Mission directors announced for the Climate Project at MIT

The Climate Project at MIT has appointed leaders for each of its six focal areas, or Climate Missions, President Sally Kornbluth announced in a letter to the MIT community today.

Introduced in February, the Climate Project at MIT is a major new effort to change the trajectory of global climate outcomes for the better over the next decade. The project will focus MIT’s strengths on six broad climate-related areas where progress is urgently needed. The mission directors in these fields, representing diverse areas of expertise, will collaborate with faculty and researchers across MIT, as well as each other, to accelerate solutions that address climate change.

“The mission directors will be absolutely central as the Climate Project seeks to marshal the Institute’s talent and resources to research, develop, deploy and scale up serious solutions to help change the planet’s climate trajectory,” Kornbluth wrote in her letter, adding: “To the faculty members taking on these pivotal roles: We could not be more grateful for your skill and commitment, or more enthusiastic about what you can help us all achieve, together.”

The Climate Project will expand and accelerate MIT’s efforts to both reduce greenhouse gas emissions and respond to climate effects such as extreme heat, rising sea levels, and reduced crop yields. At the urgent pace needed, the project will help the Institute create new external collaborations and deepen existing ones to develop and scale climate solutions.

The Institute has pledged an initial $75 million to the project, including $25 million from the MIT Sloan School of Management to launch a complementary effort, the new MIT Climate Policy Center. MIT has more than 300 faculty and senior researchers already working on climate issues, in collaboration with their students and staff. The Climate Project at MIT builds on their work and the Institute’s 2021 “Fast Forward” climate action plan.

Richard Lester, MIT’s vice provost for international activities and the Japan Steel Industry Professor of Nuclear Science and Engineering, has led the Climate Project’s formation; MIT will shortly hire a vice president for climate to oversee the project. The six Climate Missions and the new mission directors are as follows:

Decarbonizing energy and industry

This mission supports advances in the electric power grid as well as the transition across all industry — including transportation, computing, heavy production, and manufacturing — to low-emissions pathways.

The mission director is Elsa Olivetti PhD ’07, who is MIT’s associate dean of engineering, the Jerry McAfee Professor in Engineering, and a professor of materials science and engineering since 2014.

Olivetti analyzes and improves the environmental sustainability of materials throughout the life cycle and across the supply chain, by linking physical and chemical processes to systems impact. She researches materials design and synthesis using natural language processing, builds models of material supply and technology demand, and assesses the potential for recovering value from industrial waste through experimental approaches. Olivetti has experience building partnerships across the Institute and working with industry to implement large-scale climate solutions through her role as co-director of the MIT Climate and Sustainability Consortium (MCSC) and as faculty lead for PAIA, an industry consortium on the carbon footprinting of computing.

Restoring the atmosphere, protecting the land and oceans

This mission is centered on removing or storing greenhouse gases that have already been emitted into the atmosphere, such as carbon dioxide and methane, and on protecting ocean and land ecosystems, including food and water systems.

MIT has chosen two mission directors: Andrew Babbin and Jesse Kroll. The two bring together research expertise from two critical domains of the Earth system, oceans and the atmosphere, as well as backgrounds in both the science and engineering underlying our understanding of Earth’s climate. As co-directors, they jointly link MIT’s School of Science and School of Engineering in this domain.

Babbin is the Cecil and Ida Green Career Development Professor in MIT’s Program in Atmospheres, Oceans, and Climate. He is a marine biogeochemist whose specialty is studying the carbon and nitrogen cycle of the oceans, work that is related to evaluating the ocean’s capacity for carbon storage, an essential element of this mission’s work. He has been at MIT since 2017.

Kroll is a professor in MIT’s Department of of Civil and Environmental Engineering, a professor of chemical engineering, and the director of the Ralph M. Parsons Laboratory. He is a chemist who studies organic compounds and particulate matter in the atmosphere, in order to better understand how perturbations to the atmosphere, both intentional and unintentional, can affect air pollution and climate.

Empowering frontline communities

This mission focuses on the development of new climate solutions in support of the world’s most vulnerable populations, in areas ranging from health effects to food security, emergency planning, and risk forecasting.

The mission director is Miho Mazereeuw, an associate professor of architecture and urbanism in MIT’s Department of Architecture in the School of Architecture and Planning, and director of MIT’s Urban Risk Lab. Mazereeuw researches disaster resilience, climate change, and coastal strategies. Her lab has engaged in design projects ranging from physical objects to software, while exploring methods of engaging communities and governments in preparedness efforts, skills she brings to bear on building strong collaborations with a broad range of stakeholders.

Mazereeuw is also co-lead of one of the five projects selected in MIT’s Climate Grand Challenges competition in 2022, an effort to help communities prepare by understanding the risk of extreme weather events for specific locations.

Building and adapting healthy, resilient cities

A majority of the world’s population lives in cities, so urban design and planning is a crucial part of climate work, involving transportation, infrastructure, finance, government, and more.

