — kind, patient, open-minded, and always willing to make the best of things.”

That quality came through in a story by another DMSE faculty member, Assistant Professor Thomas J. Wallin, who recalled a photo Rose had sent him of the apartment he had found overlooking the Charles River. Wallin was impressed, but Rose said with characteristic directness that it was quite small, but “very long and narrow” — which turned out to be a good thing. 

“He explained that because it was narrow, he could stand at the edge of his balcony and shoot at his archery target through his hallway, past his kitchen, at his door, and it would be the proper distance,” Wallin said, drawing chuckles in the chapel. “I don’t think that’s a request many Boston realtors get, but that was Thomas.”

Rose’s connections also extended beyond DMSE. Katya Tiukhtikova, joined by fellow MIT Archery Club member Kiarra Ning, remembered how Rose treated others. 

“He only focused on the good in people, and gave everyone a chance to show the best parts of themselves,” Tiukhtikova said.

She also recalled the contagious excitement Rose brought to archery practice, saying he “always brightened the days of everyone who was lucky enough to share the archery range with him.”

Tarkanian ended his remarks with a reflection on academic life and the importance of generosity and humility alongside brilliance and achievement.

“If part of Thomas’s legacy here at MIT is that we become a little more generous with our time, a little more humble about our accomplishments, and a little more deliberate about building communities in which other people can thrive and grow, that would be a legacy worthy of him.”

de MIT News https://ift.tt/xd7zw9B

Researchers improve weather forecasts looking two to six weeks out

In 2021, Texas suffered the worst energy infrastructure failure in state history, following three severe winter storms in February. At least 246 people were killed as a result of the disruptions, which led to food, water, heat and electricity shortages.

If energy officials had had more time to prepare for the storms, for example by bringing in more trucks of natural gas to disrupted areas, damage from the event could have been mitigated. Unfortunately, weather forecasts beyond two weeks are notoriously unreliable, making for a tighter window for energy operators, farmers, and emergency response officials around the globe.

A new technique by a team including researchers at MIT significantly improves weather forecasting beyond two weeks. The researchers developed a machine-learning model that learns to correct errors in existing forecasting models to improve their output over the two-to-six-week time frame. When applied to a gold-standard forecasting model, it doubled forecasting accuracy over that time frame, placing first in a prominent international forecasting competition.

In a new paper in Nature, the researchers describe their approach and show how it enables forecasts that for the first time beat climatologic averages between two and six weeks.

“Not only does this do better on time averaged forecasts for this range, which is already very useful for farmers and members of the energy industry trying to match supply with demand, but it also improves predictions of extreme weather events, whether it’s flooding or cold spells or something else,” co-author and MIT Research Scientist Judah Cohen says. “It’s really important to predict both, but extreme events might be the most impactful part of this for society.” 

Joining Cohen on the paper are lead authors Hannah Guan, a Harvard College undergraduate; Soukayna Mouatadid, a University of Toronto postdoc; Genevieve Flaspohler SM ’18, PhD ’22; Paulo Orenstein, an associate professor at Brazil’s National Institute for Pure and Applied Mathematics; Jeremy Berman of the data company DTN; Jakob Schloer and Joshua Talib, researchers with the European Center for Medium-Range Weather Forecasts; and Alex Lu, Haiyu Dong, Jonathan Weyn, Zekun Ni, and Lester Mackey, all researchers at Microsoft.

Longer forecasts

In the 1970s, longtime MIT Professor Edward Lorenz came up with the idea of chaos theory to describe how tiny changes in initial weather conditions lead to completely different outcomes.

“He came up with this theoretical two-week limit where beyond that, we just can’t predict weather because the noise in the system grows so much that it overwhelms any signal you start with,” Cohen says. “That had a strong influence on the field that may have held people back from exploring forecasting beyond the two-week range. I never really believed that completely. In theory the atmosphere can create an infinite number of patterns, but it likes to fall into these favorite patterns over and over again. Once one thing happens, I think you can predict the domino that will fall after that, and the domino that will fall after that.”

Today, the best weather forecasts come from either dynamical models, which numerically simulate how the atmosphere and oceans evolve over time, or artificial intelligence models, which learn to make predictions from historical weather data. AI models have shown improvements in predicting weather at short ranges, but they’re even worse than dynamical models at longer time ranges.

Cohen, who is based in the MIT Department of Civil and Environmental Engineering, focuses on forecasting weather two to six weeks in advance, which is known as subseasonal forecasting. “That range is referred to as the ‘predictability desert,’ because in the short range you can capitalize on the initial conditions in the climate system and the atmosphere, and on the longer term you have these slow, varying boundary forces, like the super El Niño everyone’s focused on this winter,” Cohen says. “The subseasonal range can’t take advantage of either of those. In most cases forecasting in the subseasonal ranges is no different than going to a casino and rolling the dice.”

