Most work is new work, long-term study of U.S. census data shows

This is part 1 of a two-part MIT News feature examining new job creation in the U.S. since 1940, based on new research from Ford Professor of Economics David Autor. Part 2 is available here.

In 1900, Orville and Wilbur Wright listed their occupations as “Merchant, bicycle” on the U.S. census form. Three years later, they made their famous first airplane flight in Kitty Hawk, North Carolina. So, on the next U.S. census, in 1910, the brothers each called themselves “Inventor, aeroplane.” There weren’t too many of those around at the time, however, and it wasn’t until 1950 that “Airplane designer” became a recognized census category.

Distinctive as their case may be, the story of the Wright brothers tells us something important about employment in the U.S. today. Most work in the U.S. is new work, as U.S. census forms reveal. That is, a majority of jobs are in occupations that have only emerged widely since 1940, according to a major new study of U.S. jobs led by MIT economist David Autor.

“We estimate that about six out of 10 jobs people are doing at present didn’t exist in 1940,” says Autor, co-author of a newly published paper detailing the results. “A lot of the things that we do today, no one was doing at that point. Most contemporary jobs require expertise that didn’t exist back then, and was not relevant at that time.”

This finding, covering the period 1940 to 2018, yields some larger implications. For one thing, many new jobs are created by technology. But not all: Some come from consumer demand, such as health care services jobs for an aging population.

On another front, the research shows a notable divide in recent new-job creation: During the first 40 years of the 1940-2018 period, many new jobs were middle-class manufacturing and clerical jobs, but in the last 40 years, new job creation often involves either highly paid professional work or lower-wage service work.

Finally, the study brings novel data to a tricky question: To what extent does technology create new jobs, and to what extent does it replace jobs?

The paper, “New Frontiers: The Origins and Content of New Work, 1940-2018,” appears in the Quarterly Journal of Economics. The co-authors are Autor, the Ford Professor of Economics at MIT; Caroline Chin, a PhD student in economics at MIT; Anna Salomons, a professor in the School of Economics at Utrecht University; and Bryan Seegmiller SM ’20, PhD ’22, an assistant professor at the Kellogg School of Northwestern University.

“This is the hardest, most in-depth project I’ve ever done in my research career,” Autor adds. “I feel we’ve made progress on things we didn’t know we could make progress on.”

“Technician, fingernail”

To conduct the study, the scholars dug deeply into government data about jobs and patents, using natural language processing techniques that identified related descriptions in patent and census data to link innovations and subsequent job creation. The U.S. Census Bureau tracks the emerging job descriptions that respondents provide — like the ones the Wright brothers wrote down. Each decade’s jobs index lists about 35,000 occupations and 15,000 specialized variants of them.

Many new occupations are straightforwardly the result of new technologies creating new forms of work. For instance, “Engineers of computer applications” was first codified in 1970, “Circuit layout designers” in 1990, and “Solar photovoltaic electrician” made its debut in 2018.

“Many, many forms of expertise are really specific to a technology or a service,” Autor says. “This is quantitatively a big deal.”

He adds: “When we rebuild the electrical grid, we’re going to create new occupations — not just electricians, but the solar equivalent, i.e., solar electricians. Eventually that becomes a specialty. The first objective of our study is to measure [this kind of process]; the second is to show what it responds to and how it occurs; and the third is to show what effect automation has on employment.”

On the second point, however, innovations are not the only way new jobs emerge. The wants and needs of consumers also generate new vocations. As the paper notes, “Tattooers” became a U.S. census job category in 1950, “Hypnotherapists” was codified in 1980, and “Conference planners” in 1990. Also, the date of U.S. Census Bureau codification is not the first time anyone worked in those roles; it is the point at which enough people had those jobs that the bureau recognized the work as a substantial employment category. For instance, “Technician, fingernail” became a category in 2000.

“It’s not just technology that creates new work, it’s new demand,” Autor says. An aging population of baby boomers may be creating new roles for personal health care aides that are only now emerging as plausible job categories.

All told, among “professionals,” essentially specialized white-collar workers, about 74 percent of jobs in the area have been created since 1940. In the category of “health services” — the personal service side of health care, including general health aides, occupational therapy aides, and more — about 85 percent of jobs have emerged in the same time. By contrast, in the realm of manufacturing, that figure is just 46 percent.

Differences by degree

The fact that some areas of employment feature relatively more new jobs than others is one of the major features of the U.S. jobs landscape over the last 80 years. And one of the most striking things about that time period, in terms of jobs, is that it consists of two fairly distinct 40-year periods.

In the first 40 years, from 1940 to about 1980, the U.S. became a singular postwar manufacturing powerhouse, production jobs grew, and middle-income clerical and other office jobs grew up around those industries.

But in the last four decades, manufacturing started receding in the U.S., and automation started eliminating clerical work. From 1980 to the present, there have been two major tracks for new jobs: high-end and specialized professional work, and lower-paying service-sector jobs, of many types. As the authors write in the paper, the U.S. has seen an “overall polarization of occupational structure.”

That corresponds with levels of education. The study finds that employees with at least some college experience are about 25 percent more likely to be working in new occupations than those who possess less than a high school diploma.

“The real concern is for whom the new work has been created,” Autor says. “In the first period, from 1940 to 1980, there’s a lot of work being created for people without college degrees, a lot of clerical work and production work, middle-skill work. In the latter period, it’s bifurcated, with new work for college graduates being more and more in the professions, and new work for noncollege graduates being more and more in services.”

Still, Autor adds, “This could change a lot. We’re in a period of potentially consequential technology transition.”

At the moment, it remains unclear how, and to what extent, evolving technologies such as artificial intelligence will affect the workplace. However, this is also a major issue addressed in the current research study: How much does new technology augment employment, by creating new work and viable jobs, and how much does new technology replace existing jobs, through automation? In their paper, Autor and his colleagues have produced new findings on that topic, which are outlined in part 2 of this MIT News series.

Support for the research was provided, in part, by the Carnegie Corporation; Google; Instituut Gak; the MIT Work of the Future Task Force; Schmidt Futures; the Smith Richardson Foundation; and the Washington Center for Equitable Growth.

de MIT News https://ift.tt/NMC7L9q

A first-ever complete map for elastic strain engineering

Without a map, it can be just about impossible to know not just where you are, but where you’re going, and that’s especially true when it comes to materials properties.

For decades, scientists have understood that while bulk materials behave in certain ways, those rules can break down for materials at the micro- and nano-scales, and often in surprising ways. One of those surprises was the finding that, for some materials, applying even modest strains — a concept known as elastic strain engineering — on materials can dramatically improve certain properties, provided those strains stay elastic and do not relax away by plasticity, fracture, or phase transformations. Micro- and nano-scale materials are especially good at holding applied strains in the elastic form.

Precisely how to apply those elastic strains (or equivalently, residual stress) to achieve certain material properties, however, had been less clear — until recently.

Using a combination of first principles calculations and machine learning, a team of MIT researchers has developed the first-ever map of how to tune crystalline materials to produce specific thermal and electronic properties.

Led by Ju Li, the Battelle Energy Alliance Professor in Nuclear Engineering and professor of materials science and engineering, the team described a framework for understanding precisely how changing the elastic strains on a material can fine-tune properties like thermal and electrical conductivity. The work is described in an open-access paper published in PNAS.

“For the first time, by using machine learning, we’ve been able to delineate the complete six-dimensional boundary of ideal strength, which is the upper limit to elastic strain engineering, and create a map for these electronic and phononic properties,” Li says. “We can now use this approach to explore many other materials. Traditionally, people create new materials by changing the chemistry.”

“For example, with a ternary alloy, you can change the percentage of two elements, so you have two degrees of freedom,” he continues. “What we’ve shown is that diamond, with just one element, is equivalent to a six-component alloy, because you have six degrees of elastic strain freedom you can tune independently.”

Small strains, big material benefits

The paper builds on a foundation laid as far back as the 1980s, when researchers first discovered that the performance of semiconductor materials doubled when a small — just 1 percent — elastic strain was applied to the material.

While that discovery was quickly commercialized by the semiconductor industry and today is used to increase the performance of microchips in everything from laptops to cellphones, that level of strain is very small compared to what we can achieve now, says Subra Suresh, the Vannevar Bush Professor of Engineering Emeritus.

In a 2018 Science paper, Suresh, Dao, and colleagues demonstrated that 1 percent strain was just the tip of the iceberg.

As part of a 2018 study, Suresh and colleagues demonstrated for the first time that diamond nanoneedles could withstand elastic strains of as much as 9 percent and still return to their original state. Later on, several groups independently confirmed that microscale diamond can indeed elastically deform by approximately 7 percent in tension reversibly.

“Once we showed we could bend nanoscale diamonds and create strains on the order of 9 or 10 percent, the question was, what do you do with it,” Suresh says. “It turns out diamond is a very good semiconductor material … and one of our questions was, if we can mechanically strain diamond, can we reduce the band gap from 5.6 electron-volts to two or three? Or can we get it all the way down to zero, where it begins to conduct like a metal?”

To answer those questions, the team first turned to machine learning in an effort to get a more precise picture of exactly how strain altered material properties.

“Strain is a big space,” Li explains. “You can have tensile strain, you can have shear strain in multiple directions, so it’s a six-dimensional space, and the phonon band is three-dimensional, so in total there are nine tunable parameters. So, we’re using machine learning, for the first time, to create a complete map for navigating the electronic and phononic properties and identify the boundaries.”