Christoph Reinhart, the Alan and Terri Spoon Professor of Architecture and Climate and director of MIT’s Building Technology Program in the School of Architecture and Planning, is the mission director in this area. The Sustainable Design Lab that Reinhart founded when he joined MIT in 2012 has launched several technology startups, including Mapdwell Solar System, now part of Palmetto Clean Technology, as well as Solemma, makers of an environmental building design software used in architectural practice and education worldwide. Reinhart’s online course on Sustainable Building Design has an enrollment of over 55,000 individuals and forms part of MIT’s XSeries Program in Future Energy Systems.

Inventing new policy approaches

Climate change is a unique crisis. With that in mind, this mission aims to develop new institutional structures and incentives — in carbon markets, finance, trade policy, and more — along with decision support tools and systems for scaling up climate efforts.

Christopher Knittel brings extensive knowledge of these topics to the mission director role. The George P. Shultz Professor and Professor of Applied Economics at the MIT Sloan School of Management, Knittel has produced high-impact research in multiple areas; his studies on emissions and the automobile industry have evaluated fuel-efficiency standards, changes in vehicle fuel efficiency, market responses to fuel-price changes, and the health impact of automobiles.

Beyond that, Knittel has also studied the impact of the energy transition on jobs, conducted high-level evaluations of climate policies, and examined energy market structures. He joined the MIT faculty in 2011. He also serves as the director of the MIT Climate Policy Center, which will work closely with all six missions.

Wild cards

This mission consists of what the Climate Project at MIT calls “unconventional solutions outside the scope of the other missions,” and will have a broad portfolio for innovation.

While all the missions will be charged with encouraging unorthodox approaches within their domains, this mission will seek out unconventional solutions outside the scope of the others, and has a broad mandate for promoting them.

The mission director in this case is Benedetto Marelli, the Paul M. Cook Career Development Associate Professor in MIT’s Department of Civil and Environmental Engineering. Marelli’s research group develops biopolymers and bioinspired materials with reduced environmental impact compared to traditional technologies. He engages with research at multiple scales, including nanofabrication, and the research group has conducted extensive work on food security and safety while exploring new techniques to reduce waste through enhanced food preservation and to precisely deliver agrochemicals in plants and in soil.

As Lester and other MIT leaders have noted, the Climate Project at MIT is still being shaped, and will have the flexibility to accommodate a wide range of projects, partnerships, and approaches needed for thoughtful, fast-moving change. By filling out the leadership structure, today’s announcement is a major milestone in making the project operational.

In addition to the six Climate Missions, the Climate Project at MIT includes Climate Frontier Projects, which are efforts launched by these missions, and a Climate HQ, which will support fundamental research, education, and outreach, as well as new resources to connect research to the practical work of climate response.

de MIT News https://ift.tt/j6dsQVS

Or Hen: Getting to the core of the matter

Every now and then, Or Hen, who recently received tenure as an associate professor of physics at MIT, will refer back to a file that he has kept since middle school.

The file is a comprehensive assessment of Hen’s learning disabilities, stemming from dysgraphia — a neurological condition in which someone has difficulty translating their thoughts into written form. Hen was diagnosed with a severe case of dysgraphia as a kindergartener. In middle school, due to an administrative snafu, the school was unaware of his condition and, lacking proper support, Hen failed most of his classes. It wasn’t until one teacher took a special interest that Hen was sent for a detailed assessment of his learning abilities.

That assessment, which Hen did not read until later, when he was well into his undergraduate degree in physics, was for him a revelation and validation.

“I saw a lot of ‘he’s bad at this, and not good at that,’ and it went through all the things I failed at,” Hen recalls. “But there was one test where I did extremely well, and which they had bold-faced.”

It was a test of nonverbal thinking, or abstract comprehension, assessing Hen’s ability to conceptually pare down complex ideas to their fundamental core. In this test, Hen scored in the 99th percentile. Fittingly, by the time Hen read this, he was already immersed in studies of abstract concepts and systems in nuclear physics. On his own, he had gravitated to a field that suited his strengths. Reading the assessment gave him confidence in his own instincts.

“They wrote that ‘this skill is particularly strong in him, and you should push him toward areas that utilize it,’” Hen says.

He brings this up to students at MIT today, not to brag, but as a guide.

“I try to emphasize that there is more than one path to success,” Hen says. “Try to think about what you’re good at, and then do the hard work of finding out what area, group, subfield, can utilize that. Find the thing you bring that’s unique, and build on that. Because then you really shine.”

Today, Hen and his research group are probing the inner workings of the nucleus, the interactions between protons and neutrons, and their even smaller constituents of quarks and gluons, which are the basic ingredients that hold together all the visible matter in the universe. Hen seeks connections between how these particles behave and how their interactions shape the visible universe and extreme astrophysical phenomena such as neutron stars.

“A lot of physics, in my mind, is taking complex systems with lots of details and abstracting away the details to seek the main principles that drive everything,” he says.

A frontier of ideas

Hen grew up in the countryside of Jerusalem, Israel, as part of a moshav — a small Jewish village, where his family worked as farmers, raising chickens for their eggs. In kindergarten, once Hen was diagnosed with dysgraphia, his parents sought out any available resources to help him with his writing.

“I think my mother’s entire salary went toward corrective lessons,” Hen recalls.

