Transistors, the building blocks of modern electronics, are typically made of silicon. Because it’s a semiconductor, this material can control the flow of electricity in a circuit. But silicon has fundamental physical limits that restrict how compact and energy-efficient a transistor can be.

MIT researchers have now replaced silicon with a magnetic semiconductor, creating a magnetic transistor that could enable smaller, faster, and more energy-efficient circuits. The material’s magnetism strongly influences its electronic behavior, leading to more efficient control of the flow of electricity. 

The team used a novel magnetic material and an optimization process that reduces the material’s defects, which boosts the transistor’s performance.

The material’s unique magnetic properties also allow for transistors with built-in memory, which would simplify circuit design and unlock new applications for high-performance electronics.

“People have known about magnets for thousands of years, but there are very limited ways to incorporate magnetism into electronics. We have shown a new way to efficiently utilize magnetism that opens up a lot of possibilities for future applications and research,” says Chung-Tao Chou, an MIT graduate student in the departments of Electrical Engineering and Computer Science (EECS) and Physics, and co-lead author of a paper on this advance.

Chou is joined on the paper by co-lead author Eugene Park, a graduate student in the Department of Materials Science and Engineering (DMSE); Julian Klein, a DMSE research scientist; Josep Ingla-Aynes, a postdoc in the MIT Plasma Science and Fusion Center; Jagadeesh S. Moodera, a senior research scientist in the Department of Physics; and senior authors Frances Ross, TDK Professor in DMSE; and Luqiao Liu, an associate professor in EECS, and a member of the Research Laboratory of Electronics; as well as others at the University of Chemistry and Technology in Prague. The paper appears today in Physical Review Letters.

Overcoming the limits

In an electronic device, silicon semiconductor transistors act like tiny light switches that turn a circuit on and off, or amplify weak signals in a communication system. They do this using a small input voltage.

But a fundamental physical limit of silicon semiconductors prevents a transistor from operating below a certain voltage, which hinders its energy efficiency.

To make more efficient electronics, researchers have spent decades working toward magnetic transistors that utilize electron spin to control the flow of electricity. Electron spin is a fundamental property that enables electrons to behave like tiny magnets.

So far, scientists have mostly been limited to using certain magnetic materials. These lack the favorable electronic properties of semiconductors, constraining device performance.

“In this work, we combine magnetism and semiconductor physics to realize useful spintronic devices,” Liu says.

The researchers replace the silicon in the surface layer of a transistor with chromium sulfur bromide, a two-dimensional material that acts as a magnetic semiconductor.

Due to the material’s structure, researchers can switch between two magnetic states very cleanly. This makes it ideal for use in a transistor that smoothly switches between “on” and “off.”

“One of the biggest challenges we faced was finding the right material. We tried many other materials that didn’t work,” Chou says.

They discovered that changing these magnetic states modifies the material’s electronic properties, enabling low-energy operation. And unlike many other 2D materials, chromium sulfur bromide remains stable in air.

To make a transistor, the researchers pattern electrodes onto a silicon substrate, then carefully align and transfer the 2D material on top. They use tape to pick up a tiny piece of material, only a few tens of nanometers thick, and place it onto the substrate.

“A lot of researchers will use solvents or glue to do the transfer, but transistors require a very clean surface. We eliminate all those risks by simplifying this step,” Chou says.

Leveraging magnetism

This lack of contamination enables their device to outperform existing magnetic transistors. Most others can only create a weak magnetic effect, changing the flow of current by a few percent or less. Their new transistor can switch or amplify the electric current by a factor of 10.

They use an external magnetic field to change the magnetic state of the material, switching the transistor using significantly less energy than would usually be required.

The material also allows them to control the magnetic states with electric current. This is important because engineers cannot apply magnetic fields to individual transistors in an electronic device. They need to control each one electrically.

The material’s magnetic properties could also enable transistors with built-in memory, simplifying the design of logic or memory circuits.

A typical memory device has a magnetic cell to store information and a transistor to read it out. Their method can combine both into one magnetic transistor.

“Now, not only are transistors turning on and off, they are also remembering information. And because we can switch the transistor with greater magnitude, the signal is much stronger so we can read out the information faster, and in a much more reliable way,” Liu says.

Building on this demonstration, the researchers plan to further study the use of electrical current to control the device. They are also working to make their method scalable so they can fabricate arrays of transistors.

This research was supported, in part, by the Semiconductor Research Corporation, the U.S. Defense Advanced Research Projects Agency (DARPA), the U.S. National Science Foundation (NSF), the U.S. Department of Energy, the U.S. Army Research Office, and the Czech Ministry of Education, Youth, and Sports. The work was partially carried out at the MIT.nano facilities.

MIT formally dedicated its newly transformed Met Warehouse (Building W41) on Wednesday, in an energetic evening ceremony heralding the start of a new era for the practice of design on campus. 

The ceremony highlighted “the essential vision of the Met: connecting people, disciplines, and ideas, and inviting everyone to see where those connections might be,” said Mark Gorenberg ’76, chair of the MIT Corporation, during introductory remarks.

MIT President Sally Kornbluth called the new building “a testament to the transformative power of design, which has been central to MIT from the very beginning.” She also heralded the building as a cornerstone of MIT’s “magnetic new West Campus district for art and design,” which includes the Edward and Joyce Linde Music Building (W18) and the MIT Theater Arts building (W97). 

Hashim Sarkis, dean of MIT’s School of Architecture and Planning, delivered keynote remarks at the event, linking together many themes of the evening and recounting the project’s development in the late 2010s.

In envisioning moving into the Met Warehouse, Sarkis said, the “building spoke to us. It offered us possibilities that we could clearly envision both outside and inside the thick brick walls. … It helped us imagine how we can live and work together as a campus and as a school. Over the months of discussions among the faculty and students across the school, it became clear that it was an idea whose time had come.”

