Waiting in an airport for a connecting flight is often tedious. A new study by MIT researchers shows it’s bad for business, too.

Looking at air travel and multinational firm formation over a 30-year period, the researchers measured how much a strong network of airline connections matters for economic growth. They found that multinational firms are more likely to locate their subsidiaries in cities they can reach with direct flights, and that this trend is particularly pronounced in knowledge industries. The degree to which a city is embedded within a larger network of high-use flights matters notably for business expansion too.

The team examined 142 countries over the period from 1993 through 2023 and concluded that pairs of cities reachable only by flights with one stopover had 20 percent fewer multinational firm subsidiaries than cities with direct flights. If two changes of planes were needed to connect cities, they had 34 percent fewer subsidiaries. That equates to 1.8 percent and 3.0 percent fewer new firms per year, respectively.

“What we found is how much it matters for a city to be embedded within the global air transportation network,” says Ambra Amico, an MIT researcher and co-author of a new paper detailing the study’s results. “And we also highlight the importance of this for knowledge-intensive business sectors.”

Siqi Zheng, an MIT professor and co-author of the paper, adds: “We found a very strong empirical result about the relationship of parent and subsidiary firms, and how much connectivity matters. The important role that connectivity plays to facilitate face-to-face interactions, build trust, and reduce information asymmetry between such firms is crucial.”

The paper, “Air Connectivity Boosts Urban Attractiveness for Global Firms,” is published today in Nature Cities.

The co-authors are Amico, a postdoc at the MIT-Singapore Alliance for Research and Technology (SMART); Fabio Duarte, associate director of MIT’s Senseable City Lab; Wen-Chi Liao, a visiting associate professor at the MIT Center for Real Estate (CRE) and an associate professor at NUS Business School at the National University of Singapore; and Zheng, the STL Champion Professor of Urban and Real Estate Sustainability at CRE and MIT’s Department of Urban Studies and Planning.

The study analyzes 7.5 million firms in 800 cities with airports, comprising a total of over 400,000 international flight routes. The research focused only on multinational firms, and thus international flights, excluding domestic flights in large countries.

To conduct the analysis and build their new database, the researchers used flight data from the International Civil Aviation Organization as well as firm data from the Orbis database, run by Moody’s, which has company data for over 469 million firms globally. That includes ownership data, allowing the researchers to track relationships between companies. The study included firms located within 37 miles (60 kilometers) of an airport, and accounted for additional factors influencing new-firm location, including city size.

By analyzing industry types, the researchers observed that air connectivity matters relatively more in knowledge industries, such as finance, where face-to-face activity seems to matter more. Alternately, a knowledge-industry firm with auditors periodically showing up to conduct work can lower costs by being more reachable.

“We were fascinated by the heterogenity across industries,” Liao says. “The results are intuitive, but it surprised us that the pattern is so consistent. If the nature of the industy requires face-to-face interaction, air connectivity matters more.” By contrast, for manufacturing, he notes, road infrastructure and ocean shipping will matter relatively more.

To be sure, there are multiple ways to define how connected a city is within the global air transportation network, and the study examines how specific measures relate to firm growth. One measure is what the paper calls “degree centrality,” or how many other places a city is connected to by direct flights. Over a 10-year period, a 10 percent increase in a city’s degree centrality leads to a 4.3 percent increase in the number of subsidiaries located there.

However, another kind of connectedness is even more strongly associated with subsidiary growth. It’s not just how many cities one place is linked to, but in turn, how many direct connections those linked cities themselves have. This turns out to be the strongest predictor of subsidiary growth.

“What matters is not just how many neighbor [directly linked] cities you have,” Duarte says. “It’s important to choose strategically which ones you’re connected to, as well. If you tell me who you are connected to, I tell you how successful your city will be.”

Intriguingly, the relationship between direct flights and multinational firm growth patterns has held up throughout the 30-year study period, despite the rise of teleconferencing, the Covid-19 pandemic, shifts in global growth, and other factors.

“There is consistency across a 30-year period, which is not something to underestimate,” Amico says. “We needed face-to-face interaction 30 years ago, 20 years ago, and 10 years ago, and we need it now, despite all the big changes we have seen.”