In middle school, once the records of his condition were transferred to the school, and an assessment was made of his abilities, Hen was given permission to take his exams orally. Rather than writing his answers, he would sit with a teacher and talk it out. To this day, he credits his loquacious nature to those early, formative years.

“Everyone who knows me knows I talk a lot,” Hen says. “And part of that was the way I was able to learn by talking about the material.”

Once he could work around his dysgraphia, however, Hen soon became bored by the content of his lessons, and in high school, he routinely skipped class. A teacher, seeing his potential, told his parents about an outreach program at the nearby Hebrew University, which Hen’s teacher thought might challenge the boy in ways that high school could not.

In his last two years of high school, Hen took part in the program and enrolled in a couple university classes in programming, which he quickly took to. After graduation, he attended Hebrew University full-time, double-majoring in computer engineering and physics — a topic that he thought he might like, as his older brother had also majored in the topic.

One class, early on in his first year, was especially motivating. The class explored ideas in modern physics, and students got to hear from different physicists about the concepts and phenomena they were tackling in the moment.

“It showed us that the beginning of our studies may be hard and annoying, but this is what you’re working toward,” Hen says. “In that class, we learned about quantum mechanics and nonlinear solids and astrophysics, and it just gave us a view of the frontier of the field.”

At the core

After completing his undergraduate degrees, Hen joined the Israeli Defense Forces, which is a mandatory service for all Israeli citizens. He spent seven years in the army, working as a researcher in a physics laboratory.

In tandem with his military service, Hen was also pursuing a PhD in physics, and would make the short trip to Tel-Aviv University once a week and on weekends to work on his degree. There, he got to know an eccentric and beloved professor who took a chance on Hen and offered him a rare opportunity: to travel to the United States to help build a new particle detector. The detector would be based at Jefferson Laboratory, a facility funded by the U.S. Department of Energy that houses a huge particle accelerator, designed to collide beams of electrons with various atomic nuclei.

With a particle detector, physicists could essentially snap pictures of a collision and its aftermath, to tease out the subatomic constituents and their properties, and how they interact to make up an atom’s nuclear structure.

Hen spent a summer at the facility, helping to build a neutron detector that physicists hoped would shed light on “short-range correlations” — extremely brief, quantum-mechanical fluctuations that can occur between some protons and neutrons within an atom’s nucleus. When these particles get so close as to touch each other, their interactions become stronger, though only for a moment before they flit away. It’s thought that these short-range correlations are the source of most of the kinetic energy in a nucleus, which itself is the basis of all visible matter in the universe.

“More than half the kinetic energy in a nucleus comes from these weird states,” Hen says. “If you ever want to understand atomic nuclei and visible matter at its core, you have to also understand short-range correlations.”

Helping to build the neutron detector was a gratifying combination of hands-on work and abstract thinking, and from then on, Hen was hooked on experimental nuclear physics.

After completing his PhD, and a thesis on short-range correlations, Hen headed to MIT, where he interviewed for a postdoc position as a Pappalardo Fellow in the Laboratory of Nuclear Science. As he chatted with one person after another, he eventually found himself in the office of then-department head Peter Fisher, who encouraged Hen to also apply for an open faculty position.

A few months later, in 2015, he found himself in the fortunate position of starting at MIT as a postdoc, having already accepted a junior faculty position he would start at MIT 18 months later.

“MIT took a bet on me,” he says. “That’s the unique thing about MIT. They saw something in me that I didn’t see back then, and they supported me.”

Particle connections

In his first years on campus, Hen continued his work in short-range correlations. His group used data from particle accelerators around the world to develop a universal understanding of short-range correlations in a way that can be applied across many scales. It could, for instance, predict how the interactions would determine correlations in one type of atom versus another, and shape the behavior of much more dense and extreme phenomena such as neutron stars.

Hen also expanded into the field of neutrinos, which are nearly massless particles that are the most abundant particles in the universe. The properties of neutrinos are thought to be the key to the origins of matter, though neutrinos are notoriously difficult to study in detail because their detection requires detailed understanding of their interaction with atomic nuclei. Hen found that, instead of depending on the elusive interactions of neutrinos, there might be a way to abstract that behavior to that of a more detectable particle, to better understand the neutrino itself.

By analyzing data from electron-beam accelerators around the world, his group founded the “electrons-for-neutrinos” effort, which developed a framework that essentially transposed the interactions of an electron to describe how a neutrino would behave under similar circumstances — a tool that will help physicists interpret data from hard-to-pin-down neutrino experiments.

Reflecting on how he determines which direction to take his research, Hen says: “I have a big nose, and I like to talk to people and understand what they’re doing and whether can I do something there or not. I like to build communities, bring in people with different abilities, and do something big together, where the whole is greater than the sum of its parts.”

Big science

Hen got a chance to start up a big-science collaboration, as part of the Electron-Ion Collider (EIC), a concept for a particle accelerator that collides electrons with protons, neutrons, and nuclei to study the particles’ internal structures and how they are held together by the “strong nuclear force,” which is known as the strongest force in nature.