Cohen credits his co-author Lester Mackey, whom he has been collaborating with for more than a decade, with the initial idea of using machine learning to reduce errors in forecasting models.

“We built this wrapper — Lester calls it postprocessing,” Cohen says. “We know AI models have a lot of errors, so we thought maybe we can quantify those errors and remove them.”

While other forecasting systems do more simplistic error correction, the research team’s postprocessing layer uses AI to learn from a forecasting system’s mistakes based on historic data and shifts its forecasts. In that time range, forecasters use so-called probabilistic forecasts, where the likelihood of a range of possibilities if given.

“As more forecasts and data become available, our model adaptively improves its correction,” Cohen says.

To build the model, the researchers used MIT Engaging, a supercomputing cluster hosted by MIT’s Office of Research Computing and Data, but the resulting model does not require a supercomputer to run.

To put their model to the test, last year the research team submitted three models that included their post-processing layer to the European Center for Medium-Range Weather Forecasts (ECMWF), which is considered the best forecasting center in the world. Each season ECMWF holds forecasting competitions between models.

The submission that won first place combined ECMWF’s forecasting model with another, and then applied the postprocessing correction on top. That approach ended up doubling the original model’s predictive power in the subseasonal range. For the first time, the model offered an improvement over relying on climatologic averages.

“Up until now, no one had been able to show a model predicts weather better than long term averages for these subseasonal forecasts,” Cohen says. “They’ve pretty much all been worse than that — you could literally flip a coin and do better. People have questioned if it’s even worth investing in models at all for this time range. We consistently beat climatology averages with our models.”

The model also showed a statistically significant improvement in the ability to predict extreme weather.

“That’s one of the more important findings, because it’s so difficult to accomplish,” Cohen says. “When the ECMWF developed their own, non-AI, error-correction to their model, it got worse at predicting extreme weather. I think that’s because traditional error, or bias, correction is always trying to get to the mean, whereas extreme weather is on the fringe of probabilities.”

Preparing for disaster

In the Nature paper, the team showed their technique could be used to improve land forecasts in most regions across the globe. They also applied their bias correction technique to other models and showed it improved those as well, putting even lesser models on par with the ECMWF’s forecasts.

“What’s nice about our model is it’s an add-on wrapper,” Cohen explains. “We used it with the ECMWF’s dynamical model because that was considered the best, but it can be used with any model. It even improves models that are already bias corrected using another AI algorithm. So it shows a robust improvement. And it’s open source and low cost.”

Cohen says the meteorological office for the Republic of Kenya has already reached out about implementing the model to improve forecasting of the annual monsoon season to improve agricultural yields this winter. He hopes other groups will also use the model.

“This could help you decide when to fertilize around rain, when there’s a freeze coming to protect crops during winter harvests, or help you avoid droughts,” Cohen says. “It could also be useful for energy and emergency management. I’ve talked to energy officials who say, ‘If I’m told two days before that an ice storm’s coming, I don’t have enough time, but if I had two weeks, that could be enough to make adequate preparations.”

The work was supported, in part, by the U.S. National Science Foundation, the Carlos Chagas Filho Foundation for Research Support, and the European Union.

de MIT News https://ift.tt/LWYC3kR

MIT Farm is growing

“This tomato is blowing my mind,” says Nermine Hassanin. The MIT graduate student in architecture is marveling at the fruit in her salad. 

She and several MIT School of Architecture and Planning classmates had spent their morning working at Hannan Healthy Foods Farm in Lincoln, Massachusetts, and then harvested the contents of the salad they were sharing at a picnic table: lettuce, kale, cucumbers, green onions, and tomatoes. 

“It doesn’t get much fresher unless you’re eating this in the field,” responds Zachary Rapaport, a classmate and co-founder of MIT Farm, a community-driven program at MIT that is radically changing what it means to eat farm-fresh food on campus and gain practical experience in urban farming. 

In fact, the program resonated so quickly with so many that it was awarded the Karl Taylor Compton Prize for a group this spring — the highest award presented by the Institute to students in recognition of excellent achievements in citizenship and devotion to the welfare of MIT. 

Seated with Hassanin and Rapaport were classmates Aisha Cheema and MIT Farm co-founder Aleks Banaś, undergraduate architecture major Amy Kim, Sanne Wright MCP ’24, and Mohammed Hannan, the owner of Hannan Healthy Foods Farm. The story of how they came to enjoy this meal on a brilliant summer Sunday is equal parts vision, hard work, and serendipity. 