Indeed, Zheng adds, “Ironically, I think even with trade and geopolitical frictions, it’s more and more important to have face-to-face interactions to build trust for global trade and business. You still need to reach an actual place and see your business partners, so air connectivity really influences how global business copes with global uncertainties.”

The research was supported by the National Research Foundation of Singapore within the Office of the Prime Minister of Singapore, under its Campus for Research Excellence and Technological Enterprise program, and the MIT Asia Real Estate Initiative. 

Can an organization ever be truly meritocratic? That’s a question Emilio J. Castilla, the NTU Professor of Management at the MIT Sloan School of Management, explores in his new book, “The Meritocracy Paradox: Where Talent Management Strategies Go Wrong and How to Fix Them” (Columbia University Press, 2025). Castilla, who is co-director of MIT’s Institute for Work and Employment Research (IWER), researches how talent is managed inside organizations and why — even with the best intentions — workplace practices often fail to deliver fairness and effectiveness.

Castilla’s book brings together decades of research to explain why organizations struggle to achieve meritocracy in practice — and what leaders can do to build fairer, more effective, and higher-performing workplaces. In the following Q&A, he unpacks how bias can unintentionally seep into hiring, evaluation, promotion, and reward systems, and offers concrete strategies to counteract these dynamics and design processes that recognize and support merit.

Q: One central argument of your book is that true meritocracy is not easy for organizations to achieve in practice. Why is that? 

A. A large body of research has found that bias and unfairness can creep into the workplace, affecting talent management processes such as who gets interviewed for jobs, who gets hired, what kind of performance evaluations employees receive, and how employees are rewarded. So it’s not easy for an organization to be truly meritocratic.

In fact, research I conducted with Stephen Benard found that, ironically, emphasizing that an organization is a meritocracy may lead decision-makers to behave in more biased ways. Specifically, in our study, we found that when participants were told they were making decisions for an organization that emphasized meritocracy, they were more likely to recommend higher bonuses for male employees than for their equally-performing female peers, compared to when meritocracy wasn’t emphasized. We called this phenomenon the “paradox of meritocracy,” and it may stem from managers paying less attention to monitoring their own biases when they are assured the organization is fair.

A study I conducted with Aruna Ranganathan PhD ’14 further showed that managers’ understanding of what constitutes “merit” varies widely — even within the same organization. There is no universally agreed-upon definition, and our research found that managers often apply the concept of merit in ways that reflect their own experiences as employees. This variability can lead to inconsistent, and sometimes inequitable, outcomes.

Q. What are some of the things organizations can do to make their talent management practices more meritocratic?

A. The encouraging news is that making your organization’s talent management processes fairer and more meritocratic doesn’t have to be complex or expensive. It does, however, require buy-in from top management. The key factors, my research in organizations has shown, are organizational transparency and accountability.

To improve organizational transparency, you need to be very explicit and open about the criteria and procedures you use in talent management processes such as hiring, evaluation, promotion, and reward decisions. That’s because research has shown that having clear and specific merit-based criteria and well-defined processes can help reduce biases.

On the accountability side, you need to have at least one person responsible for monitoring the organization’s talent management processes and outcomes to ensure fairness and effectiveness. In practice, companies often give this responsibility to a group from different parts of the organization. Research has shown that knowing that your decisions will be reviewed by others causes managers to think carefully about their decisions — something that can reduce the impact of unconscious biases in the workplace.

Q. How realistic is it to think that organizations can ever be true meritocracies — and why do you nonetheless believe meritocracy is worth striving for?

A. It’s true that organizations are unlikely to ever be perfectly meritocratic. Still, striving for meritocracy and fairness in your talent management strategies is beneficial, and you should be aware of the pitfalls. Employers that hire, reward, and advance the most talented and hard-working employees, regardless of their demographic background, are likely to benefit in the long run. That’s the promise and enduring appeal of meritocracy.

Many in the United States may not realize that one of the world’s earliest formal meritocracies emerged in China during the Han and Qin dynasties more than 2,000 years ago. As early as 200 B.C.E., the Chinese empire began developing a system of civil service exams in order to identify and appoint competent and talented officials to help administer government operations throughout the empire.

Those Chinese emperors were on to something. Once an organization reaches a certain size, leaders won’t achieve the most effective performance if they make talent management decisions based on non-meritocratic factors such as nepotism, aristocracy/social class, corruption, or friendship. When it comes to choosing a guiding principle for people management decisions within an organization, meritocracy beats a lot of the alternatives.