In late 2019, the EIC was a focus of a meeting at MIT, in which physicists from around the world gathered to discuss the project’s recent go-ahead, granted by the U.S. Department of Energy (DoE). The next step was to design a detector, and multiple versions were considered. Hen, who stopped in out of curiosity, wound up joining a community-wide effort to develop a menu of possible detectors that could be built at the EIC.

Hen then took a leadership role in the next step to put forward a specific detector design for the DoE to fund. Working closely with physicists Tanja Horn and John Lajoie, they called experts to join the effort, eventually gathering physicists from 98 institutions. Thanks to their collaborative efforts the DoE ultimately chose their design over a competing one. Hen and his colleagues subsequently reached out to that other group to join forces to further evolve and fine-tune the design.

“We combined strengths,” Hen says. “We are doing big science. And when you do big science, there’s lots of talented people involved. I’ve learned through this process that it’s all about how you interact with people and adapt yourself to do science, together.”

Today, Hen is overseeing aspects of the EIC’s science community that is leading its development, which is projected to break ground in the next few years. In the meantime, he continues to expand projects in his research group, and works to mentor his students and postdocs, just as he was supported through his early career.

“I’m a very fortunate person, in that I had so many mentors in my life, and they all believed in me and saw things I didn’t,” Hen says. “They noticed that I’m different in whichever capacity, and tried to squeeze that lemon. That was a big thing for me.” 

de MIT News https://ift.tt/vYDkt8x

Collaborating to advance LEADing-edge digital financial infrastructure

MIT’s Laboratory for Economic Analysis and Design (LEAD) has been awarded a 400,000-euro grant from the Deutsche Gesellschaft für Internationale Zusammenarbeit (GIZ) GmbH, a German service provider focused on international cooperation for sustainable development and international education. The grant aims to create knowledge sharing opportunities for central bank leaders and help low- and middle-income countries (LMICs) design and scale central bank operations and digital public infrastructure (DPI).

"Increased research between leading economists and computer scientists is critical, and an equal exchange between academics and central bankers is required to mitigate the risks and realize the innovative potential of this emerging field,” says LEAD director Robert M. Townsend, the Elizabeth and James Killian Professor of Economics, who is principal investigator for the funded project.

Townsend will combine computer science, economic theory, and data to help LMICs implement projects while centering them in central bank digital currency (CBDC) and DPI research.

“For such systemic technologies, large-scale interdisciplinary research is rare,” notes Shira Frank, director of Maiden Labs, a research organization with which LEAD intends to collaborate through the grant. “This collaboration aims to change that.”

Townsend is an expert in developmental economics, economic theory, and macroeconomics. He is known for his influential work on costly state verification, the revelation principle, optimal multi-period contracts, decentralization of economies with private information, models of money with spatially separated agents, forecasting the forecasts of others, and insurance and credit in developing countries. He is also a research associate at the National Bureau of Economics, a member of the National Academy of Sciences, and the only two-time winner of the Frisch Medal.

Research and participant focus

The grant, which funds the project through June 2025, will help create opportunities for LMIC leaders to engage in research and collaboration among policymakers, technologists, and economists.

The pilot program aims to produce:

• in-depth research collaborations with two to three LMIC central banks;
• a series of peer-knowledge exchange workshops with 10-20 LMICs;
• a published CBDC and DPI curriculum for central banks, integrating research from economics, computer science, and user research; and
• a published report of research findings.

“We want LMICs to lead the charge into information sharing and technological infrastructure scaling alongside subject matter experts,” Townsend said. “It’s important to meet LMIC leaders where they are.”

Rapid changes to digital tools and the infrastructure necessary to implement, monitor, and protect them will require reliable, effective products. These can include smart contracts, which are open access digital agreements to be signed and stored on a blockchain network; distributed ledgers, platforms that use ledgers stored on separate, connected devices in a network to ensure data accuracy and security; and, encryption, which is necessary to protect electronic data from intrusion and capture and safeguard transmission by reducing exposure to bad actors.

Outcomes

The program hopes to contribute to the development of flexible financial wholesale platforms for use in improving both financial operations in LMICs and the operation of high-valued asset markets. Townsend also hopes to help participants investigate and establish digital transmission infrastructure in countries that primarily communicate and encrypt communications on mobile device networks at the retail level. The distinction between legacy communications infrastructure and mobile data transmission is blurred with new technologies.

“We want to increase interest and investment in these countries’ financial well-being,” Townsend said. “If we can identify the best learning model, this work offers multiple opportunities to foster collaboration among global financial partners.” 

de MIT News https://ift.tt/ApuJ0jT

Proton-conducting materials could enable new green energy technologies

As the name suggests, most electronic devices today work through the movement of electrons. But materials that can efficiently conduct protons — the nucleus of the hydrogen atom — could be key to a number of important technologies for combating global climate change.

Most proton-conducting inorganic materials available now require undesirably high temperatures to achieve sufficiently high conductivity. However, lower-temperature alternatives could enable a variety of technologies, such as more efficient and durable fuel cells to produce clean electricity from hydrogen, electrolyzers to make clean fuels such as hydrogen for transportation, solid-state proton batteries, and even new kinds of computing devices based on iono-electronic effects.