Armed with that map, the team subsequently demonstrated how strain could be used to dramatically alter diamond’s semiconductor properties.

“Diamond is like the Mt. Everest of electronic materials,” Li says, “because it has very high thermal conductivity, very high dielectric breakdown strengths, a very big carrier mobility. What we have shown is we can controllably squish Mt. Everest down … so we show that by strain engineering you can either improve diamond’s thermal conductivity by a factor of two, or make it much worse by a factor of 20.”

New map, new applications

Going forward, the findings could be used to explore a host of exotic material properties, Li says, from dramatically reduced thermal conductivity to superconductivity.

“Experimentally, these properties are already accessible with nanoneedles and even microbridges,” he says. “And we have seen exotic properties, like reducing diamond’s (thermal conductivity) to only a few hundred watts per meter-Kelvin. Recently, people have shown that you can produce room-temperature superconductors with hydrides if you squeeze them to a few hundred gigapascals, so we have found all kinds of exotic behavior once we have the map.”

The results could also influence the design of next-generation computer chips capable of running much faster and cooler than today’s processors, as well as quantum sensors and communication devices. As the semiconductor manufacturing industry moves to denser and denser architectures, Suresh says the ability to tune a material’s thermal conductivity will be particularly important for heat dissipation.

While the paper could inform the design of future generations of microchips, Zhe Shi, a postdoc in Li’s lab and first author of the paper, says more work will be needed before those chips find their way into the average laptop or cellphone.

“We know that 1 percent strain can give you an order of magnitude increase in the clock speed of your CPU,” Shi says. “There are a lot of manufacturing and device problems that need to be solved in order for this to become realistic, but I think it’s definitely a great start. It’s an exciting beginning to what could lead to significant strides in technology.”

This work was supported with funding from the Defense Threat Reduction Agency, an NSF Graduate Research Fellowship, the Nanyang Technological University School of Biological Sciences, the National Science Foundation (NSF), the MIT Vannevar Bush Professorship, and a Nanyang Technological University Distinguished University Professorship.

de MIT News https://ift.tt/1qxL2Tc

“Life is short, so aim high”

When Rafael Jaramillo talks about his favorite accomplishments, it quickly becomes clear that he has the right temperament for a researcher — he is energized by a challenge and the prospect of hard work.

“I am proudest of things that required risky strategic thinking, followed by years of technical slog, followed by validation,” says Jaramillo, the Thomas Lord Career Development Associate Professor in the Department of Materials Science and Engineering.

Not even the fear of failure deters him. Referring to his work developing new semiconductor materials, he says, “It’s often a fool’s errand to try to replace silicon in any particular application. Time will tell if I spend a career making a fool of myself.”

Of course, Jaramillo is being modest. He has received several significant awards, and in 2021 he and other researchers in his lab succeeded in creating thin, high-quality films using a new family of semiconductor materials, which could be useful in such products as solar cells and environmentally benign LEDs. The materials, called chalcogenide perovskites, are extremely stable and are made of inexpensive, nontoxic elements.

The son of two musicians, Jaramillo grew up attending schools in Brookline, Massachusetts. A second-grade classmate was the son of MIT professor and cosmologist Alan Guth, who volunteered to meet with students in the class and answer their questions about space. Having made a point to check out every library book on space and astronomy, Jaramillo didn’t hold back, and asked Guth about the size of the universe at the earliest stages of the Big Bang.

“He was very kind and patient,” Jaramillo says.

Over the years, Jaramillo’s fascination with space transformed into a love of physics, and he earned his bachelor’s degree in applied and engineering physics at Cornell University and his PhD in physics at the University of Chicago.

Jaramillo says he studied physics “because it satisfied a compulsive need to understand and explain things at a certain level of simplicity.”

“I like physics because I like the methods of physics — the habits of mind, the problem-solving strategies, the experiments,” he says. “Physicists like to tell themselves that they can always figure things out from first principles, and that their field is the opposite of rote memorization.”

A longtime environmentalist with a desire to help society to move beyond reliance on fossil fuels, Jaramillo wanted to focus his knowledge on low-carbon energy after earning his PhD.

“I want to pass on to my kids a world at least as lovely and diverse as I’ve enjoyed and, like most people, I’m worried for the future of our planet,” he says. “Different people can and should bring different disciplinary backgrounds and skillsets to bear on problems of shared importance — it takes a village to solve the hardest ones.”

Jaramillo says that his having switched fields from physics into materials science highlighted some beneficial connections in his work: “I’m sure that some of my ideas, if they contain originality, it’s because I may have a different perspective than others in my field.”

Nonetheless, the switch involved some heavy lifting during two postdocs.

Wanting to engage in solar cell research, “I had to be intentional about seeing postdoc opportunities where I would learn a thing or two about semiconductors, materials science, device optimization, energy technologies, and techno-economies,” Jaramillo says, adding that he read and took notes on hundreds of pages of textbooks “in an attempt to catch up to people around me who always seemed to know more useful things than I did, probably because they did their PhD work in the field.”

Jaramillo conducted postdoctoral research at the Harvard University Center for the Environment and the Harvard School of Engineering and Applied Sciences, as well as at MIT with Tonio Buonassisi, a professor of mechanical engineering and an expert in solar photovoltaics. Jaramillo joined the MIT faculty in 2015 and recently earned tenure.

His current research on new materials could improve the economics and reduce the environmental footprint of semiconductors used in such applications as telecommunications, microelectronics, and photovoltaics.

“We’re butting up against the limitations of the tried-and-true materials,” Jaramillo said in a previous interview with MIT News. “That’s exciting because it means you get to dive in and think about new materials.”

Also exciting to Jaramillo is the increasing worldwide attention devoted to the kind of research he and his lab have been conducting on chalcogenide perovskites for solar cells.

“It used to be a quiet and somewhat lonely field, so I welcome the new community and the competition,” he says. “We took on a lot of risk and delayed gratification for a long time for that project. Now it’s churning out results. If we continue to work quite hard, and if we catch a lot of breaks, it’s possible that chalcogenide perovskite solar cells will contribute meaningfully to the continued expansion of global solar power generation.”

Always the determined researcher, Jaramillo encourages MIT students — who, he is quick to point out, share his high level of motivation — to shoot for the stars.

“I’d say life is short, so aim high,” he says. “As scientists and engineers, that means tackling the hard problems because the easy ones have been solved and, besides, there’s little satisfaction in them.”

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

Atmospheric observations in China show rise in emissions of a potent greenhouse gas

To achieve the aspirational goal of the Paris Agreement on climate change — limiting the increase in global average surface temperature to 1.5 degrees Celsius above preindustrial levels — will require its 196 signatories to dramatically reduce their greenhouse gas (GHG) emissions. Those greenhouse gases differ widely in their global warming potential (GWP), or ability to absorb radiative energy and thereby warm the Earth’s surface. For example, measured over a 100-year period, the GWP of methane is about 28 times that of carbon dioxide (CO₂), and the GWP of sulfur hexafluoride (SF₆) is 24,300 times that of CO₂, according to the Intergovernmental Panel on Climate Change (IPCC) Sixth Assessment Report. 

Used primarily in high-voltage electrical switchgear in electric power grids, SF₆ is one of the most potent greenhouse gases on Earth. In the 21st century, atmospheric concentrations of SF₆ have risen sharply along with global electric power demand, threatening the world’s efforts to stabilize the climate. This heightened demand for electric power is particularly pronounced in China, which has dominated the expansion of the global power industry in the past decade. Quantifying China’s contribution to global SF₆ emissions — and pinpointing its sources in the country — could lead that nation to implement new measures to reduce them, and thereby reduce, if not eliminate, an impediment to the Paris Agreement’s aspirational goal. 

To that end, a new study by researchers at the MIT Joint Program on the Science and Policy of Global Change, Fudan University, Peking University, University of Bristol, and Meteorological Observation Center of China Meteorological Administration determined total SF₆ emissions in China over 2011-21 from atmospheric observations collected from nine stations within a Chinese network, including one station from the Advanced Global Atmospheric Gases Experiment (AGAGE) network. For comparison, global total emissions were determined from five globally distributed, relatively unpolluted “background” AGAGE stations, involving additional researchers from the Scripps Institution of Oceanography and CSIRO, Australia's National Science Agency.

The researchers found that SF₆ emissions in China almost doubled from 2.6 gigagrams (Gg) per year in 2011, when they accounted for 34 percent of global SF₆ emissions, to 5.1 Gg per year in 2021, when they accounted for 57 percent of global total SF₆ emissions. This increase from China over the 10-year period — some of it emerging from the country’s less-populated western regions — was larger than the global total SF₆ emissions rise, highlighting the importance of lowering SF₆ emissions from China in the future.

The open-access study, which appears in the journal Nature Communications, explores prospects for future SF₆ emissions reduction in China.

“Adopting maintenance practices that minimize SF₆ leakage rates or using SF₆-free equipment or SF₆ substitutes in the electric power grid will benefit greenhouse-gas mitigation in China,” says Minde An, a postdoc at the MIT Center for Global Change Science (CGCS) and the study’s lead author. “We see our findings as a first step in quantifying the problem and identifying how it can be addressed.”

Emissions of SF₆ are expected to last more than 1,000 years in the atmosphere, raising the stakes for policymakers in China and around the world.