Massachusetts is uniquely positioned to become a leader in climate tech, said Emily Reichert MBA ’12, the CEO of the Massachusetts Clean Energy Center (MassCEC) and former CEO of Greentown Labs, to members of the MIT community at a seminar in November. 

Reichert outlined the interconnectedness of economic development and clean energy innovation in MassCEC’s efforts to advance the energy transition and address climate change, as part of the MITEI Presents: Advancing the Energy Transition speaker series. An MIT Sloan School of Management alumna, Reichert stepped aside as the agency’s CEO in late November and the MITEI speaker series was her final presentation in that role.

“There’s not another [agency] in the country exactly like us focused on innovation and economic development for clean energy and climate tech,” stated Reichert. Created in 2008, MassCEC is the state’s economic development agency dedicated to the growth of the clean energy and climate tech sector. Reichert stressed that economic development is just as much about businesses as it is about the jobs they create.

The organization’s economic development plan is built on its knowledge of the commonwealth’s infrastructure, talent capabilities, academic resources, startup culture, and regional strengths. Reichert explained that there are four areas at the core of MassCEC’s work.

First, bringing emerging climate-tech ideas out of the laboratory and into the world. To do this, MassCEC provides grants, internships, and has a small investment fund that is co-invested with different investors in the area. “We are increasingly focusing on the longer-term growth trajectory of these young companies,” said Reichert, adding that the hope is for these startups to stay, grow, and create jobs in Massachusetts.

Second, MassCEC aims to accelerate decarbonization by taking commercialized technologies and helping to get them into as many homes and businesses as possible. This can often require specialized knowledge of Massachusetts’ infrastructure, given that the state has relatively older buildings and unique structures, such as triple-deckers. One example is finding a way to make charging technology available to electric vehicle owners when they don’t have a single-family home with a garage.

MassCEC is also focused on enabling the large-scale deployment of offshore wind. “It’ll be 400,000 homes that are powered by the clean energy that’s being generated by offshore wind right off the coast of Martha’s Vineyard. MassCEC’s role is to support the port infrastructure from which we marshal those offshore wind projects,” stated Reichert. “We also support innovation that is needed to do all the things that support the offshore wind industry, in general.”

Finally, Reichert reiterated that MassCEC’s overarching goal is to support clean energy workforce development through job creation, as well as professional development opportunities such as providing internships, training for high school and community college students, and supporting students returning to school for a second career in clean energy.

Reichert emphasized that Massachusetts is particularly well-equipped to house this level of climate-tech innovation since the state is already a leader in the life sciences. The Healey-Driscoll administration charged MassCEC with spearheading the state’s Climatetech Economic Development Strategy and Implementation Plan, a 10-year strategy to position Massachusetts as a global climate tech leader and drive a more equitable and resilient climate future.

To complement this plan and further position the state as an epicenter for energy innovation, the Healey-Driscoll administration also passed the Mass Leads Act, which established the Climatetech Tax Incentive Program, an annual tax incentive to be administered by MassCEC. “This is the money piece,” said Reichert. “How we do it. How we implement it.”

To unlock Massachusetts’ full potential, MassCEC uses a regional approach to take advantage of the strengths held in each area of the state. “We have a fantastic ecosystem. We have more startups per capita than any other state,” said Reichert. The quantity of startups is in large part due to the strengths of the Greater Boston region, with its strong venture capital community and good research institutions, said Reichert, who also highlighted MIT as a key factor. MIT spinout companies like Sublime Systems, Commonwealth Fusion Systems, Boston Metal, and The Engine are all part of MassCEC’s ecosystem.

For the agency, retaining talent in Massachusetts is just as important as supporting its development. “How can we help companies to do their processes, find their facilities, build their facilities, do their demonstrations, do their testing, and find the talent?” asked Reichert. “How can we reduce the time and money barriers to all of that, and therefore make it as easy as possible and as inexpensive as possible for the company to stay here and grow here?”

Reichert expressed her confidence in climate-tech innovation’s ability to endure the changing energy landscape. “The rest of the world is going in this direction. We can decide not to compete as a country, or we can decide that we want to compete and that we want to be part of the future,” said Reichert. “Innovation isn’t going anywhere. I think when you have places like MIT, who are very focused on climate innovation and the energy transition, that activity helps move the ball forward.”