Many members of the MIT Corporation were in attendance for the event, held in the Met Warehouse’s Sidara Auditorium. The ceremony was also an occasion for giving thanks to those who made the new Met Warehouse possible: campus leaders, donors, faculty, architects and designers, contractors and specialized builders, and many others who worked on the remarkable structure in different ways. 

“The list, I promise you, runs longer than the credits at the end of “The Odyssey,” and what an odyssey this has been,” Sarkis quipped. 

From fortress to workshop

The Met Warehouse was originally constructed from 1884 to 1923, as a massive private storage facility with 2-foot thick brick walls, a corner turret, slit-like windows, and other features making it look like a fortress. About 500 feet long, with five stories, the building was long an imposing local curiosity.

MIT acquired the Met Warehouse in the 1970s and in the 2010s began exploring possible new uses for it. By 2018, the idea of making it the new home of MIT’s School of Architecture and Planning had gained enough traction to move forward. 

The high-profile firm Diller Scofidio + Renfro won the competition to become the project architects, and created a variety of solutions to bring natural light into the warehouse and revamp its interior. With permission from the Cambridge Historical Commission, the designers replaced four segments of brick wall on the building’s north side with top-to-bottom glass sheets, which along with skylights bring in abundant light.

At the same time, the architects — led by Elizabeth Diller and Benjamin Gilmartin, who were at the dedication last night — overhauled the building’s interior, while working within many of its structural features. The refurbished Met Warehouse now features double-height design studios, an auditorium, a large entrance lobby, a ground-floor café, offices, and many flexible, reconfigurable classroom spaces designed to help faculty and students collaborate on an immense array of projects. 

The Met Warehouse also features building-long open corridors on all five floors and a central staircase connecting all of them, as elements designed to enhance circulation and connectivity within the building.

In her remarks, Kornbluth heralded the Met Warehouse’s educational potential, noting the challenges artificial intelligence presents for education, as highlighted in an MIT-wide report released in August. 

That report, she outlined, emphasized that education comes from “helping students to value the process of learning as a ‘productive struggle’ and from engaging them in ‘human settings where they … learn how to work with others, communicate their ideas, receive criticism constructively, build confidence, develop judgment, and act as members of a community.’”

With that in mind, Kornbluth said, “that sounds exactly like the kind of hands-on, in-person, collaborative problem-solving the new Met was made for. … This community is not only ready to withstand the educational challenges of AI … it’s also primed to help the rest of MIT meet the moment. And that is very good news for us all!”

Diller also addressed the audience, highlighting some of the key design challenges involved in the project and thanking many of those who worked on it, including Leers Weinzapfel Associates, the project’s associate architects, and Shawmut Design and Construction. The Met Warehouse, she emphasized, is meant to be used in many ways in the future, and was designed with enough flexibility so that it can continue to evolve. 

“This building should remain a work in progress,” Diller said, adding that she would continue to regard it as “definitively unfinished.” 

Professors John Ochsendorf, Caroline Jones, and Lawrence Vale — MIT faculty who are all associate deans in the School of Architecture and Planning — also spoke at the ceremony, outlining the implications of the building for the school’s many forms of research and collaborative work. 

“It’s working,” Ochsendorf said, now that the building is inhabited on an everyday basis by students, faculty, and staff. 

Giving more thanks

As with almost any large, long-term project, credit can be spread in many directions, and the speakers at the dedication ceremony gave ample thanks to those involved — and to the important supporters of the project. 

“Hashim Sarkis has been its greatest champion,” Kornbluth said. “His leadership and imagination shaped not only this building, but the ambitious future it makes possible for the school.” She also thanked MIT President Emeritus L. Rafael Reif, a project supporter during his tenure, “for your foresight, your perseverance, and your insistence that the music, theater, and design communities at MIT deserve facilities worthy of the quality of their world-class work.”

For his part, Sarkis also gave credit to former MIT Corporation Chair Robert Millard ’73, saying, “Without Bob and his wife Bethany, this building would not be here today.” Gorenberg, in his remarks, also made a point of thanking the City of Cambridge for its extensive cooperation with MIT on the project. 

Gorenberg also expressed his deep “gratitude to the Morningside Foundation,” the philanthropic arm of the T.H. Chan family. He cited the “extraordinary generosity” of the founding gift, from family members Gerald and Beryl Chan and Ronnie and Barbara Chan, establishing the Morningside Academy for Design (MAD), a major interdisciplinary center at MIT located in the Met Warehouse. 

Speaking of Gerald L. Chan, Gorenberg added, “We cherish his wisdom and belief in MIT as a leading institution that can do good for the world.”

In a statement sent to MIT News, Chan said: “In this day and age when disciplinary boundaries are ever dissolving, it is important to have initiatives that tie all departments of the Institute together so that students can be facilitated to have broad exposures. Design provides such a possibility, and MAD is the venue.”

Sidara (formerly the Dar Group), a global collaborative of specialist design, engineering, and consulting firms owned by Maha and Talal Shair, supported the establishment of two central public spaces in the building, the Sidara Auditorium, and the Sidara Gallery, on the ground floor. 

“At Sidara, we share MIT’s commitment to improving lives, solving critical challenges, and making room for cultures to shine,” Talal Shair said in a statement to MIT News. “We feel a profound resonance with SA+P across all three dimensions — so it was only fitting that we would support the transformation of the MET into a hub for education, research, and innovation. We trust the Sidara Auditorium and the Sidara Gallery will serve as spaces for idea exchange, inspiring future generations to think broadly and act boldly.”

The LUMA Foundation, a Zurich-based nonprofit founded by Maja Hoffmann in 2004 to support artistic production and the organization behind LUMA Arles, an interdisciplinary creative campus in southern France, also gave an establishing gift for the MIT LUMA Lab, based in the Met Warehouse, for projects combining art, science, technology, conservation, and design. 

“LUMA Foundation has always been grounded in the belief that meaningful change begins by creating the field and conditions for people, disciplines, and forms of knowledge to encounter one another freely, critically, and with mutual respect,” Hoffmann said in a statement for MIT News. “The MET gives this principle a remarkable new context. To see faculty, students, researchers, artists, designers, scientists, and technologists connected through the MIT-LUMA Lab working alongside one another is a powerful expression of what such a place can enable. I am excited and proud to be contributing to this journey and be part of the MIT community and its future.”