“Any increase in SF₆ emissions this century will effectively alter our planet’s radiative budget — the balance between incoming energy from the sun and outgoing energy from the Earth — far beyond the multi-decadal time frame of current climate policies,” says MIT Joint Program and CGCS Director Ronald Prinn, a coauthor of the study. “So it’s imperative that China and all other nations take immediate action to reduce, and ultimately eliminate, their SF₆ emissions.”

The study was supported by the National Key Research and Development Program of China and Shanghai B&R Joint Laboratory Project, the U.S. National Aeronautics and Space Administration, and other funding agencies.  

de MIT News https://ift.tt/M16Rygn

VIAVI Solutions joins MIT.nano Consortium

VIAVI Solutions, a global provider of communications test and measurement and optical technologies, has joined the MIT.nano Consortium.

With roots going back to 1923 as Wandell and Goltermann and to 1948 as Optical Coating Laboratory Inc., VIAVI is a global enterprise supporting innovation in communication networks, hyperscale and enterprise data centers, consumer electronics, automotive sensing, mission-critical avionics, aerospace, and anti-counterfeiting technologies.

“VIAVI is an exciting new member of the MIT.nano Consortium. The company’s innovations overlap with MIT’s research interests in a variety of applications — electronics, 3D sensing, optics, data analysis, artificial intelligence, and more,” says Vladimir Bulović, the founding faculty director of MIT.nano and the Fariborz Maseeh (1990) Professor of Emerging Technologies. “VIAVI’s awareness of industry needs will make them a valuable collaborator as we at MIT.nano work to develop new technologies in the lab that can successfully transition to the real world.”

With over 3,600 employees in 22 countries, VIAVI is poised to contribute global insights to the MIT.nano Consortium and MIT research community.

“VIAVI is delighted to be part of the extraordinary MIT.nano ecosystem,” says Oleg Khaykin, president and CEO of VIAVI. “MIT.nano occupies a unique position at the intersection of academia, industry, and government. We look forward to collaborating with the organization and its stakeholders focused on innovation in materials and processes that will enable the photonics applications of the future.”

The MIT.nano Consortium is a platform for academia-industry collaboration centered around research and innovation emerging from nanoscale science and engineering at MIT. Through activities that include quarterly industry consortium meetings, VIAVI will gain insight into the work of MIT.nano’s community of users and provide advice to help guide and advance nanoscale innovations at MIT alongside the 11 other consortium companies:

• Analog Devices
• Edwards
• Fujikura
• IBM Research
• Lam Research
• Lockheed Martin
• NC
• NEC
• Raith
• Shell
• UpNano

MIT.nano continues to welcome new companies as sustaining members. For more details, visit the MIT.nano Consortium page.

de MIT News https://ift.tt/wFQb0pC

Second round of seed grants awarded to MIT scholars studying the impact and applications of generative AI

Last summer, MIT President Sally Kornbluth and Provost Cynthia Barnhart issued a call for papers to “articulate effective roadmaps, policy recommendations, and calls for action across the broad domain of generative AI.” The response to the call far exceeded expectations with 75 proposals submitted. Of those, 27 proposals were selected for seed funding.

In light of this enthusiastic response, Kornbluth and Barnhart announced a second call for proposals this fall.

“The groundswell of interest and the caliber of the ideas overall made clear that a second round was in order,” they said in their email to MIT’s research community this fall. This second call for proposals resulted in 53 submissions.

Following the second call, the faculty committee from the first round considered the proposals and selected 16 proposals to receive exploratory funding. Co-authored by interdisciplinary teams of faculty and researchers affiliated with all five of the Institute’s schools and the MIT Schwarzman College of Computing, the proposals offer insights and perspectives on the potential impact and applications of generative AI across a broad range of topics and disciplines.

Each selected research group will receive between $50,000 and $70,000 to create 10-page impact papers. Those papers will be shared widely via a publication venue managed and hosted by the MIT Press under the auspices of the MIT Open Publishing Services program.

As with the first round of papers, Thomas Tull, a member of the MIT School of Engineering Dean’s Advisory Council and a former innovation scholar at the School of Engineering, contributed funding to support the effort.

The selected papers are:

• “A Road-map for End-to-end Privacy and Verifiability in Generative AI,” led by Alex Pentland, Srini Devadas, Lalana Kagal, and Vinod Vaikuntanathan;
• “A Virtuous Cycle: Generative AI and Discovery in the Physical Sciences,” led by Philip Harris and Phiala Shanahan;
• “Artificial Cambrian Intelligence: Generating New Forms of Visual Intelligence,” led by Ramesh Raskar and Tomaso A. Poggio;
• “Artificial Fictions and the Value of AI-Generated Art,” led by Justin Khoo;
• “GenAI for Improving Human-to-human Interactions with a Focus on Negotiations,” led by Lawrence Susskind;
• “Generative AI as a New Applications Platform and Ecosystem,” led by Michael Cusumano;
• “Generative AI for Cities: A Civic Engagement Playbook,” led by Sarah Williams, Sara Beery, and Eden Medina;
• “Generative AI for Textile Engineering: Advanced Materials from Heritage Lace Craft,” led by Svetlana V. Boriskina;
• “Generative AI Impact for Biomedical Innovation and Drug Discovery,” led by Manolis Kellis, Brad Pentelute, and Marinka Zitnik;
• “Impact of Generative AI on the Creative Economy,” led by Ashia Wilson and Dylan Hadfield-Menell;
• “Redefining Virtuosity: The Role of Generative AI in Live Music Performances,” led by Joseph A. Paradiso and Eran Egozy;
• “Reflection-based Learning with Generative AI,” led by Stefanie Mueller;
• “Robust and Reliable Systems for Generative AI,” led by Shafi Goldwasser, Yael Kalai, and Vinod Vaikuntanathan;
• “Supporting the Aging Population with Generative AI,” led by Pattie Maes;
• “The Science of Language in the Era of Generative AI,” led by Danny Fox, Yoon Kim, and Roger Levy; and
• “Visual Artists, Technological Shock, and Generative AI,” led by Caroline Jones and Huma Gupta.

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

Is it the school, or the students?

Are schools that feature strong test scores highly effective, or do they mostly enroll students who are already well-prepared for success? A study co-authored by MIT scholars concludes that widely disseminated school quality ratings reflect the preparation and family background of their students as much or more than a school’s contribution to learning gains.

Indeed, the study finds that many schools that receive relatively low ratings perform better than these ratings would imply. Conventional ratings, the research makes clear, are highly correlated with race. Specifically, many published school ratings are highly positively correlated with the share of the student body that is white.

“A school’s average outcomes reflect, to some extent, the demographic mix of the population it serves,” says MIT economist Josh Angrist, a Nobel Prize winner who has long analyzed education outcomes. Angrist is co-author of a newly published paper detailing the study’s results.

The study, which examines the Denver and New York City school districts, has the potential to significantly improve the way school quality is measured. Instead of raw aggregate measures like test scores, the study uses changes in test scores and a statistical adjustment for racial composition to compute more accurate measures of the causal effects that attending a particular school has on students’ learning gains. This methodologically sophisticated research builds on the fact that Denver and New York City both assign students to schools in ways that allow the researchers to mimic the conditions of a randomized trial.

In documenting a strong correlation between currently used rating systems and race, the study finds that white and Asian students tend to attend higher-rated schools, while Black and Hispanic students tend to be clustered at lower-rated schools.

“Simple measures of school quality, which are based on the average statistics for the school, are invariably highly correlated with race, and those measures tend to be a misleading guide of what you can expect by sending your child to that school,” Angrist says.

The paper, “Race and the Mismeasure of School Quality,” appears in the latest issue of the American Economic Review: Insights. The authors are Angrist, the Ford Professor of Economics at MIT; Peter Hull, a professor of economics at Brown University; Parag Pathak, the Class of 1922 Professor of Economics at MIT; and Christopher Walters PhD ’13, an associate professor of economics at the University of California at Berkeley. Angrist and Pathak are both professors in the MIT Department of Economics and co-founders of MIT’s Blueprint Labs, a research group that often examines school performance.

The study uses data provided by the Denver and New York City public school districts, where 6th-graders apply for seats at certain middle schools, and the districts use a school-assignment system. In these districts, students can opt for any school in the district, but some schools are oversubscribed. In these circumstances, the district uses a random lottery number to determine who gets a seat where.

By virtue of the lottery inside the seat-assignment algorithm, otherwise-similar sets of students randomly attend an array of different schools. This facilitates comparisons that reveal causal effects of school attendance on learning gains, as in a randomized clinical trial of the sort used in medical research. Using math and English test scores, the researchers evaluated student progress in Denver from the 2012-2013 through the 2018-2019 school years, and in New York City from the 2016-2017 through 2018-2019 school years.

Those school-assignment systems, it happens, are mechanisms some of the researchers have helped construct, allowing them to better grasp and measure the effects of school assignment.

“An unexpected dividend of our work designing Denver and New York City’s centralized choice systems is that we see how students are rationed from [distributed among] schools,” says Pathak. “This leads to a research design that can isolate cause and effect.”

Ultimately, the study shows that much of the school-to-school variation in raw aggregate test scores stems from the types of students at any given school. This is a case of what researchers call “selection bias.” In this case, selection bias arises from the fact that more-advantaged families tend to prefer the same sets of schools.