In order to advance the development of proton conductors, MIT engineers have identified certain traits of materials that give rise to fast proton conduction. Using those traits quantitatively, the team identified a half-dozen new candidates that show promise as fast proton conductors. Simulations suggest these candidates will perform far better than existing materials, although they still need to be conformed experimentally. In addition to uncovering potential new materials, the research also provides a deeper understanding at the atomic level of how such materials work.

The new findings are described in the journal Energy and Environmental Sciences, in a paper by MIT professors Bilge Yildiz and Ju Li, postdocs Pjotrs Zguns and Konstantin Klyukin, and their collaborator Sossina Haile and her students from Northwestern University. Yildiz is the Breene M. Kerr Professor in the departments of Nuclear Science and Engineering, and Materials Science and Engineering.

“Proton conductors are needed in clean energy conversion applications such as fuel cells, where we use hydrogen to produce carbon dioxide-free electricity,” Yildiz explains. “We want to do this process efficiently, and therefore we need materials that can transport protons very fast through such devices.”

Present methods of producing hydrogen, for example steam methane reforming, emit a great deal of carbon dioxide. “One way to eliminate that is to electrochemically produce hydrogen from water vapor, and that needs very good proton conductors,” Yildiz says. Production of other important industrial chemicals and potential fuels, such as ammonia, can also be carried out through efficient electrochemical systems that require good proton conductors.

But most inorganic materials that conduct protons can only operate at temperatures of 200 to 600 degrees Celsius (roughly 450 to 1,100 Fahrenheit), or even higher. Such temperatures require energy to maintain and can cause degradation of materials. “Going to higher temperatures is not desirable because that makes the whole system more challenging, and the material durability becomes an issue,” Yildiz says. “There is no good inorganic proton conductor at room temperature.” Today, the only known room-temperature proton conductor is a polymeric material that is not practical for applications in computing devices because it can’t easily be scaled down to the nanometer regime, she says.

To tackle the problem, the team first needed to develop a basic and quantitative understanding of exactly how proton conduction works, taking a class of inorganic proton conductors, called solid acids. “One has to first understand what governs proton conduction in these inorganic compounds,” she says. While looking at the materials’ atomic configurations, the researchers identified a pair of characteristics that directly relates to the materials’ proton-carrying potential.

As Yildiz explains, proton conduction first involves a proton “hopping from a donor oxygen atom to an acceptor oxygen. And then the environment has to reorganize and take the accepted proton away, so that it can hop to another neighboring acceptor, enabling long-range proton diffusion.” This process happens in many inorganic solids, she says. Figuring out how that last part works — how the atomic lattice gets reorganized to take the accepted proton away from the original donor atom — was a key part of this research, she says.

The researchers used computer simulations to study a class of materials called solid acids that become good proton conductors above 200 degrees Celsius. This class of materials has a substructure called the polyanion group sublattice, and these groups have to rotate and take the proton away from its original site so it can then transfer to other sites. The researchers were able to identify the phonons that contribute to the flexibility of this sublattice, which is essential for proton conduction. Then they used this information to comb through vast databases of theoretically and experimentally possible compounds, in search of better proton conducting materials.

As a result, they found solid acid compounds that are promising proton conductors and that have been developed and produced for a variety of different applications but never before studied as proton conductors; these compounds turned out to have just the right characteristics of lattice flexibility. The team then carried out computer simulations of how the specific materials they identified in their initial screening would perform under relevant temperatures, to confirm their suitability as proton conductors for fuel cells or other uses. Sure enough, they found six promising materials, with predicted proton conduction speeds faster than the best existing solid acid proton conductors.

“There are uncertainties in these simulations,” Yildiz cautions. “I don’t want to say exactly how much higher the conductivity will be, but these look very promising. Hopefully this motivates the experimental field to try to synthesize them in different forms and make use of these compounds as proton conductors.”

Translating these theoretical findings into practical devices could take some years, she says. The likely first applications would be for electrochemical cells to produce fuels and chemical feedstocks such as hydrogen and ammonia, she says.

The work was supported by the U.S. Department of Energy, the Wallenberg Foundation, and the U.S. National Science Foundation.

de MIT News https://ift.tt/t3x7ryc

Large language models don’t behave like people, even though we may expect them to

One thing that makes large language models (LLMs) so powerful is the diversity of tasks to which they can be applied. The same machine-learning model that can help a graduate student draft an email could also aid a clinician in diagnosing cancer.

However, the wide applicability of these models also makes them challenging to evaluate in a systematic way. It would be impossible to create a benchmark dataset to test a model on every type of question it can be asked.

In a new paper, MIT researchers took a different approach. They argue that, because humans decide when to deploy large language models, evaluating a model requires an understanding of how people form beliefs about its capabilities.

For example, the graduate student must decide whether the model could be helpful in drafting a particular email, and the clinician must determine which cases would be best to consult the model on.

Building off this idea, the researchers created a framework to evaluate an LLM based on its alignment with a human’s beliefs about how it will perform on a certain task.

They introduce a human generalization function — a model of how people update their beliefs about an LLM’s capabilities after interacting with it. Then, they evaluate how aligned LLMs are with this human generalization function.