Referencing MIT’s motto, “mens et manus,” which is Latin for “mind and hand,” Gorenberg wrapped up the dedication ceremony last night with an additional thought: “This is ‘mens et manus’ at its finest.” 

The Massachusetts Institute of Technology Investment Management Company (MITIMCo) announced today that MIT’s unitized pool of endowment and other MIT funds generated an investment return of 10.3 percent during the fiscal year ending June 30, 2026, as measured using valuations received within one month of fiscal year end. At the end of the fiscal year, MIT’s endowment funds totaled $29.2 billion, excluding pledges. Over the 10 years ending June 30, 2026, MIT generated an annualized return of 11.7 percent.

The endowment is the bedrock of MIT’s finances, made possible by gifts from alumni and friends for more than a century. The use of the endowment is governed by a state law that requires MIT to maintain each endowed gift as a permanent fund, preserve its purchasing power, and spend it as directed by its original donor. Most of the endowment’s funds are restricted and must be used for a specific purpose. MIT uses the bulk of the income these endowed gifts generate to support financial aid, research, and education.

The endowment supports about half of undergraduate tuition, helping to enable the Institute’s need-blind and full-need undergraduate admissions policy, which ensures that an MIT education is accessible to the most talented students in the nation and the world regardless of their financial resources. 

In fiscal 2026, MIT enhanced undergraduate financial aid, ensuring that all students from families with incomes below $200,000 and typical assets have tuition fully covered by scholarships, and that families with incomes below $100,000 and typical assets owe nothing toward their students’ MIT education. Eighty-eight percent of the Class of 2026 graduated with no debt. With our investments in financial aid, parents of MIT undergraduates receiving financial aid now pay on average 10 percent less on a real basis than at the end of the Great Recession.

MIT Student Financial Services works closely with all families of undergraduates who need financial aid to make MIT affordable for them. In 2025-26, the average need-based MIT undergraduate scholarship was $66,155. Fifty-eight percent of MIT undergraduates received need-based financial aid, and 44 percent of MIT undergraduate students received scholarship funding from MIT and other sources sufficient to cover the total cost of tuition.

MIT’s endowment enables it to do more cutting-edge research. Fueled by funding from the endowment, the Institute more than matches the amount of campus-based research funded by the U.S. government and other sponsors — expanding its beneficial impact without asking more from taxpayers. 

MITIMCo is a unit of MIT, created to manage and oversee the investment of the Institute’s endowment, retirement, and operating funds.

MIT’s Report of the Treasurer for fiscal year 2026, which details the Institute’s annual financial performance, was made available publicly today.

As far back as she can remember, Cathy Wu ’12, MNG ’13 wanted to find ways to solve problems to improve people’s lives. Her parents were Taiwanese immigrants, and her father had a long commute to his job, which took him away from the family. On a tight budget, the rest of the family often stayed home on a street that was too busy for playing outdoors. Wu and her siblings ended up playing a lot of computer games. 

Wu says her desire to make the world a better place, her dad’s daily battle against traffic, and the games she played, like “SimCity,” were the seeds of her motivation to design safe, efficient transportation systems. 

Wu is an associate professor in the MIT Department of Civil and Environmental Engineering (CEE) and the Institute for Data, Systems, and Society (IDSS), and a principal investigator in the Laboratory for Information and Decision Systems. Her research focuses on using machine learning and reinforcement learning (RL) to advance reliable strategies for improving a range of complex systems, including transportation.

“Designing transportation systems consists of modeling and analyzing dozens, if not hundreds or thousands, of variants, which means that an evidence-driven approach to designing those systems is simply not within reach of today’s tools,” Wu says. “This is the role that RL plays. If successful, it would free transportation researchers and enable their practitioner partners to design the systems they want.”

Wu credits her older sister with instilling in her the desire to improve people’s lives, and Wu’s interest in transportation fits neatly into that ideal.

“I like transportation because it connects everyone. We all use it, we all experience it, we all have issues with it. So, at some level, we’re all interested in the system being better,” she says.

Wu got interested in applying artificial intelligence to transportation while earning her undergraduate degree at MIT, after attending a lecture on autonomous vehicles by the late professor Seth Teller. The lecture, which Teller gave during an Independent Activities Period robotics competition (that Wu actually won), was the event that honed her particular approach to transportation research, Wu says. She began working with Teller, and when he stopped concentrating on autonomous vehicles, he encouraged Wu to transfer to Professor Daniela Rus, who had done research on robotaxis.

“I’m very grateful to the people who helped me explore those interests and helped me become the person I am now,” she says, specifically naming Teller, Rus, and “my friends at Dropbox,” who invited her to do a second internship focused on transportation issues.

After her master’s degree at MIT, Wu went on to earn her PhD at the University of California at Berkeley. During that time, she observed that transportation researchers were spending years developing optimization methods to model and analyze a single new variant of a system. Her approach as a computer scientist working to develop RL and optimization methodologies to address transportation challenges held the promise of exponentially improved efficiency.

In 2018, Wu’s last year of her PhD at UC Berkeley, she successfully applied RL to a traffic problem: automatically analyzing the potential traffic flow impact of autonomous vehicles in a range of different traffic networks. The research went viral.

While this could have been a “the rest is history” moment for Wu, RL turned out to be a flighty friend. Wu worked on RL theory in a postdoc at Microsoft and came back to MIT as faculty drawn, she says, by the sustainability focus of CEE, and IDSS’s emphasis on infusing data science into other disciplines.

Yet over the next two years, Wu’s further attempts to apply RL to traffic problems failed.

“That was stressful,” Wu says, “it was unclear whether the problem was me (the advisor), my students, the traffic domain, or RL itself.”

Still, the earlier research was a proof-of-concept demonstration that RL could be applied to transportation systems.