“The fundamental problem here is selection bias,” Angrist says. “In the case of schools, selection bias is very consequential and a big part of American life. A lot of decision-makers, whether they’re families or policymakers, are being misled by a kind of naïve interpretation of the data.”

Indeed, Pathak notes, the preponderance of more simplistic school ratings today (found on many popular websites) not only creates a deceptive picture of how much value schools add for students, but has a self-reinforcing effect — since well-prepared and better-off families bid up housing costs near highly-rated schools. As the scholars write in the paper, “Biased rating schemes direct households to low-minority rather than high-quality schools, while penalizing schools that improve achievement for disadvantaged groups.”

The research team hopes their study will lead districts to examine and improve the way they measure and report on school quality. To that end, Blueprint Labs is working with the New York City Department of Education to pilot a new ratings system later this year. They also plan additional work examining the way families respond to different sorts of information about school quality.

Given that the researchers are proposing to improve ratings in what they believe is a straightforward way, by accounting for student preparation and improvement, they think more officials and districts may be interested in updating their measurement practices.

“We’re hopeful that the simple regression adjustment we propose makes it relatively easy for school districts to use our measure in practice,” Pathak says.

The research received support from the Walton Foundation and the National Science Foundation.

de MIT News https://ift.tt/iy4op8e

Persistent “hiccups” in a far-off galaxy draw astronomers to new black hole behavior

At the heart of a far-off galaxy, a supermassive black hole appears to have had a case of the hiccups.

Astronomers from MIT, Italy, the Czech Republic, and elsewhere have found that a previously quiet black hole, which sits at the center of a galaxy about 800 million light years away, has suddenly erupted, giving off plumes of gas every 8.5 days before settling back to its normal, quiet state.

The periodic hiccups are a new behavior that has not been observed in black holes until now. The scientists believe the most likely explanation for the outbursts stems from a second, smaller black hole that is zinging around the central, supermassive black hole and slinging material out from the larger black hole’s disk of gas every 8.5 days.

The team’s findings, which are published today in the journal Science Advances, challenge the conventional picture of black hole accretion disks, which scientists had assumed are relatively uniform disks of gas that rotate around a central black hole. The new results suggest that accretion disks may be more varied in their contents, possibly containing other black holes and even entire stars.

“We thought we knew a lot about black holes, but this is telling us there are a lot more things they can do,” says study author Dheeraj “DJ” Pasham, a research scientist in MIT’s Kavli Institute for Astrophysics and Space Research. “We think there will be many more systems like this, and we just need to take more data to find them.”

The study’s MIT co-authors include postdoc Peter Kosec, graduate student Megan Masterson, Associate Professor Erin Kara, Principal Research Scientist Ronald Remillard, and former research scientist Michael Fausnaugh, along with collaborators from multiple institutions, including the Tor Vergata University of Rome, the Astronomical Institute of the Czech Academy of Sciences, and Masaryk University in the Czech Republic.

“Use it or lose it”

The team’s findings grew out of an automated detection by ASAS-SN (the All Sky Automated Survey for SuperNovae), a network of 20 robotic telescopes situated in various locations across the Northern and Southern Hemispheres. The telescopes automatically survey the entire sky once a day for signs of supernovae and other transient phenomena.

In December of 2020, the survey spotted a burst of light in a galaxy about 800 million light years away. That particular part of the sky had been relatively quiet and dark until the telescopes’ detection, when the galaxy suddenly brightened by a factor of 1,000. Pasham, who happened to see the detection reported in a community alert, chose to focus in on the flare with NASA’s NICER (the Neutron star Interior Composition Explorer), an X-ray telescope aboard the International Space Station that continuously monitors the sky for X-ray bursts that could signal activity from neutron stars, black holes, and other extreme gravitational phenomena. The timing was fortuitous, as it was getting toward the end of the yearlong period during which Pasham had permission to point, or “trigger,” the telescope.

“It was either use it or lose it, and it turned out to be my luckiest break,” he says.

He trained NICER to observe the far-off galaxy as it continued to flare. The outburst lasted about four months before petering out. During that time, NICER took measurements of the galaxy’s X-ray emissions on a daily, high-cadence basis. When Pasham looked closely at the data, he noticed a curious pattern within the four-month flare: subtle dips, in a very narrow band of X-rays, that seemed to reappear every 8.5 days.

It seemed that the galaxy’s burst of energy periodically dipped every 8.5 days. The signal is similar to what astronomers see when an orbiting planet crosses in front of its host star, briefly blocking the star’s light. But no star would be able to block a flare from an entire galaxy.

“I was scratching my head as to what this means because this pattern doesn’t fit anything that we know about these systems,” Pasham recalls.

Punch it

As he was looking for an explanation to the periodic dips, Pasham came across a recent paper by theoretical physicists in the Czech Republic. The theorists had separately worked out that it would be possible, in theory, for a galaxy’s central supermassive black hole to host a second, much smaller black hole. That smaller black hole could orbit at an angle from its larger companion’s accretion disk.

As the theorists proposed, the secondary would periodically punch through the primary black hole’s disk as it orbits. In the process, it would release a plume of gas, like a bee flying through a cloud of pollen. Powerful magnetic fields, to the north and south of the black hole, could then slingshot the plume up and out of the disk. Each time the smaller black hole punches through the disk, it would eject another plume, in a regular, periodic pattern. If that plume happened to point in the direction of an observing telescope, it might observe the plume as a dip in the galaxy’s overall energy, briefly blocking the disk’s light every so often.

“I was super excited by this theory, and I immediately emailed them to say, ‘I think we’re observing exactly what your theory predicted,’” Pasham says.

He and the Czech scientists teamed up to test the idea, with simulations that incorporated NICER’s observations of the original outburst, and the regular, 8.5-day dips. What they found supports the theory: The observed outburst was likely a signal of a second, smaller black hole, orbiting a central supermassive black hole, and periodically puncturing its disk.

Specifically, the team found that the galaxy was relatively quiet prior to the December 2020 detection. The team estimates the galaxy’s central supermassive black hole is as massive as 50 million suns. Prior to the outburst, the black hole may have had a faint, diffuse accretion disk rotating around it, as a second, smaller black hole, measuring 100 to 10,000 solar masses, was orbiting in relative obscurity.

The researchers suspect that, in December 2020, a third object — likely a nearby star — swung too close to the system and was shredded to pieces by the supermassive black hole’s immense gravity — an event that astronomers know as a “tidal disruption event.” The sudden influx of stellar material momentarily brightened the black hole’s accretion disk as the star’s debris swirled into the black hole. Over four months, the black hole feasted on the stellar debris as the second black hole continued orbiting. As it punched through the disk, it ejected a much larger plume than it normally would, which happened to eject straight out toward NICER’s scope.

The team carried out numerous simulations to test the periodic dips. The most likely explanation, they conclude, is a new kind of David-and-Goliath system — a tiny, intermediate-mass black hole, zipping around a supermassive black hole.

“This is a different beast,” Pasham says. “It doesn’t fit anything that we know about these systems. We’re seeing evidence of objects going in and through the disk, at different angles, which challenges the traditional picture of a simple gaseous disk around black holes. We think there is a huge population of these systems out there.”

“This is a brilliant example of how to use the debris from a disrupted star to illuminate the interior of a galactic nucleus which would otherwise remain dark. It is akin to using fluorescent dye to find a leak in a pipe,” says Richard Saxton, an X-ray astronomer from the European Space Astronomy Centre (ESAC) in Madrid, who was not involved in the study. “This result shows that very close super-massive black hole binaries could be common in galactic nuclei, which is a very exciting development for future gravitational wave detectors.”

This research was supported, in part, by NASA.

de MIT News https://ift.tt/ZJCjV1d

Unlocking mRNA’s cancer-fighting potential

What if training your immune system to attack cancer cells was as easy as training it to fight Covid-19? Many people believe the technology behind some Covid-19 vaccines, messenger RNA, holds great promise for stimulating immune responses to cancer.

But using messenger RNA, or mRNA, to get the immune system to mount a prolonged and aggressive attack on cancer cells — while leaving healthy cells alone — has been a major challenge.

The MIT spinout Strand Therapeutics is attempting to solve that problem with an advanced class of mRNA molecules that are designed to sense what type of cells they encounter in the body and to express therapeutic proteins only once they have entered diseased cells.

“It’s about finding ways to deal with the signal-to-noise ratio, the signal being expression in the target tissue and the noise being expression in the nontarget tissue,” Strand CEO Jacob Becraft PhD ’19 explains. “Our technology amplifies the signal to express more proteins for longer while at the same time effectively eliminating the mRNA’s off-target expression.”

Strand is set to begin its first clinical trial in April, which is testing a proprietary, self-replicating mRNA molecule’s ability to express immune signals directly from a tumor, eliciting the immune system to attack and kill the tumor cells directly. It’s also being tested as a possible improvement for existing treatments to a number of solid tumors.

As they work to commercialize its early innovations, Strand’s team is continuing to add capabilities to what it calls its “programmable medicines,” improving mRNA molecules’ ability to sense their environment and generate potent, targeted responses where they’re needed most.

“Self-replicating mRNA was the first thing that we pioneered when we were at MIT and in the first couple years at Strand,” Becraft says. “Now we’ve also moved into approaches like circular mRNAs, which allow each molecule of mRNA to express more of a protein for longer, potentially for weeks at a time. And the bigger our cell-type specific datasets become, the better we are at differentiating cell types, which makes these molecules so targeted we can have a higher level of safety at higher doses and create stronger treatments.”