This speaker series highlights energy experts and leaders at the forefront of the scientific, technological, and policy solutions needed to transform our energy systems. Visit MITEI’s Events page for more information on this and additional events.

Detecting cancer in the earliest stages could dramatically reduce cancer deaths because cancers are usually easier to treat when caught early. To help achieve that goal, MIT and Microsoft researchers are using artificial intelligence to design molecular sensors for early detection.

The researchers developed an AI model to design peptides (short proteins) that are targeted by enzymes called proteases, which are overactive in cancer cells. Nanoparticles coated with these peptides can act as sensors that give off a signal if cancer-linked proteases are present anywhere in the body.

Depending on which proteases are detected, doctors would be able to diagnose the particular type of cancer that is present. These signals could be detected using a simple urine test that could even be done at home.

“We’re focused on ultra-sensitive detection in diseases like the early stages of cancer, when the tumor burden is small, or early on in recurrence after surgery,” says Sangeeta Bhatia, the John and Dorothy Wilson Professor of Health Sciences and Technology and of Electrical Engineering and Computer Science at MIT, and a member of MIT’s Koch Institute for Integrative Cancer Research and the Institute for Medical Engineering and Science (IMES).

Bhatia and Ava Amini ’16, a principal researcher at Microsoft Research and a former graduate student in Bhatia’s lab, are the senior authors of the study, which appears today in Nature Communications. Carmen Martin-Alonso PhD ’23, a founding scientist at Amplifyer Bio, and Sarah Alamdari, a senior applied scientist at Microsoft Research, are the paper’s lead authors.

Amplifying cancer signals

More than a decade ago, Bhatia’s lab came up with the idea of using protease activity as a marker of early cancer. The human genome encodes about 600 proteases, which are enzymes that can cut through other proteins, including structural proteins such as collagen. They are often overactive in cancer cells, as they help the cells escape their original locations by cutting through proteins of the extracellular matrix, which normally holds cells in place.

The researchers’ idea was to coat nanoparticles with peptides that can be cleaved by a specific protease. These particles could then be ingested or inhaled. As they traveled through the body, if they encountered any cancer-linked proteases, the peptides on the particles would be cleaved.

Those peptides would be secreted in the urine, where they could be detected using a paper strip similar to a pregnancy test strip. Measuring those signals would reveal the overactivity of proteases deep within the body.

“We have been advancing the idea that if you can make a sensor out of these proteases and multiplex them, then you could find signatures of where these proteases were active in diseases. And since the peptide cleavage is an enzymatic process, it can really amplify a signal,” Bhatia says.

The researchers have used this approach to demonstrate diagnostic sensors for lung, ovarian, and colon cancers.

However, in those studies, the researchers used a trial-and-error process to identify peptides that would be cleaved by certain proteases. In most cases, the peptides they identified could be cleaved by more than one protease, which meant that the signals that were read could not be attributed to a specific enzyme.

Nonetheless, using “multiplexed” arrays of many different peptides yielded distinctive sensor signatures that were diagnostic in animal models of many different types of cancer, even if the precise identity of the proteases responsible for the cleavage remained unknown.

In their new study, the researchers moved beyond the traditional trial-and-error process by developing a novel AI system, named CleaveNet, to design peptide sequences that could be cleaved efficiently and specifically by target proteases of interest.

Users can prompt CleaveNet with design criteria, and CleaveNet will generate candidate peptides likely to fit those criteria. In this way, CleaveNet enables users to tune the efficiency and specificity of peptides generated by the model, opening a path to improving the sensors’ diagnostic power.

“If we know that a particular protease is really key to a certain cancer, and we can optimize the sensor to be highly sensitive and specific to that protease, then that gives us a great diagnostic signal,” Amini says. “We can leverage the power of computation to try to specifically optimize for these efficiency and selectivity metrics.”

For a peptide that contains 10 amino acids, there are about 10 trillion possible combinations. Using AI to search that immense space allows for prediction, testing, and identification of useful sequences much faster than humans would be able to find them, while also considerably reducing experimental costs.

Predicting enzyme activity

To create CleaveNet, the researchers developed a protein language model to predict the amino acid sequences of peptides, analogous to how large language models can predict sequences of text. For the training data, they used publicly available data on about 20,000 peptides and their interactions with different proteases from a family known as matrix metalloproteinases (MMPs).