And in 2022, she and her students identified that RL algorithms are so sensitive that an algorithm that works on one problem may not on even a closely related one. A key result, which Wu says she is proudest of “because it was like the light at the end of a long tunnel of negative results,” came in 2023. She and her team of researchers devised a way to work around the sensitivity of RL. The team found that while RL may not train well on 90 percent of a group of problems, it can train quite well on 10 percent. And by training RL models on those problems that solve and generalize well, the resultant models collectively perform well on a set of related problems, even those that would not have been solved through direct training. The researchers designed an algorithm to determine which problems to use RL to train, and that algorithm improved training efficiency by up to 30 times, meaning that what would normally have required 100 training models may only require three models.

“This work gave me back the confidence that reinforcement learning can play an important role in solving hard optimization problems, including in transportation,” Wu says. “Now, a good chunk of my group works on the topic of contextual RL, which is the setting where RL seeks to solve a space of related problems.”

Wu’s more recent research applies RL to solve a hard transportation optimization problem with important policy implications: the work shows that eco-driving measures in which vehicle speeds are intelligently controlled to reduce excessive stopping and starting could reduce vehicle emissions by between 11 and 22 percent. The system provides evidence that policies instituting such measures could significantly improve system efficiency, and is “a demonstration that RL can be used to inform transportation policy on problems of practical importance,” Wu says.

“I am a big fan of evidence-based policy and believe it’s the basis for a thriving democratic society, yet our societal systems are so complex,” Wu says. “People can bicker forever about what’s better or worse, but I do believe that there are questions we bicker about that can be analyzed systematically using data and have objective answers. A large part of the reason I am in academia is to better understand how technology can support democratic societal decision-making.”

Wu says that much of the work she and her team have done over the last several years has produced algorithms “to streamline the development of solvers for hard optimization problems, whether they are related to transportation or to other systems, such as logistics, supply chains, manufacturing, and resource allocation.

“This alludes to my preferred style of work,” Wu says, “which is called use-inspired basic research,” explaining that such research addresses a practical problem, developing fundamental knowledge that often translates to other practical problems. Her students start by probing consequential problems ranging from safety to congestion to accessibility, identifying where existing methods fall short, and allowing the problems themselves to shape the direction of the research.

At the same time, Wu’s desire to help others on a more personal level plays out in her teaching.

“I love working with students, both in the classroom and research mentoring,” she says. “It makes my day when I am able to teach someone something — when I see that light bulb go on in a student.”

In addition to earning academic honors, including a 2023 National Science Foundation Faculty Early Career Development Award, Wu has also been formally celebrated for her teaching and mentoring, including with the Ole Madsen Mentoring Award in 2025.

What does she tell students confronting extremely complicated problems?

“Be patient. Start small. Societal impact is a lifelong endeavor, not something to be accomplished in a few years,” Wu says. “It will take years to really understand what’s going on and where the real problems are. In the meantime, try to be helpful. Be curious. Ask many questions.”

Despite the “dot-com crash” in 2000, the tech industry remained an attractive career destination for many who believed technology represented the future. The digital age — defined by global connectivity and computers — had firmly taken hold, and over time, the tech industry emerged as a dominant force in the labor market. High-paying jobs for engineers, designers, and professionals across a wide range of fields became increasingly common. 

MIT PhD student JS Tan SM ’22 was among those who, upon graduating from Brown University and Rhode Island School of Design in 2015, joined the tech industry. 

“A lot of us had this idea that technology, and in particular the technologies related to the internet, had the potential for bringing about a more progressive version of the world,” Tan says. 

Google echoed this ethos to its employees with its once-famous motto, “don’t be evil,” as their informal corporate philosophy and code of conduct guideline. They’ve since dropped the tagline.

But in time, particularly with the start of the first Trump administration, some tech industry employees found themselves questioning if their employers were really intent on supporting policies to support a more democratic world. According to Tan, their willingness to publicly oppose their employers’ actions — at first successfully — is currently experiencing an anti-worker backlash. 

Now, Tan and his former tech industry colleague Clarissa Redwine have published a book on the rise and fall of the tech worker labor movement. “Against Tech Oligarchy: Worker Resistance in the World’s Most Powerful Industry” (Haymarket Books, 2026) chronicles how tech workers organized themselves, the effective strategies they used, and the effect the movement had on Silicon Valley labor politics over the past decade.

Documenting a movement from within

In 2017, Tan was working for Microsoft and Redwine for Kickstarter, when President Donald Trump signed an executive order suspending entry into the United States for nationals from seven predominantly Muslim countries for 90 days, and suspending Syrian refugees from entering the country indefinitely. 

“As a whole, I think the tech sector really pushed back against this,” says Tan. “[OpenAI co-founder] Sam Altman participated in protests of this ban at the airport. In fact, the day before he joined these protests, he wrote in his blog that the tech industry needed to take a stand against the Trump administration, and particularly its immigration policies.” 

Altman’s Jan. 28, 2017, blog post read, in part, “Tech companies go to extraordinary lengths to recruit and retain employees; those employees have a lot of leverage. If employees push companies to do something, I believe they’ll have to. At a minimum, companies should take a public stance. But talking is only somewhat effective, and employees should push their companies to figure out what actions they can take.”

Tan says Altman’s words inspired tech workers across the industry to publicly voice their opposition and “push for the values they believed in.” For the next several years, workers staged walkouts and protested their employers’ contracts with U.S. military and immigration enforcement agencies, as well as workplace policies they considered sexist. 

“That was a time in which a lot of tech workers felt that their companies were walking back the values that they had initially promised,” says Tan.

Their objections initially met with some success. In 2018, following protests by Google employees, the company decided not to renew its contract for Project Maven, a Pentagon initiative using artificial intelligence to analyze drone surveillance footage. Nearly 4,000 employees signed an open letter to their CEO stating, “Google should not be in the business of war.”

Fast forward eight years. Earlier this year, more than 600 employees signed an open letter urging Google’s CEO to reject classified AI work with the Pentagon, citing concerns about potential uses including lethal autonomous weapons and domestic surveillance. 