Making mRNA smarter

Becraft got his first taste of MIT as an undergraduate at the University of Illinois when he secured a summer internship in the lab of MIT Institute Professor Bob Langer.

“That’s where I learned how lab research could be translated into spinout companies,” Becraft recalls.

The experience left enough of an impression on Becraft that he returned to MIT the next fall to earn his PhD, where he worked in the Synthetic Biology Center under professor of bioengineering and electrical engineering and computer science Ron Weiss. During that time, he collaborated with postdoc Tasuku Kitada to create genetic “switches” that could control protein expression in cells.

Becraft and Kitada realized their research could be the foundation of a company around 2017 and started spending time in the Martin Trust Center for MIT Entrepreneurship. They also received support from MIT Sandbox and eventually worked with the Technology Licensing Office to establish Strand’s early intellectual property.

“We started by asking, where is the highest unmet need that also allows us to prove out the thesis of this technology? And where will this approach have therapeutic relevance that is a quantum leap forward from what anyone else is doing?” Becraft says. “The first place we looked was oncology.”

People have been working on cancer immunotherapy, which turns a patient’s immune system against cancer cells, for decades. Scientists in the field have developed drugs that produce some remarkable results in patients with aggressive, late-stage cancers. But most next-generation cancer immunotherapies are based on recombinant (lab-made) proteins that are difficult to deliver to specific targets in the body and don’t remain active for long enough to consistently create a durable response.

More recently, companies like Moderna, whose founders also include MIT alumni, have pioneered the use of mRNAs to create proteins in cells. But to date, those mRNA molecules have not been able to change behavior based on the type of cells they enter, and don’t last for very long in the body.

“If you’re trying to engage the immune system with a tumor cell, the mRNA needs to be expressing from the tumor cell itself, and it needs to be expressing over a long period of time,” Becraft says. “Those challenges are hard to overcome with the first generation of mRNA technologies.”

Strand has developed what it calls the world’s first mRNA programming language that allows the company to specify the tissues its mRNAs express proteins in.

“We built a database that says, ‘Here are all of the different cells that the mRNA could be delivered to, and here are all of their microRNA signatures,’ and then we use computational tools and machine learning to differentiate the cells,” Becraft explains. “For instance, I need to make sure that the messenger RNA turns off when it's in the liver cell, and I need to make sure that it turns on when it's in a tumor cell or a T-cell.”

Strand also uses techniques like mRNA self-replication to create more durable protein expression and immune responses.

“The first versions of mRNA therapeutics, like the Covid-19 vaccines, just recapitulate how our body’s natural mRNAs work,” Becraft explains. “Natural mRNAs last for a few days, maybe less, and they express a single protein. They have no context-dependent actions. That means wherever the mRNA is delivered, it’s only going to express a molecule for a short period of time. That’s perfect for a vaccine, but it’s much more limiting when you want to create a protein that’s actually engaging in a biological process, like activating an immune response against a tumor that could take many days or weeks.”

Technology with broad potential

Strand’s first clinical trial is targeting solid tumors like melanoma and triple-negative breast cancer. The company is also actively developing mRNA therapies that could be used to treat blood cancers.

“We’ll be expanding into new areas as we continue to de-risk the translation of the science and create new technologies,” Becraft says.

Strand plans to partner with large pharmaceutical companies as well as investors to continue developing drugs. Further down the line, the founders believe future versions of its mRNA therapies could be used to treat a broad range of diseases.

“Our thesis is: amplified expression in specific, programmed target cells for long periods of time,” Becraft says. “That approach can be utilized for [immunotherapies like] CAR T-cell therapy, both in oncology and autoimmune conditions. There are also many diseases that require cell-type specific delivery and expression of proteins in treatment, everything from kidney disease to types of liver disease. We can envision our technology being used for all of that.”

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

A revolutionary, bold educational endeavor for Belize

When 14-year-old Jahzhia Moralez played a vocabulary game that involved jumping onto her friend like a backpack, she knew Itz'at STEAM Academy wasn’t like other schools in Belize. Transferring from a school that assigned nearly four hours of homework every night, Moralez found it strange that her first week at Itz'at was focused on having fun. 

“I was very excited,” Moralez says. “I want to be an architect or a vet, and this school has the curriculum for that and other technology-based stuff.”

The name “Itz’at” translates to “wise one” in Maya, honoring the local culture that studied mathematics and astronomy for over a thousand years. Launched in September 2023, Itz’at STEAM Academy is a secondary school that prepares students between the ages of 13 and 16 to build sustainable futures for themselves and their communities, using science, technology, engineering, arts, and mathematics (STEAM). The school’s mission is to create a diverse and inclusive community for all, especially girls, students with special educational needs, and learners from marginalized social, economic, and cultural groups.

The school’s launch is the culmination of a three-year project between MIT and the Ministry of Education, Culture, Science, and Technology of Belize. “The Itz’at STEAM Academy represents a revolutionary and bold educational endeavor for us in Belize,” a ministry representative says. “Serving as an institution championing the pedagogy of STEAM through inventive and imaginative methodologies, its primary aim is to push the boundaries of educational norms within our nation.”

Itz’at is one of the first Belizean schools to use competency-based programs and individualized, authentic learning experiences. The Itz’at pedagogical framework was co-created by MIT pK-12 — part of MIT Open Learning — with members of the ministry and the school. The framework’s foundation has three core pillars: social-emotional and cultural learning, transdisciplinary academics, and community engagement.

“The school's core pillars inform the students' growth and development by fostering empathy, cultural awareness, strong interpersonal skills, holistic thinking, and a sense of responsibility and civic-mindedness,” says Vice Principal Christine Coc.

Building student confidence and connecting with community

The teaching and learning framework developed for Itz’at is rooted in proven learning science research. A student-centered, hands-on learning approach helps students develop critical thinking, creativity, and problem-solving skills. 

“The curriculum places emphasis on fostering student competence and cultivating a culture where it's acceptable not to have all the answers,” says teacher Lionel Palacio.

Instead of measuring students’ understanding through tests and quizzes, which focus on memorization of content, teachers assess each stage of students’ project-based work. Teachers are reporting increased student engagement and deeper understanding of concepts.

“It’s like night and day,” says Moralez’s father, Alejandro. “I enjoy seeing her happy while working on a project. She’s not too stressed.”

The transdisciplinary approach encourages students to think beyond the boundaries of traditional school subjects. This holistic educational experience reinforces students’ understanding. For example, Moralez first learned about conversions in her Quantitative Reasoning course, and later applied that knowledge to convert centimeters to kilometers for a Belizean Studies project.

Students are also encouraged to consider their roles in and outside of school through community engagement initiatives. Connections with outside organizations like the Belize Zoo and the Belize Institute of Archaeology open avenues for collaboration and mutual growth.

“We have seen a positive impact on students’ confidence and self-esteem as they take on challenges and see the real-world relevance of their learning,” says Coc. 

Assignments that engage in real-world problem-solving are practical, offering students insight into future careers. The school aims to create career pathways to strengthen Belize’s existing industries, such as agriculture and food systems, while also supporting the development of new ones, such as cybersecurity.

Students’ sense of belonging is readily apparent to teachers, which positively correlates with their learning. “There's a noticeable companionship among students, with a willingness to assist one another and an openness to the novel learning approach,” says Palacio.

Parents see the impact of the safe learning environment that Itz’at creates for their children. Izaya Lovell, for example, gets to embrace his whole self. “I get to speak my mother tongue, Kriol,” he says. “I can be like my dad — get dreads and grow out my hair. I can play sports and be physical.”

Izaya’s mother, Odessa Lovell, says her son was a completely different person after one month of studying at Itz’at. “He’s so independent, he’s saving money, and he’s doing things on his own,” she says.

A vision for Belize

The development of Itz’at emerged from a 2019 agreement between MIT's Abdul Latif Jameel World Education Lab (J-WEL) and the ministry for the implementation of a STEAM laboratory school in Belize, with funding from the Inter-American Development Bank. MIT had a proven track record of projects and partnerships that transformed education globally. For example, MIT collaborated with administrators in India, which trained 3,300 teachers to launch a large-scale education system focusing on hands-on learning and competencies in values, citizenship, and professional skills that would prepare Indian students for further academic studies or the workforce. The Belize program is the first time that groups across the Institute have come together to develop a school from the ground up, and MIT pK-12 led the charge.

“One of the key aspects of the project has been the approach to co-design and co-creation of the school,” says Claudia Urrea, principal investigator for the Itz’at project at MIT and senior associate director of MIT pK-12. “This approach has not only allowed us to create a relevant school for the country, but to build the local capacity for innovation to sustain beyond the time of the project.”

Working with an extended team at MIT and stakeholders from the ministry, the school, parents, the community, and businesses, Urrea oversaw the development of the school’s mission, vision, values, governance structure, and internship program. The MIT pK-12 team — Urrea; Emily Glass, senior learning innovation designer; and Joe Diaz, program coordinator — led a collaborative effort on the school’s pedagogical framework and curriculum. Other core MIT team members include Brandon Muramatsu, associate director of special projects at Open Learning, and Judy Perry, director of the MIT Scheller Teacher Education Program, who created operational guidance for finances, policies, and teacher professional development. By sharing insights with J-WEL, the MIT pK-12 team is fueling shared thinking and innovations that improve students’ learning and pathways from early to higher education to the workforce. 