Using these data, the researchers trained one model to generate peptide sequences that are predicted to be cleaved by proteases. These sequences could then be fed into another model that predicted how efficiently each peptide would be cleaved by any protease of interest.

To demonstrate this approach, the researchers focused on a protease called MMP13, which cancer cells use to cut through collagen and help them metastasize from their original locations. Prompting CleaveNet with MMP13 as a target allowed the models to design peptides that could be cut by MMP13 with considerable selectivity and efficiency. This cleavage profile is particularly useful for diagnostic and therapeutic applications.

“When we set the model up to generate sequences that would be efficient and selective for MMP13, it actually came up with peptides that had never been observed in training, and yet these novel sequences did turn out to be both efficient and selective,” Martin-Alonso says. “That was very exciting to see.”

This kind of selectivity could help to reduce the number of different peptides needed to diagnose a given type of cancer, to identify novel biomarkers, and to provide insight into specific biological pathways for study and therapeutic testing, the researchers say.

Bhatia’s lab is currently part of an ARPA-H funded project to create reporters for an at-home diagnostic kit that could potentially detect and distinguish between 30 different types of cancer, in early stages of disease, based on measurements of protease activity. These sensors could include detection of not only MMP-mediated cleavage, but other enzymes such as serine proteases and cysteine proteases.

Peptides designed using CleaveNet could also be incorporated into cancer therapeutics such as antibody treatments. Using a specific peptide to attach a therapeutic such as a cytokine or small molecule drug to a targeting antibody could enable the medicine to be released only when the peptides are exposed to proteases in the tumor environment, improving efficacy and reducing side effects.

Beyond direct applications in diagnostics and therapeutics, combining efforts from the ARPA-H work with this modeling framework could enable the creation of a comprehensive “protease activity atlas” that spans multiple protease classes and cancers. Such a resource could further accelerate research in early cancer detection, protease biology, and AI models for peptide design.

The research was funded by La Caixa Foundation, the Ludwig Center at MIT, and the Marble Center for Cancer Nanomedicine.

In elementary school, Sean Luk loved donning an oversized lab coat and helping her mom pipette chemicals at Johns Hopkins University. A few years later, she started a science blog and became fascinated by immunoengineering, which is now her concentration as a biological engineering major at MIT.

Her grandparents’ battles with cancer made Luk, now a senior, realize how urgently patients need advancements in immunotherapy, which leverages a patient’s immune system to fight tumors or pathogens.

“The idea of creating something that is actually able to improve human health is what really drives me now. You want to fight that sense of helplessness when you see a loved one suffering through this disease, and it just further motivates me to be excellent at what I do,” Luk says.

A varsity athlete and entrepreneur as well as a researcher, Luk thrives when bringing people together for a common cause.

Working with immunotherapies

Luk was introduced to immunotherapies in high school after she listened to a seminar about using components of the immune system, such as antibodies and cytokines, to improve graft tolerance.

“The complexity of the immune system really fascinated me, and it is incredible that we can build antibodies in a very logical way to address disease,” Luk says.

She worked in several Johns Hopkins labs as a high school student in Maryland, and a professor there connected her to MIT Professor Dane Wittrup. Luk has worked in the Wittrup lab throughout her time at MIT. One of her main projects involves developing ultra-stable cyclic peptide drugs to help treat autoimmune diseases, which could potentially be taken orally instead of injected.

Luk has been a co-author on two published articles and has become increasingly interested in the intersection between computational and experimental protein design. Currently, she is working on engineering an interferon gamma construct that preferentially targets myeloid cells in the tumor microenvironment.

“We're trying to target and reprogram the immunosuppressive myeloid cells surrounding the cancer cells, so that they can license T cells to attack cancer cells and kickstart the cancer immunity cycle,” she explains.

Communication for all

Through her work in high school with Best Buddies, an organization that aims to promote one-on-one friendships between students with and without intellectual and developmental disabilities, Luk became passionate about empowering people with special needs. At MIT, she started a project focusing on children with Down syndrome, with support from the Sandbox Innovation Fund.