“The way Google responded to them this time was to say, basically, ‘Too bad, we’re committed to working with the Pentagon,’” says Tan. “So, there is this kind of shift as to how Google is positioning itself politically, as well as to their employees.”

What happened to the tech industry employee leverage? 

Tan’s book outlines several events that he argues have negatively impacted their formerly strong influence. First, following interest rate hikes in 2022, the tech industry lost hundreds of billions in market valuation and set out to cut costs, the most significant of them being the expensive salaries of their employees. In other words, the labor market soured on tech workers, giving employers opportunity to wrest back control, Tan suggests. 

Second, he points to the drastic effect of agentic AI coding systems on the nature of their work, arguing that these tools have deskilled workers and made everyone much more worried about job security.  

“Tech workers had to face these shocks on their own. With a union or some ability to coordinate across workers and bargain as a group, they might’ve had more power to actually push back,” says Tan.  

Documenting the past to support the future

Tan enrolled at MIT in 2020, earning a master’s degree at the MIT Media Lab. He is currently a doctoral student in the Department of Urban Studies and Planning with a focus on the political economy of the tech sector.

At various points in their careers, Tan and his co-author had been involved with organizing in the tech sector. Redwine was a prominent organizer in the union drive at Kickstarter. She was fired from the company in 2019; Redwine said her dismissal was retaliation for her organizing activity, while Kickstarter denied that claim. One reason Tan and Redwine wrote this book is because they saw that the bandwidth for labor organizing among tech workers had hit a new low. 

“Tech workers want to have a say over the way their technologies are designed,” says Tan. “They want to be able to push for the right guardrails around technologies that they’re building. We wanted to use this book as an opportunity to analyze why it was that, within eight years, the tech worker movement had sort of fallen into this state of paralysis.”

They began writing the book in 2024, reliving the highs and lows of the past decade. This is Tan’s first book. He says writing it was an “exhilarating experience” and one that he profoundly enjoyed.

“It’s why, in part, I’m drawn to academia. To a large extent, I believe in the power of research and of writing. To have the opportunity to do this about a subject that I care deeply about has been an amazing experience.”

A book launch and discussion about “Against Tech Oligarchy,” co-hosted by the Department of Urban Studies and Planning, will take place on Oct. 26.

On the first day of class 16.445J/STS.468J (Entrepreneurship in Aerospace and Mobility Systems), David Mindell, MIT professor of aeronautics and astronautics (AeroAstro) and the Frances and David Dibner Professor of the History of Engineering and Manufacturing, asked his students to take a look at an image of a textbook. The cover featured a bright and inspiring collage of rockets, planets, and all manner of futuristic air- and spacecraft. The title: “Entrepreneurship in Aerospace: A Guide for Founders and Investors.”

“This is the most up-to-date guide on the topic,” Mindell said, describing the book’s treatment of the concepts and procedures involved in innovating in aerospace, an industry experiencing a renaissance of entrepreneurship led by venture-backed startups and rapid innovation, and one that, whether or not we realize it, we all depend on every day. 

Konark Chopra, a graduate student in MIT’s Leaders for Global Operations Program, raised his hand. “Sounds great — how can we get a copy?”

“It doesn’t exist,” replied Mindell. “You’re going to write it this semester.”

After conducting over 50 interviews with founders, operators, engineers, and investors across the aerospace industry, the class pulled it off. The 100-page industry report, titled “Entrepreneurship in Aerospace,” provides comprehensive insight into what it actually takes to build an aerospace venture today, and makes predictions about what is needed in the near future.

The challenge to produce the report was inspired by the bestselling guide “Disciplined Entrepreneurship: 24 Steps to a Successful Startup,” by Bill Aulet, professor of the practice in the MIT Sloan School of Management and managing director of the Martin Trust Center for Entrepreneurship. Originally published in 2013, the book provides an outline for entrepreneurs in any industry to cultivate skills for success. “Entrepreneurship in Aerospace” builds on Aulet’s framework to provide an industry-specific guide. “Aerospace is its own special industry with a unique set of constraints and aspirations,” says Mindell.

“Putting together this report let the students learn what they won’t get from a regular class about aerospace entrepreneurship,” he says. “This is the view of the industry from the young people building its future. I’m incredibly proud of what they’ve accomplished.”

Betting on the known unknowns

During spring 2026, while the class was in session, major developments were shaping the aerospace sector, from NASA’s Artemis II mission to SpaceX’s launch of what would become the largest initial public offering in history. “The entire industry landscape shifted like crazy during the semester,” says Mindell, as just one of the reasons that the report was especially timely. 

The report argues that the volume of capital, talent, and policy attention directed toward aerospace is “structurally different from any prior point in the industry’s history,” creating a pivotal moment for the next generation of entrepreneurs and investors.

The report further outlines five predictions about the new space economy: terrestrial infrastructure for compute, manufacturing, and energy will move off-world; autonomous systems, robots, and humans will continue to work together, but with humans in a supervisory capacity; venture capital investment will run ahead of economic justifications; government investment will become the fastest way for young companies to fundraise; and that we are years away from a global regulatory framework for space companies to operate within. Each prediction, or “bet,” includes a section on “where serious people disagree,” laying out relevant counterarguments to their conclusions.

As a guide, the report also translates its findings into practical tools, including a diagnostic for determining how many independent breakthroughs a company needs to succeed. “One of my favorite tools from the report is the miracle count,” says Chopra. “If your company needs one breakthrough to work, that’s a venture bet. If it needs three, that’s a research project pretending to be a startup.”

The report’s findings were informed both by existing research and by interviews with current aerospace professionals. Students were graded, in part, on how many people they spoke with. Their interviews focused on what excites people about the industry, reasons for their optimism (or pessimism), and how people are working within their organizations to address the challenges they see. 

The experience also allowed the students to expand their professional networks, practicing a core entrepreneurial skill while gaining insider perspectives.