Like the students, this is the Belizean teachers’ first experience with project-based learning. The MIT team shared the skills, mindsets, and practical training needed to achieve the school’s core values. The professional development training was designed to build their capacity, so they feel confident teaching this model to students and future educators. 

Itz’at currently has 64 students, with plans to reach full capacity of 300 students by 2026. The goal is to continue to build capacity toward STEAM education in the country, expand the possibilities available to students after graduation, and foster a robust school-to-career pipeline. 

“The opening of this school marks a pioneering milestone not just within Belize but also across the broader Central American and Caribbean regions,” a ministry spokesperson says. “We are excited about the future of Itz’at STEAM Academy and the success of its students.”

de MIT News https://ift.tt/qwi3CgU

MIT-derived algorithm helps forecast the frequency of extreme weather

To assess a community’s risk of extreme weather, policymakers rely first on global climate models that can be run decades, and even centuries, forward in time, but only at a coarse resolution. These models might be used to gauge, for instance, future climate conditions for the northeastern U.S., but not specifically for Boston.

To estimate Boston’s future risk of extreme weather such as flooding, policymakers can combine a coarse model’s large-scale predictions with a finer-resolution model, tuned to estimate how often Boston is likely to experience damaging floods as the climate warms. But this risk analysis is only as accurate as the predictions from that first, coarser climate model.

“If you get those wrong for large-scale environments, then you miss everything in terms of what extreme events will look like at smaller scales, such as over individual cities,” says Themistoklis Sapsis, the William I. Koch Professor and director of the Center for Ocean Engineering in MIT’s Department of Mechanical Engineering.

Sapsis and his colleagues have now developed a method to “correct” the predictions from coarse climate models. By combining machine learning with dynamical systems theory, the team’s approach “nudges” a climate model’s simulations into more realistic patterns over large scales. When paired with smaller-scale models to predict specific weather events such as tropical cyclones or floods, the team’s approach produced more accurate predictions for how often specific locations will experience those events over the next few decades, compared to predictions made without the correction scheme.

Sapsis says the new correction scheme is general in form and can be applied to any global climate model. Once corrected, the models can help to determine where and how often extreme weather will strike as global temperatures rise over the coming years. 

“Climate change will have an effect on every aspect of human life, and every type of life on the planet, from biodiversity to food security to the economy,” Sapsis says. “If we have capabilities to know accurately how extreme weather will change, especially over specific locations, it can make a lot of difference in terms of preparation and doing the right engineering to come up with solutions. This is the method that can open the way to do that.”

The team’s results appear today in the Journal of Advances in Modeling Earth Systems. The study’s MIT co-authors include postdoc Benedikt Barthel Sorensen and Alexis-Tzianni Charalampopoulos SM ’19, PhD ’23, with Shixuan Zhang, Bryce Harrop, and Ruby Leung of the Pacific Northwest National Laboratory in Washington state.

Over the hood

Today’s large-scale climate models simulate weather features such as the average temperature, humidity, and precipitation around the world, on a grid-by-grid basis. Running simulations of these models takes enormous computing power, and in order to simulate how weather features will interact and evolve over periods of decades or longer, models average out features every 100 kilometers or so.

“It’s a very heavy computation requiring supercomputers,” Sapsis notes. “But these models still do not resolve very important processes like clouds or storms, which occur over smaller scales of a kilometer or less.”

To improve the resolution of these coarse climate models, scientists typically have gone under the hood to try and fix a model’s underlying dynamical equations, which describe how phenomena in the atmosphere and oceans should physically interact.

“People have tried to dissect into climate model codes that have been developed over the last 20 to 30 years, which is a nightmare, because you can lose a lot of stability in your simulation,” Sapsis explains. “What we’re doing is a completely different approach, in that we’re not trying to correct the equations but instead correct the model’s output.”

The team’s new approach takes a model’s output, or simulation, and overlays an algorithm that nudges the simulation toward something that more closely represents real-world conditions. The algorithm is based on a machine-learning scheme that takes in data, such as past information for temperature and humidity around the world, and learns associations within the data that represent fundamental dynamics among weather features. The algorithm then uses these learned associations to correct a model’s predictions.

“What we’re doing is trying to correct dynamics, as in how an extreme weather feature, such as the windspeeds during a Hurricane Sandy event, will look like in the coarse model, versus in reality,” Sapsis says. “The method learns dynamics, and dynamics are universal. Having the correct dynamics eventually leads to correct statistics, for example, frequency of rare extreme events.”

Climate correction

As a first test of their new approach, the team used the machine-learning scheme to correct simulations produced by the Energy Exascale Earth System Model (E3SM), a climate model run by the U.S. Department of Energy, that simulates climate patterns around the world at a resolution of 110 kilometers. The researchers used eight years of past data for temperature, humidity, and wind speed to train their new algorithm, which learned dynamical associations between the measured weather features and the E3SM model. They then ran the climate model forward in time for about 36 years and applied the trained algorithm to the model’s simulations. They found that the corrected version produced climate patterns that more closely matched real-world observations from the last 36 years, not used for training.

“We’re not talking about huge differences in absolute terms,” Sapsis says. “An extreme event in the uncorrected simulation might be 105 degrees Fahrenheit, versus 115 degrees with our corrections. But for humans experiencing this, that is a big difference.”

When the team then paired the corrected coarse model with a specific, finer-resolution model of tropical cyclones, they found the approach accurately reproduced the frequency of extreme storms in specific locations around the world.

“We now have a coarse model that can get you the right frequency of events, for the present climate. It’s much more improved,” Sapsis says. “Once we correct the dynamics, this is a relevant correction, even when you have a different average global temperature, and it can be used for understanding how forest fires, flooding events, and heat waves will look in a future climate. Our ongoing work is focusing on analyzing future climate scenarios.”

“The results are particularly impressive as the method shows promising results on E3SM, a state-of-the-art climate model,” says Pedram Hassanzadeh, an associate professor who leads the Climate Extremes Theory and Data group at the University of Chicago and was not involved with the study. “It would be interesting to see what climate change projections this framework yields once future greenhouse-gas emission scenarios are incorporated.”

This work was supported, in part, by the U.S. Defense Advanced Research Projects Agency.

de MIT News https://ift.tt/ZVtbq0W

Artificial reef designed by MIT engineers could protect marine life, reduce storm damage

The beautiful, gnarled, nooked-and-crannied reefs that surround tropical islands serve as a marine refuge and natural buffer against stormy seas. But as the effects of climate change bleach and break down coral reefs around the world, and extreme weather events become more common, coastal communities are left increasingly vulnerable to frequent flooding and erosion.

An MIT team is now hoping to fortify coastlines with “architected” reefs — sustainable, offshore structures engineered to mimic the wave-buffering effects of natural reefs while also providing pockets for fish and other marine life.

The team’s reef design centers on a cylindrical structure surrounded by four rudder-like slats. The engineers found that when this structure stands up against a wave, it efficiently breaks the wave into turbulent jets that ultimately dissipate most of the wave’s total energy. The team has calculated that the new design could reduce as much wave energy as existing artificial reefs, using 10 times less material.

The researchers plan to fabricate each cylindrical structure from sustainable cement, which they would mold in a pattern of “voxels” that could be automatically assembled, and would provide pockets for fish to explore and other marine life to settle in. The cylinders could be connected to form a long, semipermeable wall, which the engineers could erect along a coastline, about half a mile from shore. Based on the team’s initial experiments with lab-scale prototypes, the architected reef could reduce the energy of incoming waves by more than 95 percent.

“This would be like a long wave-breaker,” says Michael Triantafyllou, the Henry L. and Grace Doherty Professor in Ocean Science and Engineering in the Department of Mechanical Engineering. “If waves are 6 meters high coming toward this reef structure, they would be ultimately less than a meter high on the other side. So, this kills the impact of the waves, which could prevent erosion and flooding.”

Details of the architected reef design are reported today in a study appearing in the open-access journal PNAS Nexus. Triantafyllou’s MIT co-authors are Edvard Ronglan SM ’23; graduate students Alfonso Parra Rubio, Jose del Auila Ferrandis, and Erik Strand; research scientists Patricia Maria Stathatou and Carolina Bastidas; and Professor Neil Gershenfeld, director of the Center for Bits and Atoms; along with Alexis Oliveira Da Silva at the Polytechnic Institute of Paris, Dixia Fan of Westlake University, and Jeffrey Gair Jr. of Scinetics, Inc.

Leveraging turbulence

Some regions have already erected artificial reefs to protect their coastlines from encroaching storms. These structures are typically sunken ships, retired oil and gas platforms, and even assembled configurations of concrete, metal, tires, and stones. However, there’s variability in the types of artificial reefs that are currently in place, and no standard for engineering such structures. What’s more, the designs that are deployed tend to have a low wave dissipation per unit volume of material used. That is, it takes a huge amount of material to break enough wave energy to adequately protect coastal communities.

The MIT team instead looked for ways to engineer an artificial reef that would efficiently dissipate wave energy with less material, while also providing a refuge for fish living along any vulnerable coast.

“Remember, natural coral reefs are only found in tropical waters,” says Triantafyllou, who is director of the MIT Sea Grant. “We cannot have these reefs, for instance, in Massachusetts. But architected reefs don’t depend on temperature, so they can be placed in any water, to protect more coastal areas.”