Community in Cambridge

In their first year at MIT, Rapaport and Banaś were classmates in a design workshop 4.184 (ClimateCorps@MIT) that focused on climate action within communities. Bonding over their interest in food systems, the two shared a desire to continue the community service work that took them outside of the classroom.

“We started ideating on what it would mean to create relationships with students, faculty, and staff on campus with partners who were already growing food in the community,” says Banaś. “We saw many organizations in the community already doing inspiring work, so instead of starting from scratch we wanted to plug into this system.”

Their concept for MIT Farm explored a holistic view of community, food security, and supporting local farming in a time of climate resilience. It also capitalized on the desires of two budding architects to develop practices they see as essential for sustainable communities.

“After working in an architecture practice, I came to grad school to find a way to create an architecture practice grounded within real communities,” says Banaś. “For me, architecture is not just about creating a building, but creating an environment for other interactions. I think we have a certain level of responsibility to the people around us; we need to learn how to work together with them.”

“What I’ve learned through developing MIT Farm is that farming requires you to being accountable to seasonal change, to people, and to the environment in which we’re working,” adds Rapaport. “There are things you only begin to notice when you spend time with a place and really engage with it. In many ways, architecture is the same. We need to be very attentive to the world around us if we’re going to make good design.”

In 2025, Rapaport and Banaś actively sought out like-minded organizations on campus — including MIT Professor Miho Mazereeuw’s Urban Risk Lab, the MIT Office of Sustainability, and the Open Space Committee — to build relationships and germinate ideas. They found enthusiastic supporters for their ideas which kept them both engaged and excited. That spring, they entered their concept for MIT Farm into the iCampus competition, which recognizes innovative and creative applications of technology that improve living and learning at MIT; they were awarded one of the three cash prizes. 

Between supporters on campus and burgeoning relationships with like-minded organizations off campus, Rapaport and Banaś began to execute their vision.

One idea — a farm cart — kick-started what Rapaport terms “the growing side” of MIT Farm. He, Banaś, and classmate Courage Dzidula Kpodo MArch ’25 designed and built the two-tiered cart. It is constructed of pine with polycarbonate panels and divided into sections that support active vegetable gardens. Repurposed pipette tip containers sourced from campus labs serve as seedling planters. Supported on four wheels, the cart can be pushed or pulled anywhere.

“By introducing mobility, we began growing food without having it tethered to a specific plot of land,” says Rapaport. 

They discovered it was more than just a “moveable farm.” The cart became a DJ booth at school happy hours; was wheeled to nearby public schools as a learning tool; showcased at the “Day of Climate” event at the MIT Museum; and transformed into a tiny winter greenhouse at the East Campus dorms.

Field work

Rapaport and Banaś also located farm communities in Boston and beyond to develop volunteering and mentoring opportunities. Now, through MIT Farm, students can volunteer to work the two rooftop gardens at Boston Medical Center (BMC), a compact farm that generates 6,000 pounds of produce to serve patients and the BMC community. 

A second option for students to work at a city farm is with Eastie Farm in East Boston. A nonprofit with a focus on food security and climate justice, the organization supports local farms that struggle to compete with industrial agriculture.

Last summer, Rapaport and Banaś connected with Mohammed Hannan, the Lincoln farm owner and a laboratory manager at MIT. MIT students had already found their way to his farm, about 15 miles from Cambridge. Four years earlier, Heather Paxson, an anthropology professor and associate dean for faculty at MIT School of Humanities and Social Sciences, brought her anthropology class to visit Hannan Farm to learn about small-scale certified organic operations. 

Hannan was open to the idea of the MIT community working on the farm. He says it is “increasingly challenging” to find people who want to farm. He stresses the broader importance of community support for local agriculture, noting that food security and sustainable farming require collective effort from society — not just from farmers. 

“Many farmers around me are leaving farming to make their living elsewhere,” says Hannan. “It is a silent pandemic.” 

When classes resumed last fall, Rapaport, Banaś, and classmates began volunteering on weekends at the farm. Working with the MIT Office of Sustainability and the Anthropology Section, the group started a pilot program selling Hannan Farm’s produce for a reduced price to students on campus on Mondays. The Fresh Food Bag program offered curated bags of produce that had been harvested a day earlier. The program was met with delight among the MIT community, with nary a vegetable left behind each week. Rapaport and Banaś say they will expand the program this fall to include offering bags of fresh produce in MIT dorms.

The work on the farm changes constantly, based on the season and business needs. One day students may be in the field replanting plants that didn’t establish properly; on another, they may be maintaining tools or working on infrastructure projects.