Their results indicate that when models are misaligned with the human generalization function, a user could be overconfident or underconfident about where to deploy it, which might cause the model to fail unexpectedly. Furthermore, due to this misalignment, more capable models tend to perform worse than smaller models in high-stakes situations.

“These tools are exciting because they are general-purpose, but because they are general-purpose, they will be collaborating with people, so we have to take the human in the loop into account,” says study co-author Ashesh Rambachan, assistant professor of economics and a principal investigator in the Laboratory for Information and Decision Systems (LIDS).

Rambachan is joined on the paper by lead author Keyon Vafa, a postdoc at Harvard University; and Sendhil Mullainathan, an MIT professor in the departments of Electrical Engineering and Computer Science and of Economics, and a member of LIDS. The research will be presented at the International Conference on Machine Learning.

Human generalization

As we interact with other people, we form beliefs about what we think they do and do not know. For instance, if your friend is finicky about correcting people’s grammar, you might generalize and think they would also excel at sentence construction, even though you’ve never asked them questions about sentence construction.

“Language models often seem so human. We wanted to illustrate that this force of human generalization is also present in how people form beliefs about language models,” Rambachan says.

As a starting point, the researchers formally defined the human generalization function, which involves asking questions, observing how a person or LLM responds, and then making inferences about how that person or model would respond to related questions.

If someone sees that an LLM can correctly answer questions about matrix inversion, they might also assume it can ace questions about simple arithmetic. A model that is misaligned with this function — one that doesn’t perform well on questions a human expects it to answer correctly — could fail when deployed.

With that formal definition in hand, the researchers designed a survey to measure how people generalize when they interact with LLMs and other people.

They showed survey participants questions that a person or LLM got right or wrong and then asked if they thought that person or LLM would answer a related question correctly. Through the survey, they generated a dataset of nearly 19,000 examples of how humans generalize about LLM performance across 79 diverse tasks.

Measuring misalignment

They found that participants did quite well when asked whether a human who got one question right would answer a related question right, but they were much worse at generalizing about the performance of LLMs.

“Human generalization gets applied to language models, but that breaks down because these language models don’t actually show patterns of expertise like people would,” Rambachan says.

People were also more likely to update their beliefs about an LLM when it answered questions incorrectly than when it got questions right. They also tended to believe that LLM performance on simple questions would have little bearing on its performance on more complex questions.

In situations where people put more weight on incorrect responses, simpler models outperformed very large models like GPT-4.

“Language models that get better can almost trick people into thinking they will perform well on related questions when, in actuality, they don’t,” he says.

One possible explanation for why humans are worse at generalizing for LLMs could come from their novelty — people have far less experience interacting with LLMs than with other people.

“Moving forward, it is possible that we may get better just by virtue of interacting with language models more,” he says.

To this end, the researchers want to conduct additional studies of how people’s beliefs about LLMs evolve over time as they interact with a model. They also want to explore how human generalization could be incorporated into the development of LLMs.

“When we are training these algorithms in the first place, or trying to update them with human feedback, we need to account for the human generalization function in how we think about measuring performance,” he says.

In the meanwhile, the researchers hope their dataset could be used a benchmark to compare how LLMs perform related to the human generalization function, which could help improve the performance of models deployed in real-world situations.

“To me, the contribution of the paper is twofold. The first is practical: The paper uncovers a critical issue with deploying LLMs for general consumer use. If people don’t have the right understanding of when LLMs will be accurate and when they will fail, then they will be more likely to see mistakes and perhaps be discouraged from further use. This highlights the issue of aligning the models with people's understanding of generalization,” says Alex Imas, professor of behavioral science and economics at the University of Chicago’s Booth School of Business, who was not involved with this work. “The second contribution is more fundamental: The lack of generalization to expected problems and domains helps in getting a better picture of what the models are doing when they get a problem ‘correct.’ It provides a test of whether LLMs ‘understand’ the problem they are solving.”

This research was funded, in part, by the Harvard Data Science Initiative and the Center for Applied AI at the University of Chicago Booth School of Business.

de MIT News https://ift.tt/oFhkQ30

AI model identifies certain breast tumor stages likely to progress to invasive cancer

Ductal carcinoma in situ (DCIS) is a type of preinvasive tumor that sometimes progresses to a highly deadly form of breast cancer. It accounts for about 25 percent of all breast cancer diagnoses.

Because it is difficult for clinicians to determine the type and stage of DCIS, patients with DCIS are often overtreated. To address this, an interdisciplinary team of researchers from MIT and ETH Zurich developed an AI model that can identify the different stages of DCIS from a cheap and easy-to-obtain breast tissue image. Their model shows that both the state and arrangement of cells in a tissue sample are important for determining the stage of DCIS.

Because such tissue images are so easy to obtain, the researchers were able to build one of the largest datasets of its kind, which they used to train and test their model. When they compared its predictions to conclusions of a pathologist, they found clear agreement in many instances.

In the future, the model could be used as a tool to help clinicians streamline the diagnosis of simpler cases without the need for labor-intensive tests, giving them more time to evaluate cases where it is less clear if DCIS will become invasive.