The new effort is the result of a collaboration between researchers in MIT Sea Grant, who developed the reef structure’s hydrodynamic design, and researchers at the Center for Bits and Atoms (CBA), who worked to make the structure modular and easy to fabricate on location. The team’s architected reef design grew out of two seemingly unrelated problems. CBA researchers were developing ultralight cellular structures for the aerospace industry, while Sea Grant researchers were assessing the performance of blowout preventers in offshore oil structures — cylindrical valves that are used to seal off oil and gas wells and prevent them from leaking.

The team’s tests showed that the structure’s cylindrical arrangement generated a high amount of drag. In other words, the structure appeared to be especially efficient in dissipating high-force flows of oil and gas. They wondered: Could the same arrangement dissipate another type of flow, in ocean waves?

The researchers began to play with the general structure in simulations of water flow, tweaking its dimensions and adding certain elements to see whether and how waves changed as they crashed against each simulated design. This iterative process ultimately landed on an optimized geometry: a vertical cylinder flanked by four long slats, each attached to the cylinder in a way that leaves space for water to flow through the resulting structure. They found this setup essentially breaks up any incoming wave energy, causing parts of the wave-induced flow to spiral to the sides rather than crashing ahead.

“We’re leveraging this turbulence and these powerful jets to ultimately dissipate wave energy,” Ferrandis says.

Standing up to storms

Once the researchers identified an optimal wave-dissipating structure, they fabricated a laboratory-scale version of an architected reef made from a series of the cylindrical structures, which they 3D-printed from plastic. Each test cylinder measured about 1 foot wide and 4 feet tall. They assembled a number of cylinders, each spaced about a foot apart, to form a fence-like structure, which they then lowered into a wave tank at MIT. They then generated waves of various heights and measured them before and after passing through the architected reef.

“We saw the waves reduce substantially, as the reef destroyed their energy,” Triantafyllou says.

The team has also looked into making the structures more porous, and friendly to fish. They found that, rather than making each structure from a solid slab of plastic, they could use a more affordable and sustainable type of cement.

“We’ve worked with biologists to test the cement we intend to use, and it’s benign to fish, and ready to go,” he adds.

They identified an ideal pattern of “voxels,” or microstructures, that cement could be molded into, in order to fabricate the reefs while creating pockets in which fish could live. This voxel geometry resembles individual egg cartons, stacked end to end, and appears to not affect the structure’s overall wave-dissipating power.

“These voxels still maintain a big drag while allowing fish to move inside,” Ferrandis says.

The team is currently fabricating cement voxel structures and assembling them into a lab-scale architected reef, which they will test under various wave conditions. They envision that the voxel design could be modular, and scalable to any desired size, and easy to transport and install in various offshore locations. “Now we’re simulating actual sea patterns, and testing how these models will perform when we eventually have to deploy them,” says Anjali Sinha, a graduate student at MIT who recently joined the group.

Going forward, the team hopes to work with beach towns in Massachusetts to test the structures on a pilot scale.

“These test structures would not be small,” Triantafyllou emphasizes. “They would be about a mile long, and about 5 meters tall, and would cost something like 6 million dollars per mile. So it’s not cheap. But it could prevent billions of dollars in storm damage. And with climate change, protecting the coasts will become a big issue.”

This work was funded, in part, by the U.S. Defense Advanced Research Projects Agency.

de MIT News https://ift.tt/f9SWTPY

With a new experimental technique, MIT engineers probe the mechanisms of landslides and earthquakes

Granular materials, those made up of individual pieces, whether grains of sand or coffee beans or pebbles, are the most abundant form of solid matter on Earth. The way these materials move and react to external forces can determine when landslides or earthquakes happen, as well as more mundane events such as how cereal gets clogged coming out of the box. Yet, analyzing the way these flow events take place and what determines their outcomes has been a real challenge, and most research has been confined to two-dimensional experiments that don’t reveal the full picture of how these materials behave.

Now, researchers at MIT have developed a method that allows for detailed 3D experiments that can reveal exactly how forces are transmitted through granular materials, and how the shapes of the grains can dramatically change the outcomes. The new work may lead to better ways of understanding how landslides are triggered, as well as how to control the flow of granular materials in industrial processes. The findings are described in the journal PNAS in a paper by MIT professor of civil and environmental engineering Ruben Juanes and Wei Li SM ’14, PhD ’19, who is now on the faculty at Stony Brook University.

From soil and sand to flour and sugar, granular materials are ubiquitous. “It’s an everyday item, it’s part of our infrastructure,” says Li. “When we do space exploration, our space vehicles land on granular material. And the failure of granular media can be catastrophic, such as landslides.”

“One major finding of this study is that we provide a microscopic explanation of why a pack of angular particles is stronger than a pack of spheres,” Li says.

Juanes adds, “It is always important, at a fundamental level to understand the overall response of the material. And I can see that moving forward, this can provide a new way to make predictions of when a material will fail.”

Scientific understanding of these materials really began a few decades ago, Juanes explains, with the invention of a way to model their behavior using two-dimensional discs representing how forces are transmitted through a collection of particles. While this provided important new insights, it also faced severe limitations.

In previous work, Li developed a way of making three-dimensional particles through a squeeze-molding technique that produces plastic particles that are free of residual stresses and can be made in virtually any irregular shape. Now, in this latest research, he and Juanes have applied this method to reveal the internal stresses in a granular material as loads are applied, in a fully three-dimensional system that much more accurately represents real-world granular materials.

These particles are photoelastic, Juanes explains, which means that when under stress, they modify any light passing through them according to the amount of stress. “So, if you shine polarized light through it and you stress the material, you can see where that stress change is taking place visually, in the form of a different color and different brightness in the material.”

Such materials have been used for a long time, Juanes says, but “one of the key things that had never been accomplished was the ability to image the stresses of these materials when they are immersed in a fluid, where the fluid can flow through the material itself.”

Being able to do so is important, he stresses, because “porous media of interest — biological porous media, industrial porous media, and geological porous media — they often contain fluid in their pore spaces, and that fluid will be hydraulically transported through those pore openings. And the two phenomena are coupled: how the stress is transmitted and what the pore fluid pressure is.”

The problem was, when using a collection of two-dimensional discs for an experiment, the discs would pack in such a way as to block the fluid completely. Only with a three-dimensional mass of grains would there always be pathways for the fluid to flow through, so that the stresses could be monitored while fluid was moving.

Using this method, they were able to show that “when you compress a granular material, that force is transmitted in the form of what we would call chains, or filaments, that this new technique is able to visualize and depict in three dimensions,” Juanes says.

To get that 3D view, they use a combination of the photoelasticity to illuminate the force chains, along with a method called computed tomography, similar to that used in medical CT scans, to reconstruct a full 3D image from a series of 2,400 flat images taken as the object rotates through 360 degrees.

Because the grains are immersed in a fluid that has exactly the same refractive index as the polyurethane grains themselves, the beads are invisible when light shines through their container if they are not under stress. Then, stress is applied, and when polarized light is shone through, that reveals the stresses as light and color, Juanes says. “What’s really remarkable and exciting is that we’re not imaging the porous medium. We’re imaging the forces that are transmitted through the porous medium. This opens up, I think, a new way to interrogate stress changes in granular materials.” He adds that “this has really been a dream of mine for many years,” and he says it was realized thanks to Li’s work on the project.

Using the method, they were able to demonstrate exactly how it is that irregular, angular grains produce a stronger, more stable material than spherical ones. While this was known empirically, the new technique makes it possible to demonstrate exactly why that is, based on the way the forces are distributed, and will make it possible in future work to study a wide variety of grain types to determine exactly what characteristics are most important in producing stable structures, such as the ballast of railroad beds or the riprap on breakwaters.

Because there has been no way to observe the 3D force chains in such materials, Juanes says, “right now it is very difficult to make predictions as to when a landslide will occur precisely, because we don’t know about the architecture of the force chains for different materials.”

It will take time to develop the method to be able to make such predictions, Li says, but that ultimately could be a significant contribution of this new technique. And many other applications of the method are also possible, even in areas as seemingly unrelated as how fish eggs respond as the fish carrying them moves through the water, or in helping to design new kinds of robotic grippers that can easily adapt to picking up objects of any shape.

The work was supported by the U.S. National Science Foundation.

de MIT News https://ift.tt/thINw7x

Engineering household robots to have a little common sense

From wiping up spills to serving up food, robots are being taught to carry out increasingly complicated household tasks. Many such home-bot trainees are learning through imitation; they are programmed to copy the motions that a human physically guides them through.

It turns out that robots are excellent mimics. But unless engineers also program them to adjust to every possible bump and nudge, robots don’t necessarily know how to handle these situations, short of starting their task from the top.

Now MIT engineers are aiming to give robots a bit of common sense when faced with situations that push them off their trained path. They’ve developed a method that connects robot motion data with the “common sense knowledge” of large language models, or LLMs.

Their approach enables a robot to logically parse many given household task into subtasks, and to physically adjust to disruptions within a subtask so that the robot can move on without having to go back and start a task from scratch — and without engineers having to explicitly program fixes for every possible failure along the way.   

“Imitation learning is a mainstream approach enabling household robots. But if a robot is blindly mimicking a human’s motion trajectories, tiny errors can accumulate and eventually derail the rest of the execution,” says Yanwei Wang, a graduate student in MIT’s Department of Electrical Engineering and Computer Science (EECS). “With our method, a robot can self-correct execution errors and improve overall task success.”