“We sought out people from different corners of aerospace and were always asking, ‘Who else should we talk to?’” says Nicole Lee, a graduate student in AeroAstro. “Not only was I able to reconnect with people in my own network, but I got to introduce classmates to those contacts, and then benefit from the networks they brought in, too. That exchange was a big part of what made the interview process, and the class, so special.”

Engineers as entrepreneurs

The report’s authors — 16 classmates from the Department of Aeronautics and Astronautics, MIT Sloan School of Management, and Wellesley College — bring a range of academic backgrounds and career ambitions to the project, using those different perspectives to connect the realities of aerospace engineering with the economic forces impacting the industry. Their research interests and experiences range from spacecraft propulsion and human spaceflight to investment banking and military operations. Collectively, they have worked across organizations like NASA, SpaceX, Blue Origin, Boeing, and a range of startups.

“What I’ll remember most is the team,” says Chopra. “Everyone showed up with their own wisdom and a willingness to challenge each other, learn from each other, and simply have fun. Those are the teams we hope to keep building with.” 

For Lee, those industry experiences support the report’s predictions about where the field itself is headed. “Entrepreneurship in space is going to involve a much broader group of founders, engineers, researchers, policymakers, and operators,” says Lee. “Everyone working in space can take something away from understanding the entrepreneurial mindset and the cultural shift we’re seeing in commercial space. A much wider range of people will be shaping entrepreneurship in the future. I think that shift has already started with us.”

Mindell sees value in that entrepreneurial mindset, regardless of whether the students go on to found companies of their own. “I don’t know if every student in this class is going to found their own company, and I don’t expect them to, but I do expect them to drive their own careers forward,” he says. “And I think for the moment we’re at, providing that opportunity is the best thing MIT can be doing for our students.”

For Chopra, who has had his sights set on founding an aerospace company for as long as he can remember, the findings from the report are immediately applicable. “The heart of a business, a sustainable business, is the demand. What do the customers want? Sure, I could build cool technology, but if we don’t have anyone buying it, it’s a project, not a company.”

Now armed with a clear and evidence-backed picture of the landscape, Chopra wants the report to generate even more activity across the industry. “If our industry report inspires one person to go out and found a company, or invest in a company, or even think about entrepreneurship in aerospace, it’s a pretty big win.”

The book “Disciplined Entrepreneurship” by Bill Aulet, managing director of the Martin Trust Center for MIT Entrepreneurship and the Ethernet Inventors Professor of the Practice at the MIT Sloan School of Management, walks readers through the 24 steps of starting a venture. With more than half a million copies sold, the approach has proven remarkably effective: MIT students who use the framework in the delta v startup accelerator program have a 61 percent survival/acquisition rate and have collectively raised over $3 billion dollars.

But while the framework is taught at hundreds of colleges around the world, it is not designed for younger students who want to learn about entrepreneurship. To fill that gap, the Trust Center created a free, AI-powered youth entrepreneurship platform called Dear Dreamer, made possible through a gift from the Frank and Eileen Foundation. 

Dear Dreamer is open to all students in middle and high school. It adapts the disciplined entrepreneurship framework into a digital, self-paced learning experience featuring short educational videos, interactive exercises, and personalized feedback on the user’s idea. Aulet says the goal is to empower 50,000 young entrepreneurs by 2030.

MIT News spoke with Aulet about the mission of the project and how it came together.

Q: What was the impetus for creating Dear Dreamer?

A: We’ve had a lot of success teaching the disciplined entrepreneurship framework at MIT. Then we made it a course on the online learning platform edX, and we got hundreds of thousands of people taking the class. I used to get emails from people saying, “For the first time in my life, I see myself as an entrepreneur,” or “I see economic security.” At MIT, we often say, “MIT in the world, for the world.” At the Trust Center, we see our mission as, first, train people here at MIT, but also to create more entrepreneurs outside of MIT. 

The platform was inspired by the vision of Audrey McLoghlin, the founder and CEO of apparel brand Frank and Eileen and president of the Frank and Eileen Foundation. In my first meeting with Audrey, we were already asking ourselves, “How do we make more entrepreneurs in the world?” I showed her Jetpack, MIT’s generative AI tool trained on disciplined entrepreneurship, which walks MIT students through the entrepreneurial process, and her eyes lit up. She said, “This is how we create more entrepreneurs: We make the work that you guys are doing here accessible to young people.” She wished she had a tool like this when she first started. She also said her daughter wants to be an entrepreneur, but students don’t get much guidance on entrepreneurship at school. MIT has been a great place to keep making this more accessible.

Q: How does the platform work?

A: It takes the core disciplined entrepreneurship curriculum that we know works from our data, and makes it more interactive and engaging for young people. We spent a lot of time working with younger students to figure out how to make it more digestible to a 10-year old, 12-year old, or 14-year old. We put in different case studies and examples for each step of the framework, and we’ve designed the user interface to make it more like Instagram or YouTube, with videos featuring people that students can relate to. But ultimately the content follows the same path we know works; it’s just presented differently from what you would present to an MIT MBA or PhD.

Q: How might learning about entrepreneurship benefit students?

A: Entrepreneurship is a mindset, a skillset, and a way of operating. It allows you to deal with change, and the world’s rate of change is going faster and faster. It’s really not just about founding companies. Founding companies is a great way to learn the mentality that, ‘We can be different. We can build something. We can achieve a lot.’ Someone once said, ‘If you give a person a job, you give them dignity.’ But if you make someone an entrepreneur, it’s like giving them super dignity. They go from a job seeker to a job creator, and they can focus on the things that they’re most interested in, working with the people they want, in the culture they want. We teach students the entrepreneurial mindset and we train them to systematically take an idea and then come up with a solution in an ambiguous, uncertain environment, and iterate on that. 

Everyone might not start a company, but even if they go to work at bigger companies, there are a lot of benefits to the entrepreneurial mindset. I’m an entrepreneur now and I’m at MIT, which is not a startup. In a world that’s moving faster and faster, everyone has to deal with change and ambiguity. Starting companies is just a great way to learn. It’s kind of like learning to paint from a blank canvas. 