“We took the first step in understanding that we should be looking at the spatial organization of cells when diagnosing DCIS, and now we have developed a technique that is scalable. From here, we really need a prospective study. Working with a hospital and getting this all the way to the clinic will be an important step forward,” says Caroline Uhler, a professor in the Department of Electrical Engineering and Computer Science (EECS) and the Institute for Data, Systems, and Society (IDSS), who is also director of the Eric and Wendy Schmidt Center at the Broad Institute of MIT and Harvard and a researcher at MIT’s Laboratory for Information and Decision Systems (LIDS).

Uhler, co-corresponding author of a paper on this research, is joined by lead author Xinyi Zhang, a graduate student in EECS and the Eric and Wendy Schmidt Center; co-corresponding author GV Shivashankar, professor of mechogenomics at ETH Zurich jointly with the Paul Scherrer Institute; and others at MIT, ETH Zurich, and the University of Palermo in Italy. The open-access research was published July 20 in Nature Communications.

Combining imaging with AI   

Between 30 and 50 percent of patients with DCIS develop a highly invasive stage of cancer, but researchers don’t know the biomarkers that could tell a clinician which tumors will progress.

Researchers can use techniques like multiplexed staining or single-cell RNA sequencing to determine the stage of DCIS in tissue samples. However, these tests are too expensive to be performed widely, Shivashankar explains.

In previous work, these researchers showed that a cheap imagining technique known as chromatin staining could be as informative as the much costlier single-cell RNA sequencing.

For this research, they hypothesized that combining this single stain with a carefully designed machine-learning model could provide the same information about cancer stage as costlier techniques.

First, they created a dataset containing 560 tissue sample images from 122 patients at three different stages of disease. They used this dataset to train an AI model that learns a representation of the state of each cell in a tissue sample image, which it uses to infer the stage of a patient’s cancer.

However, not every cell is indicative of cancer, so the researchers had to aggregate them in a meaningful way.

They designed the model to create clusters of cells in similar states, identifying eight states that are important markers of DCIS. Some cell states are more indicative of invasive cancer than others. The model determines the proportion of cells in each state in a tissue sample.

Organization matters

“But in cancer, the organization of cells also changes. We found that just having the proportions of cells in every state is not enough. You also need to understand how the cells are organized,” says Shivashankar.

With this insight, they designed the model to consider proportion and arrangement of cell states, which significantly boosted its accuracy.

“The interesting thing for us was seeing how much spatial organization matters. Previous studies had shown that cells which are close to the breast duct are important. But it is also important to consider which cells are close to which other cells,” says Zhang.

When they compared the results of their model with samples evaluated by a pathologist, it had clear agreement in many instances. In cases that were not as clear-cut, the model could provide information about features in a tissue sample, like the organization of cells, that a pathologist could use in decision-making.

This versatile model could also be adapted for use in other types of cancer, or even neurodegenerative conditions, which is one area the researchers are also currently exploring.

“We have shown that, with the right AI techniques, this simple stain can be very powerful. There is still much more research to do, but we need to take the organization of cells into account in more of our studies,” Uhler says.

This research was funded, in part, by the Eric and Wendy Schmidt Center at the Broad Institute, ETH Zurich, the Paul Scherrer Institute, the Swiss National Science Foundation, the U.S. National Institutes of Health, the U.S. Office of Naval Research, the MIT Jameel Clinic for Machine Learning and Health, the MIT-IBM Watson AI Lab, and a Simons Investigator Award.

de MIT News https://ift.tt/5wYKz0l

License plates of MIT

What does your license plate say about you?

In the United States, more than 9 million vehicles carry personalized “vanity” license plates, in which preferred words, digits, or phrases replace an otherwise random assignment of letters and numbers to identify a vehicle. While each state and the District of Columbia maintains its own rules about appropriate selections, creativity reigns when choosing a unique vanity plate. What’s more, the stories behind them can be just as fascinating as the people who use them.

It might not come as a surprise to learn that quite a few MIT community members have participated in such vehicular whimsy. Read on to meet some of them and learn about the nerdy, artsy, techy, and MIT-related plates that color their rides.

A little piece of tech heaven

One of the most recognized vehicles around campus is Samuel Klein’s 1998 Honda Civic. More than just the holder of a vanity plate, it’s an art car — a vehicle that’s been custom-designed as a way to express an artistic idea or theme. Klein’s Civic is covered with hundreds of 5.5-inch floppy disks in various colors, and it sports disks, computer keys, and other techy paraphernalia on the interior. With its double-entendre vanity plate, “DSKDRV” (“disk drive”), the art car initially came into being on the West Coast.

Klein, a longtime affiliate of the MIT Media Lab, MIT Press, and MIT Libraries, first heard about the car from fellow Wikimedian and current MIT librarian Phoebe Ayers. An artistic friend of Ayers’, Lara Wiegand, had designed and decorated the car in Seattle but wanted to find a new owner. Klein was intrigued and decided to fly west to check the Civic out.

“I went out there, spent a whole afternoon seeing how she maintained the car and talking about engineering and mechanisms and the logistics of what’s good and bad,” Klein says. “It had already gone through many iterations.”

Klein quickly decided he was up to the task of becoming the new owner. As he drove the car home across the country, it “got a wide range of really cool responses across different parts of the U.S.”