Wang and his colleagues detail their new approach in a study they will present at the International Conference on Learning Representations (ICLR) in May. The study’s co-authors include EECS graduate students Tsun-Hsuan Wang and Jiayuan Mao, Michael Hagenow, a postdoc in MIT’s Department of Aeronautics and Astronautics (AeroAstro), and Julie Shah, the H.N. Slater Professor in Aeronautics and Astronautics at MIT.

Language task

The researchers illustrate their new approach with a simple chore: scooping marbles from one bowl and pouring them into another. To accomplish this task, engineers would typically move a robot through the motions of scooping and pouring — all in one fluid trajectory. They might do this multiple times, to give the robot a number of human demonstrations to mimic.

“But the human demonstration is one long, continuous trajectory,” Wang says.

The team realized that, while a human might demonstrate a single task in one go, that task depends on a sequence of subtasks, or trajectories. For instance, the robot has to first reach into a bowl before it can scoop, and it must scoop up marbles before moving to the empty bowl, and so forth. If a robot is pushed or nudged to make a mistake during any of these subtasks, its only recourse is to stop and start from the beginning, unless engineers were to explicitly label each subtask and program or collect new demonstrations for the robot to recover from the said failure, to enable a robot to self-correct in the moment.

“That level of planning is very tedious,” Wang says.

Instead, he and his colleagues found some of this work could be done automatically by LLMs. These deep learning models process immense libraries of text, which they use to establish connections between words, sentences, and paragraphs. Through these connections, an LLM can then generate new sentences based on what it has learned about the kind of word that is likely to follow the last.

For their part, the researchers found that in addition to sentences and paragraphs, an LLM can be prompted to produce a logical list of subtasks that would be involved in a given task. For instance, if queried to list the actions involved in scooping marbles from one bowl into another, an LLM might produce a sequence of verbs such as “reach,” “scoop,” “transport,” and “pour.”

“LLMs have a way to tell you how to do each step of a task, in natural language. A human’s continuous demonstration is the embodiment of those steps, in physical space,” Wang says. “And we wanted to connect the two, so that a robot would automatically know what stage it is in a task, and be able to replan and recover on its own.”

Mapping marbles

For their new approach, the team developed an algorithm to automatically connect an LLM’s natural language label for a particular subtask with a robot’s position in physical space or an image that encodes the robot state. Mapping a robot’s physical coordinates, or an image of the robot state, to a natural language label is known as “grounding.” The team’s new algorithm is designed to learn a grounding “classifier,” meaning that it learns to automatically identify what semantic subtask a robot is in — for example, “reach” versus “scoop” — given its physical coordinates or an image view.

“The grounding classifier facilitates this dialogue between what the robot is doing in the physical space and what the LLM knows about the subtasks, and the constraints you have to pay attention to within each subtask,” Wang explains.

The team demonstrated the approach in experiments with a robotic arm that they trained on a marble-scooping task. Experimenters trained the robot by physically guiding it through the task of first reaching into a bowl, scooping up marbles, transporting them over an empty bowl, and pouring them in. After a few demonstrations, the team then used a pretrained LLM and asked the model to list the steps involved in scooping marbles from one bowl to another. The researchers then used their new algorithm to connect the LLM’s defined subtasks with the robot’s motion trajectory data. The algorithm automatically learned to map the robot’s physical coordinates in the trajectories and the corresponding image view to a given subtask.

The team then let the robot carry out the scooping task on its own, using the newly learned grounding classifiers. As the robot moved through the steps of the task, the experimenters pushed and nudged the bot off its path, and knocked marbles off its spoon at various points. Rather than stop and start from the beginning again, or continue blindly with no marbles on its spoon, the bot was able to self-correct, and completed each subtask before moving on to the next. (For instance, it would make sure that it successfully scooped marbles before transporting them to the empty bowl.)

“With our method, when the robot is making mistakes, we don’t need to ask humans to program or give extra demonstrations of how to recover from failures,” Wang says. “That’s super exciting because there’s a huge effort now toward training household robots with data collected on teleoperation systems. Our algorithm can now convert that training data into robust robot behavior that can do complex tasks, despite external perturbations.”

de MIT News https://ift.tt/XFSa6Bo

Large language models use a surprisingly simple mechanism to retrieve some stored knowledge

Large language models, such as those that power popular artificial intelligence chatbots like ChatGPT, are incredibly complex. Even though these models are being used as tools in many areas, such as customer support, code generation, and language translation, scientists still don’t fully grasp how they work.

In an effort to better understand what is going on under the hood, researchers at MIT and elsewhere studied the mechanisms at work when these enormous machine-learning models retrieve stored knowledge.

They found a surprising result: Large language models (LLMs) often use a very simple linear function to recover and decode stored facts. Moreover, the model uses the same decoding function for similar types of facts. Linear functions, equations with only two variables and no exponents, capture the straightforward, straight-line relationship between two variables.

The researchers showed that, by identifying linear functions for different facts, they can probe the model to see what it knows about new subjects, and where within the model that knowledge is stored.

Using a technique they developed to estimate these simple functions, the researchers found that even when a model answers a prompt incorrectly, it has often stored the correct information. In the future, scientists could use such an approach to find and correct falsehoods inside the model, which could reduce a model’s tendency to sometimes give incorrect or nonsensical answers.

“Even though these models are really complicated, nonlinear functions that are trained on lots of data and are very hard to understand, there are sometimes really simple mechanisms working inside them. This is one instance of that,” says Evan Hernandez, an electrical engineering and computer science (EECS) graduate student and co-lead author of a paper detailing these findings.

Hernandez wrote the paper with co-lead author Arnab Sharma, a computer science graduate student at Northeastern University; his advisor, Jacob Andreas, an associate professor in EECS and a member of the Computer Science and Artificial Intelligence Laboratory (CSAIL); senior author David Bau, an assistant professor of computer science at Northeastern; and others at MIT, Harvard University, and the Israeli Institute of Technology. The research will be presented at the International Conference on Learning Representations.

Finding facts

Most large language models, also called transformer models, are neural networks. Loosely based on the human brain, neural networks contain billions of interconnected nodes, or neurons, that are grouped into many layers, and which encode and process data.

Much of the knowledge stored in a transformer can be represented as relations that connect subjects and objects. For instance, “Miles Davis plays the trumpet” is a relation that connects the subject, Miles Davis, to the object, trumpet.

As a transformer gains more knowledge, it stores additional facts about a certain subject across multiple layers. If a user asks about that subject, the model must decode the most relevant fact to respond to the query.

If someone prompts a transformer by saying “Miles Davis plays the. . .” the model should respond with “trumpet” and not “Illinois” (the state where Miles Davis was born).

“Somewhere in the network’s computation, there has to be a mechanism that goes and looks for the fact that Miles Davis plays the trumpet, and then pulls that information out and helps generate the next word. We wanted to understand what that mechanism was,” Hernandez says.

The researchers set up a series of experiments to probe LLMs, and found that, even though they are extremely complex, the models decode relational information using a simple linear function. Each function is specific to the type of fact being retrieved.

For example, the transformer would use one decoding function any time it wants to output the instrument a person plays and a different function each time it wants to output the state where a person was born.

The researchers developed a method to estimate these simple functions, and then computed functions for 47 different relations, such as “capital city of a country” and “lead singer of a band.”

While there could be an infinite number of possible relations, the researchers chose to study this specific subset because they are representative of the kinds of facts that can be written in this way.

They tested each function by changing the subject to see if it could recover the correct object information. For instance, the function for “capital city of a country” should retrieve Oslo if the subject is Norway and London if the subject is England.

Functions retrieved the correct information more than 60 percent of the time, showing that some information in a transformer is encoded and retrieved in this way.

“But not everything is linearly encoded. For some facts, even though the model knows them and will predict text that is consistent with these facts, we can’t find linear functions for them. This suggests that the model is doing something more intricate to store that information,” he says.

Visualizing a model’s knowledge

They also used the functions to determine what a model believes is true about different subjects.

In one experiment, they started with the prompt “Bill Bradley was a” and used the decoding functions for “plays sports” and “attended university” to see if the model knows that Sen. Bradley was a basketball player who attended Princeton.

“We can show that, even though the model may choose to focus on different information when it produces text, it does encode all that information,” Hernandez says.

They used this probing technique to produce what they call an “attribute lens,” a grid that visualizes where specific information about a particular relation is stored within the transformer’s many layers.

Attribute lenses can be generated automatically, providing a streamlined method to help researchers understand more about a model. This visualization tool could enable scientists and engineers to correct stored knowledge and help prevent an AI chatbot from giving false information.

In the future, Hernandez and his collaborators want to better understand what happens in cases where facts are not stored linearly. They would also like to run experiments with larger models, as well as study the precision of linear decoding functions.

“This is an exciting work that reveals a missing piece in our understanding of how large language models recall factual knowledge during inference. Previous work showed that LLMs build information-rich representations of given subjects, from which specific attributes are being extracted during inference. This work shows that the complex nonlinear computation of LLMs for attribute extraction can be well-approximated with a simple linear function,” says Mor Geva Pipek, an assistant professor in the School of Computer Science at Tel Aviv University, who was not involved with this work.

This research was supported, in part, by Open Philanthropy, the Israeli Science Foundation, and an Azrieli Foundation Early Career Faculty Fellowship.

de MIT News https://ift.tt/2XvNF1e