“Through talking to a lot of parents and caretakers, the biggest issue that people with Down syndrome face is communication. And when you think about it, communication is crucial to everything that we do,” Luk says, “We want to communicate our thoughts. We want to be able to interact with our peers. And if people are unable to do that, it’s isolating, it’s frustrating.”

Her solution was to co-found EasyComm, an online game platform that helps children with Down syndrome work on verbal communication.

“We thought it would be a great way to improve their verbal communication skills while having fun and incentivize that kind of learning through gamification,” Luk says. She and her co-founder recently filed a provisional patent and plan to make the platform available to a wider audience.

A global perspective

Luk grew up in Hong Kong before moving to Maryland in the fifth grade. She’s always been athletic; in Hong Kong, she was a competitive jump roper. At just 9 years old, she won bronze in the Asian Jump Rope Championships among children 14 years old and younger. At 7 years old, she started playing soccer on her brother’s team, despite being the only girl. She says the sport was considered “manly” in Hong Kong, and girls were discouraged from joining, but her coaches and family were supportive.

Moving to the U.S. meant that her time in competitive jump roping was cut short, and Luk focused more on soccer. Her team in the U.S. felt far more intense than boys soccer in Hong Kong, but the Luk family was in it together, Luk says. She credits her success to the combination of her hard-working nature she learned from Hong Kong, and the innovation and experiences she was exposed to in the U.S.

“We had a really close bond within the family,” Luk says, “Figuring out taxes for my dad and our family, like driving and houses and all that stuff, it was totally new. But I think we really took it in stride, just adjusting as we went.”

Luk continued soccer throughout high school and eventually committed to play on the MIT team. She likes that the team allows players to prioritize academics while still being competitive. Last season, she was elected captain.

“It’s really a pleasure to be captain, and it’s challenging, but it’s also very rewarding when you see the team be cohesive. When you see the team out there winning games through grit,” Luk says.

During her first year at MIT, Luk got back in touch with her old soccer coach from Hong Kong, who then worked on the national team. After sending over some tape, she was offered a spot on the U-20 national team, and played in the U20 Asian Football Championship Qualifiers.

“It was so, so cool to be able to represent Hong Kong because I played soccer all my life but it just carries a different weight to it when you’re wearing your country’s jersey,” Luk says.

Besides her cross-cultural background, Luk is also proud of her international experiences playing soccer, staying with host families and doing lab work in Copenhagen, Denmark; Stuttgart, Germany; and Ancona, Italy. She speaks English, Cantonese, and Mandarin fluently.

“Aside from the textbook academic knowledge, I feel like a global perspective is so important when you’re trying to collaborate with other people from different walks of life,” Luk says, “When you’re just thinking about science or the impact that you can have in general, it’s important to realize you don’t have all the answers and to learn from the world outside your little bubble.”

What is patient privacy for? The Hippocratic Oath, thought to be one of the earliest and most widely known medical ethics texts in the world, reads: “Whatever I see or hear in the lives of my patients, whether in connection with my professional practice or not, which ought not to be spoken of outside, I will keep secret, as considering all such things to be private.” 

As privacy becomes increasingly scarce in the age of data-hungry algorithms and cyberattacks, medicine is one of the few remaining domains where confidentiality remains central to practice, enabling patients to trust their physicians with sensitive information.

But a paper co-authored by MIT researchers investigates how artificial intelligence models trained on de-identified electronic health records (EHRs) can memorize patient-specific information. The work, which was recently presented at the 2025 Conference on Neural Information Processing Systems (NeurIPS), recommends a rigorous testing setup to ensure targeted prompts cannot reveal information, emphasizing that leakage must be evaluated in a health care context to determine whether it meaningfully compromises patient privacy.

Foundation models trained on EHRs should normally generalize knowledge to make better predictions, drawing upon many patient records. But in “memorization,” the model draws upon a singular patient record to deliver its output, potentially violating patient privacy. Notably, foundation models are already known to be prone to data leakage.

“Knowledge in these high-capacity models can be a resource for many communities, but adversarial attackers can prompt a model to extract information on training data,” says Sana Tonekaboni, a postdoc at the Eric and Wendy Schmidt Center at the Broad Institute of MIT and Harvard and first author of the paper. Given the risk that foundation models could also memorize private data, she notes, “this work is a step towards ensuring there are practical evaluation steps our community can take before releasing models.”