In 2017, a man named Michael walked onto stage in front of a packed Kresge Auditorium at MIT and attempted to draw a spiral, a common test doctors use to diagnose Parkinson’s disease. His tremors, caused by the disease, made the exercise difficult. 

Then, Michael put on a wristband device made by a team of MIT students as part of 2.009 (Product Engineering Processes), who were presenting their prototype that evening.

Michael pressed a button and the device produced a subtle vibration. The vibration sent signals up his wrist and into his brain. His tremors dramatically decreased, and within seconds he was able to draw the spiral with much more precision, to a roaring ovation from the audience.

“When this device is turned on, I feel like I used to feel when I didn’t have Parkinson’s disease,” Michael told the crowd. “It’s an amazing, amazing feeling.”

The performance was so impressive that the student team received requests from classmates and others asking where they could buy the device for family and friends living with tremors. Unfortunately, the students had to explain there was only one — for the time being.

The event set off a near decade-long journey that began by leveraging MIT entrepreneurial resources like MIT Sandbox, MIT FUSE, and the MIT Venture Mentoring Service. In 2020, the student team turned into an official company, Encora Therapeutics. But there were still dozens of hardware iterations ahead. Then there were clinical trials. In one trial, 78 percent of patients reported benefits after 90 days of home use.

This February, nine years after Michael’s brave demonstration, all that work finally paid off: The FDA cleared Encora’s device to help adults with essential tremor, a condition similar to Parkinson’s that causes shaking, often in the hands. 

“It’s been a long and difficult — very difficult — journey, but also very rewarding, especially when we get feedback from patients,” says Daniel Carballo ’18, SM ’20, an Encora co-founder and vice president of strategy. “We hear stories from patients about how they’ve struggled with their condition and how much they benefit from this. It reminds us why we keep going.”

From classroom to commercialization

The three founders of Encora who are still with the company are Carballo, Allison Davanzo ’18, and Kyle Pina ’18. They were each seniors in 2017 when they enrolled in 2.009, MIT’s popular product-design class.

The semester began with a brainstorming session in which groups of about 17 students were asked to come up with dozens of potential product ideas. Carballo proposed a wearable device that used mechanical vibration to send signals to the brain to reduce tremors. The initial idea was to help patients with Parkinson’s disease.

“It was one of hundreds of throwaway ideas,” Carballo recalls. “The initial concept was inspired by classes I had taken in robotics around neural control of movement. I had a preliminary understanding of how an electromechanical device might interact with the body’s control systems and feedback loops that control movement to relieve pathological control of movement.”

The team eventually whittled their long list of ideas down to a few. Carballo’s idea was finally selected by a vote of 16-1 — with Carballo the only dissenting vote.

“I tried to explain to the team that this was so far-fetched that there was no way, in one semester, we would be able to make anything,” Carballo recalls. “Thankfully, I got outvoted.”

Through most of the semester, Carballo’s pessimism looked justified. At every class milestone, the team lagged behind other teams. Then, the week before final presentations, they met Michael, whose severe hand tremors were the result of early-onset Parkinson’s.

“He was a home renovator, so he worked with his hands, but his tremors had progressed to the point that he struggled to turn a screwdriver, use a drill, or even fill out paperwork,” Carballo recalls. “He had become reliant on his wife and daughter not only to run his business, but to help him with everyday activities.”

By this point, the 2.009 team had a prototype that would vibrate to send mechanical feedback to the brain. Carballo described it as “a foamcore box with a Raspberry Pi chip and some wires coming out.” When Michael put on the device and turned it on, his tremors dramatically decreased.

“It was like a light switch turned on and his tremors stopped,” Carballo says. “His wife and daughter started crying. A week later, he was gracious enough to repeat the process on stage for the final presentations.”

The presentation — and the outpouring of interest from people who wanted it for loved ones with Parkinson’s — made the team determined.

“It showed us there were a lot of people with this problem that could really benefit from this,” Carballo says. “A subset of the team became possessed. It would have been such a shame to know this could exist and to have it never leave the classroom.”

Some team members began using MIT’s entrepreneurial resources to commercialize the technology, initially focusing on Parkinson’s. They ran their first clinical trial with 20 Parkinson’s patients in 2022 in collaboration with MassGeneral Brigham. But they soon learned more patients are living with essential tremor.

“It was a greater unmet need,” Carballo says. “There are a lot of drugs being developed for Parkinson’s disease, but essential tremor hasn’t experienced that same innovation.”

Today scientists think tremor is caused by malfunctioning signals in the regions of the brain responsible for interpreting sensory input and coordinating movement.

“In these diseases, neurotransmitter deficiencies result in this pulsed signaling in the brain that manifests as tremors that are basically pulsed motor outputs,” Carballo says.

One current approach to target those signals is surgery that involves drilling holes in the skull to insert electrodes that drive new electrical patterns in the brain. Encora’s watch-like product targets the same brain regions with mechanical vibrations at the wrist.

“We’re applying mechanical stimulus, basically vibration, to stretch receptors in the wrist, which tell your body where it is in space,” Carballo explains. “The stimulation causes the receptors to activate and send a patterned signal to peripheral nerves of the wrist, that then carry the signal to the peripheral nervous system and into the central nervous system, to the same regions of the brain targeted by surgery.”

In 2024, Encora ran a randomized control trial with 47 patients living with essential tremor. Last year, the team ran a 59-person trial where patients used the devices at home. In both trials, more than 70 percent of patients experienced meaningful improvement.

The results were promising enough to gain what’s known as 510(k) clearance from the FDA for use as a medical device, in February of this year.

Helping patients

Most patients see rapid benefit when using Encora’s device. Patients have described the device as life-changing. Some say it allows them to do tasks they haven’t been able to do in years.

“We see some patients using the device 12, 14 hours a day,” Carballo says.

Today, Encora is focused on building a national sales force and working to secure coverage from insurers. As the company ramps up production, the product will finally become available for patients who qualify.