Back in Massachusetts, Klein made a few adjustments: “We painted the hubcaps, we added racing stripes, we added a new generation of laser-etched glass circuits and, you know, I had my own collection of antiquated technology disks that seemed to fit.”

The vanity plate also required a makeover. In Washington state it was “DISKDRV,” but, Klein says, “we had to shave the license plate a bit because there are fewer letters in Massachusetts.”

Today, the car has about 250,000 miles and an Instagram account. “The biggest challenge is just the disks have to be resurfaced, like a lizard, every few years,” says Klein, whose partner, an MIT research scientist, often parks it around campus. “There’s a small collection of love letters for the car. People leave the car notes. It’s very sweet.”

Marking his place in STEM history

Omar Abudayyeh ’12, PhD ’18, a recent McGovern Fellow at the McGovern Institute for Brain Research at MIT who is now an assistant professor at Harvard Medical School, shares an equally riveting story about his vanity plate, “CRISPR,” which adorns his sport utility vehicle.

The plate refers to the genome-editing technique that has revolutionized biological and medical research by enabling rapid changes to genetic material. As an MIT graduate student in the lab of Professor Feng Zhang, a pioneering contributor to CRISPR technologies, Abudayyeh was highly involved in early CRISPR development for DNA and RNA editing. In fact, he and Jonathan Gootenberg ’13, another recent McGovern Fellow and assistant professor at Harvard Medical School who works closely with Abudayyeh, discovered many novel CRISPR enzymes, such as Cas12 and Cas13, and applied these technologies for both gene therapy and CRISPR diagnostics.

So how did Abudayyeh score his vanity plate? It was all due to his attendance at a genome-editing conference in 2022, where another early-stage CRISPR researcher, Samuel Sternberg, showed up in a car with New York “CRISPR” plates. “It became quite a source of discussion at the conference, and at one of the breaks, Sam and his labmates egged us on to get the Massachusetts license plate,” Abudayyeh explains. “I insisted that it must be taken, but I applied anyway, paying the 70 dollars and then receiving a message that I would get a letter eight to 12 weeks later about whether the plate was available or not. I then returned to Boston and forgot about it until a couple months later when, to my surprise, the plate arrived in the mail.”

While Abudayyeh continues his affiliation with the McGovern Institute, he and Gootenberg recently set up a lab at Harvard Medical School as new faculty members. “We have continued to discover new enzymes, such as Cas7-11, that enable new frontiers, such as programmable proteases for RNA sensing and novel therapeutics, and we’ve applied CRISPR technologies for new efforts in gene editing and aging research,” Abudayyeh notes.

As for his license plate, he says, “I’ve seen instances of people posting about it on Twitter or asking about it in Slack channels. A number of times, people have stopped me to say they read the Walter Isaacson book on CRISPR, asking how I was related to it. I would then explain my story — and describe how I’m actually in the book, in the chapters on CRISPR diagnostics.”

Displaying MIT roots, nerd pride

For some, a connection to MIT is all the reason they need to register a vanity plate — or three. Jeffrey Chambers SM ’06, PhD ’14, a graduate of the Department of Aeronautics and Astronautics, shares that he drives with a Virginia license plate touting his “PHD MIT.” Professor of biology Anthony Sinskey ScD ’67 owns several vehicles sporting vanity plates that honor Course 20, which is today the Department of Biological Engineering but has previously been known by Food Technology, Nutrition and Food Science, and Applied Biological Sciences. Sinskey says he has both “MIT 20” and “MIT XX” plates in Massachusetts and New Hampshire.

At least two MIT couples have had dual vanity plates. Says Laura Kiessling ’83, professor of chemistry: “My plate is ‘SLEX.’ This is the abbreviation for a carbohydrate called sialyl Lewis X. It has many roles, including a role in fertilization (sperm-egg binding). It tends to elicit many different reactions from people asking me what it means. Unless they are scientists, I say that my husband [Ron Raines ’80, professor of biology] gave it to me as an inside joke. My husband’s license plate is ‘PROTEIN.’”

Professor of the practice emerita Marcia Bartusiak of MIT Comparative Media Studies/Writing and her husband, Stephen Lowe PhD ’88, previously shared a pair of related license plates. When the couple lived in Virginia, Lowe working as a mathematician on the structure of spiral galaxies and Bartusiak a young science writer focused on astronomy, they had “SPIRAL” and “GALAXY” plates. Now retired in Massachusetts, while they no longer have registered vanity plates, they’ve named their current vehicles “Redshift” and “Blueshift.”

Still other community members have plates that make a nod to their hobbies — such as Department of Earth, Atmospheric and Planetary Sciences and AeroAstro Professor Sara Seager’s “ICANOE” — or else playfully connect with fellow drivers. Julianna Mullen, communications director in the Plasma Science and Fusion Center, says of her “OMGWHY” plate: “It’s just an existential reminder of the importance of scientific inquiry, especially in traffic when someone cuts you off so they can get exactly two car lengths ahead. Oh my God, why did they do it?”

Are you an MIT affiliate with a unique vanity plate? We’d love to see it!

de MIT News https://ift.tt/2glOWzu