To conduct research on the potential risk EHR foundation models could pose in medicine, Tonekaboni approached MIT Associate Professor Marzyeh Ghassemi, who is a principal investigator at the Abdul Latif Jameel Clinic for Machine Learning in Health (Jameel Clinic) and a member of the Computer Science and Artificial Intelligence Lab. Ghassemi, a faculty member in the MIT Department of Electrical Engineering and Computer Science and Institute for Medical Engineering and Science, runs the Healthy ML group, which focuses on robust machine learning in health.

Just how much information does a bad actor need to expose sensitive data, and what are the risks associated with the leaked information? To assess this, the research team developed a series of tests that they hope will lay the groundwork for future privacy evaluations. These tests are designed to measure various types of uncertainty, and assess their practical risk to patients by measuring various tiers of attack possibility.  

“We really tried to emphasize practicality here; if an attacker has to know the date and value of a dozen laboratory tests from your record in order to extract information, there is very little risk of harm. If I already have access to that level of protected source data, why would I need to attack a large foundation model for more?” says Ghassemi. 

With the inevitable digitization of medical records, data breaches have become more commonplace. In the past 24 months, the U.S. Department of Health and Human Services has recorded 747 data breaches of health information affecting more than 500 individuals, with the majority categorized as hacking/IT incidents.

Patients with unique conditions are especially vulnerable, given how easy it is to pick them out. “Even with de-identified data, it depends on what sort of information you leak about the individual,” Tonekaboni says. “Once you identify them, you know a lot more.”

In their structured tests, the researchers found that the more information the attacker has about a particular patient, the more likely the model is to leak information. They demonstrated how to distinguish model generalization cases from patient-level memorization, to properly assess privacy risk. 

The paper also emphasized that some leaks are more harmful than others. For instance, a model revealing a patient’s age or demographics could be characterized as a more benign leakage than the model revealing more sensitive information, like an HIV diagnosis or alcohol abuse. 

The researchers note that patients with unique conditions are especially vulnerable given how easy it is to pick them out, which may require higher levels of protection. “Even with de-identified data, it really depends on what sort of information you leak about the individual,” Tonekaboni says. The researchers plan to expand the work to become more interdisciplinary, adding clinicians and privacy experts as well as legal experts. 

“There’s a reason our health data is private,” Tonekaboni says. “There’s no reason for others to know about it.”

This work supported by the Eric and Wendy Schmidt Center at the Broad Institute of MIT and Harvard, Wallenberg AI, the Knut and Alice Wallenberg Foundation, the U.S. National Science Foundation (NSF), a Gordon and Betty Moore Foundation award, a Google Research Scholar award, and the AI2050 Program at Schmidt Sciences. Resources used in preparing this research were provided, in part, by the Province of Ontario, the Government of Canada through CIFAR, and companies sponsoring the Vector Institute.

A long stretch of humid heat followed by intense thunderstorms is a weather pattern historically seen mostly in and around the tropics. But climate change is making humid heat waves and extreme storms more common in traditionally temperate midlatitude regions such as the midwestern U.S., which has seen episodes of unusually high heat and humidity in recent summers.

Now, MIT scientists have identified a key condition in the atmosphere that determines how hot and humid a midlatitude region can get, and how intense related storms can become. The results may help climate scientists gauge a region’s risk for humid heat waves and extreme storms as the world continues to warm.

In a study appearing this week in the journal Science Advances, the MIT team reports that a region’s maximum humid heat and storm intensity are limited by the strength of an “atmospheric inversion”— a weather condition in which a layer of warm air settles over cooler air.

Inversions are known to act as an atmospheric blanket that traps pollutants at ground level. Now, the MIT researchers have found atmospheric inversions also trap and build up heat and moisture at the surface, particularly in midlatitude regions. The more persistent an inversion, the more heat and humidity a region can accumulate at the surface, which can lead to more oppressive, longer-lasting humid heat waves.

And, when an inversion eventually weakens, the accumulated heat energy is released as convection, which can whip up the hot and humid air into intense thunderstorms and heavy rainfall.

The team says this effect is especially relevant for midlatitude regions, where atmospheric inversions are common. In the U.S., regions to the east of the Rocky Mountains often experience inversions of this kind, with relatively warm air aloft sitting over cooler air near the surface.