“There isn't really an ‘off season’ since the farm operates year-round, and we've worked to bring MIT students out in every season and semester,” says Rapaport. “During the winter, we repaired greenhouses, prepared seedling trays, and started getting ready for the next growing season.” 

Hannan describes his collaboration with MIT students as “game changing,” highlighting how students gain practical, problem-solving experience while advising him on how to improve the farm’s infrastructure.

“It’s hard to visualize real-world problems when you are working in a lab [on campus],” says Hannan. “But the students are seeing the day-to-day challenges farmers are facing here. They are coming up with solutions that can alleviate these problems — and not just on my farm. We need more people from different backgrounds to see how we can locally grow more efficiently using the tools and technologies that we have. Sometimes we farmers don’t know where the technologies are — such as using solar irrigation. We’re not always tech-savvy, or we don’t have the expertise. Farming needs all of it.”

To everything, there is a season

In January, during MIT’s Independent Activities Period, Rapaport and Banaś conducted a weeklong workshop to design a root cellar to support food storage and preservation for the farm. The group also developed a curriculum for K-12 students to teach how to design a community garden with the support of a Day of Climate Grant they were awarded.

Kahlen Wheaton, a Department of Urban Studies and Planning (DUSP) undergrad, was among the students who worked on the curriculum and helped organize volunteers at the farm last spring.

“I really love the club,” says Wheaton. “It’s a great opportunity for MIT students to get out of their bubble and do some service work. It’s also great just to bring fresh produce to campus because, as an undergrad, I’m not always eating the most healthy or local foods.”

Wheaton had originally planned to study environmental engineering at MIT, but found urban planning and the opportunity to impact policy — especially as it influences designing cities for healthy lifestyles — exciting. She has planned her thesis on urban farming and will use the connections she is making in local farming to inform her work.

Pastures of plenty

This past spring saw dozens of volunteers spending weekends at Hannan Farm. Among them was Rania Ghosn, an MIT associate professor of architecture and urbanism, who brought her family. She says she enjoyed working shoulder-to-shoulder with colleagues in the field, and that MIT Farm provides a “tangible public good while strengthening a sense of belonging and joy.

“Working in the fields with my husband and our 10-year-old son became a kind of Sunday ritual: purposeful, physical, joyful, and full of meaning. For my son, the farm was a lesson no classroom could quite reproduce.”

On a recent Sunday, five architecture students tended to rows and rows of vegetables. Weeds had grown high with the recent rain, but after some tugging and digging, they released their hold in the dirt and lay in piles. The hot sun, the smell of fresh herbs, and the sounds of birds and insects made Cambridge and MIT feel very far away.

Nermine Hassanin has spent several Sundays here. She says she enjoys meeting new people on the farm each time.

“It’s a good sense of community here,” she says. “And being out of the city is very therapeutic.”

Growing up, Aisha Cheema worked with her dad on their family garden. She says working on the farm is “a nice way to stay connected” to those experiences.

“I’m interested in how this environment can be pulled into architecture,” says Cheema. “So many spaces seem siloed from the outdoors. I’d like to connect the outdoors to indoor spaces. I think some of the structures here — like the greenhouse — are great inspiration for me.”

Rapaport and Banaś spent their summer designing additional interventions for the Lincoln farm: an improved washing station and an upgraded farm stand that allows for cold storage to keep produce fresh longer. The design has been approved by the Lincoln Conservation Commission and permits are pending. 

Also, this summer, Amy Kim oversaw the design of a dormant garden plot on campus near Tang Hall. The undergraduate performed the work as part of MIT’s Undergraduate Research Opportunity Program. Kim says she designed the project because it allowed her “to get hands-on experience building a space that would foster community and collaboration, with the added environmental benefit of repurposing material such as scrap brick and scrap wood. 

“The garden's primary goal is to be a pleasant space to sit and relax with friends while surrounded by aromatic herbs and flowers,” Kim adds.

With their theses underway in preparation for Commencement next year, Rapaport and Banaś will hand off their oversight of MIT Farm to Wheaton and others this fall. They could not have guessed how the popularity of MIT Farm would have connected with so many students, or how many friends they cultivated off campus — and they are overjoyed with the results. They believe the community they have developed and fostered will keep MIT Farm vibrant into the future. 

Ghosn agrees.

“MIT Farm feels deeply MIT: mind and hand, ethics and experimentation, systems thinking and dirt under the fingernails,” she says. “Aleks and Zach have created a student-led project with moral clarity, organizational intelligence, and genuine public purpose. It is modest in its means and expansive in its imagination and impact."

de MIT News https://ift.tt/KjWl0JD