As climate change further warms the atmosphere in general, the team suspects that inversions may become more persistent and harder to break. This could mean more frequent humid heat waves and more intense storms for places that are not accustomed to such extreme weather.

“Our analysis shows that the eastern and midwestern regions of U.S. and the eastern Asian regions may be new hotspots for humid heat in the future climate,” says study author Funing Li, a postdoc in MIT’s Department of Earth, Atmospheric and Planetary Sciences (EAPS).

“As the climate warms, theoretically the atmosphere will be able to hold more moisture,” adds co-author and EAPS Assistant Professor Talia Tamarin-Brodsky. “Which is why new regions in the midlatitudes could experience moist heat waves that will cause stress that they weren’t used to before.”

Air energetics

The atmosphere’s layers generally get colder with altitude. In these typical conditions, when a heat wave comes through a region, it warms the air at ground level. Since warm air is lighter than cold air, it will eventually rise, like a hot air balloon, prompting colder air to sink. This rise and fall of air sets off convection, like bubbles in boiling water. When warm air hits colder altitudes, it condenses into droplets that rain out, typically as a thunderstorm, that can often relieve a heat wave.

For their new study, Li and Tamarin-Brodsky wondered: What would it take to get air at the surface to convect and ultimately end a heat wave? Put another way: What sets the limit to how hot a region can get before air begins to convect to eventually rain?

The team treated the question as a problem of energy. Heat is energy that can be thought of in two forms: the energy that comes from dry heat (i.e., temperature), and the energy that comes from latent, or moist, heat. The scientists reasoned that, for a given portion or “parcel” of air, there is some amount of moisture that, when condensed, contributes to that air parcel’s total energy. Depending on how much energy an air parcel has, it could start to convect, rise up, and eventually rain out.

“Imagine putting a balloon around a parcel of air and asking, will it stay in the same place, will it go up, or will it sink?” Tamarin-Brodsky says. “It’s not just about warm air that’s lifting. You also have to think about the moisture that’s there. So we consider the energetics of an air parcel while taking into account the moisture in that air. Then we can find the maximum ‘moist energy’ that can accumulate near the surface before the air becomes unstable and convects.”

Heat barrier

As they worked through their analysis, the researchers found that the maximum amount of moist energy, or the highest level of heat and humidity that the air can hold, is set by the presence and strength of an atmospheric inversion. In cases where atmospheric layers are inverted (when a layer of warm or light air settles over colder or heavier, ground-level air), the air has to accumulate more heat and moisture in order for an air parcel to build up enough energy to lift up and break through the inversion layer. The more persistent the inversion is, the hotter and more humid air must get before it can rise up and convect.

Their analysis suggests that an atmospheric inversion can increase a region’s capacity to hold heat and humidity. How high this heat and humidity can get depends on how stable the inversion is. If a blanket of warm air parks over a region without moving, it allows more humid heat to build up, versus if the blanket is quickly removed. When the air eventually convects, the accumulated heat and moisture will generate stronger, more intense storms.

“This increasing inversion has two effects: more severe humid heat waves, and less frequent but more extreme convective storms,” Tamarin-Brodsky says.

Inversions in the atmosphere form in various ways. At night, the surface that warmed during the day cools by radiating heat to space, making the air in contact with it cooler and denser than the air above. This creates a shallow layer in which temperature increases with height, called a nocturnal inversion. Inversions can also form when a shallow layer of cool marine air moves inland from the ocean and slides beneath warmer air over the land, leaving cool air near the surface and warmer air above. In some cases, persistent inversions can form when air heated over sun-warmed mountains is carried over colder low-lying regions, so that a warm layer aloft caps cooler air near the ground.

“The Great Plains and the Midwest have had many inversions historically due to the Rocky Mountains,” Li says. “The mountains act as an efficient elevated heat source, and westerly winds carry this relatively warm air downstream into the central and midwestern U.S., where it can help create a persistent temperature inversion that caps colder air near the surface.”

“In a future climate for the Midwest, they may experience both more severe thunderstorms and more extreme humid heat waves,” Tamarin-Brodsky says. “Our theory gives an understanding of the limit for humid heat and severe convection for these communities that will be future heat wave and thunderstorm hotspots.”

This research is part of the MIT Climate Grand Challenge on Weather and Climate Extremes. Support was provided by Schmidt Sciences.