Further down the line, Encora hopes to fulfill its original mission of helping mitigate tremors in patients with Parkinson’s. Carballo believes the approach also holds promise for many other patients.

“There is a surprisingly long list of diseases that this could work for,” Carballo says.  “The most obvious are neurological movement disorders, but the bigger picture of wearable neuromodulation is a rapidly growing field that has seen therapeutic benefit across a broad range of diseases. We see this as a platform technology.”

“What you see is what you get” is a guiding principle for many software engineers — create programs where the content you’re editing looks the same as the final product. But when you’re using generative artificial intelligence systems to 3D print, say, a mug, you’ll likely get a cup that can’t hold your coffee. Why is that?

The issue is that AI models understand how an object should look, but not how it works, leading to impractical designs that undermine an item’s intended use. Even if you want to fix these errors, the models are typically hard to edit, especially for users new to 3D design.

A new approach called “InstructMesh” makes it much easier to design and print household items, accessories, and robots that work in the real world. The design software, which was developed by researchers at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL), Google, and Northeastern University, can be prompted to generate a 3D design for a pair of glasses, for instance, and users can then highlight specific parts of the blueprint they’d like refined before 3D printing. It’s an AI-driven interface designed to understand how these designs should look and which edits experts and novices alike want to make, helping them create the objects they actually want to see.

CSAIL researchers used InstructMesh to put a personalized, creative spin on otherwise regular items. For example, the tool produced a mug that appears to be enveloped by a dragon, with its tail being the handle. It also fabricated a shiny blue whistle resembling a shell and a pair of glasses with butterfly wings spreading out just above the lens. Getting even more creative, it made an octopus-like dispenser, with liquids flowing out of each tentacle to distribute drinks into several cups at once.

What makes InstructMesh so adept at following such unique prompts? It pairs Microsoft’s TRELLIS system, which creates 3D models from text and image prompts, with the large language model (LLM) GPT-4, which supports ChatGPT — in other words, visual and textual knowledge combined.

“We wanted to bring together the talents of 3D generators and the reasoning skills of LLMs in an interactive space to make objects that people actually want,” says Faraz Faruqi SM ’22, PhD ’26, lead author on a paper presenting the project, graduate of the Department of Electrical Engineering and Computer Science, and recent CSAIL affiliate. “Language models are great at text and images, while TRELLIS’s talent lies in its ability to create 3D models, since it’s seen so many.”

InstructMesh’s strengths come in handy in other, more surprising areas. MIT scientists used the program to fabricate a knee brace that looks like denim to match a patient’s jeans. InstructMesh can even help create robots — that is, clever enclosures that house wireless components. The researchers made a “bristle bot” that resembles a colorful shrimp to demonstrate this. It has a motor hidden inside, and when switched on, it can slide across surfaces, sort of like a wind-up toy.

Faruqi and his colleagues found that InstructMesh could easily make their desired items. But what would someone who’s never 3D modeled anything think of their program? And could they really detect design flaws before fabrication?

The team has TRELLIS recreate popular 3D models found on Thingiverse, a platform home to millions of 3D printable models, to help them find out. Nearly 80 percent of the models it generated were structurally flawed in some way. CSAIL researchers then asked novices to identify and fix these issues in InstructMesh — and they were able to do both around 90 percent of the time, as reviewed by an expert. What these newcomers lacked in expertise, they made up for in intuition.

InstructMesh scaffolds the actual modeling process, which previously required domain expertise in 3D modeling tools. “With manipulation happening in the latent space of the generative model, InstructMesh supports natural language description of issues, and creates interpretive changes in the geometry for the user to evaluate and approve,” says Faruqi. 

Users then created items resembling things like phone stands and vases, noting that InstructMesh was easy to use. They also found that InstructMesh enabled them to express a wide range of ideas, while the sliders gave them more precision to make certain tweaks, such as enlarging or extruding a particular part of the model.

“The users got what they prompted for and easily tweaked designs where needed,” adds Faruqi. “What they saw is what they got, and the items worked as advertised, so to speak.” 

While users enjoyed using the InstructMesh, Faruqi has an even grander vision for the project. He now works at Google, where he may soon incorporate InstructMesh into an augmented reality (AR) platform. The idea: Prompt the system by explaining what you need using the context of your surroundings, then it’ll rapidly 3D print it (e.g., making a phone case that matches your wallet).

InstructMesh may also begin to incorporate physics simulations to model how your design may react to specific uses, such as whether a bowl breaks when dropped, and which materials would work best. The software might also integrate the more recent TRELLIS.2 to refine even smaller features in 3D models.

Stefanie Mueller, an associate professor of electrical engineering and computer science (EECS) and mechanical engineering at MIT, and a member of CSAIL, is a senior author on the paper. Faruqi and Mueller wrote the paper with Google researchers Ahmed Katary ’23; Fabian Manhardt; Vrushank Phadnis MEng ’13, PhD ’20; Ruofei Du; and Federico Tombari. Other co-authors were Northeastern University Assistant Professor Megan Hofmann along with several CSAIL colleagues: Demircan Tas SM ’24 and SMArchS ’24, a PhD student in EECS and architecture; former visiting researcher Theresa Hradilak; Ning Zhang ’25, a graduate student in EECS; postdoc Jiaji Li; and Martin Nisser SM ’19, PhD ’24.

The researchers’ work was supported, in part, by Google and the MIT-HPI Collaborative Research Program. They will present it at the ACM Symposium on User Interface Software and Technology in November.

MIT Future Fest launched on Wednesday afternoon, featuring a visit from Massachusetts Governor Maura Healey before an opening panel at Kresge Auditorium.

Both Healey and MIT President Sally Kornbluth emphasized the power of curiosity and innovation during introductory remarks, while formally kicking off MIT Future Fest as a new kind of event that opens MIT’s doors to the world.

Spanning five days and featuring talks from leading MIT scholars, panel discussions, open labs, exhibitions, performances, screenings, and much more, MIT Future Fest adds up to a demonstration of campus research, teaching, thinking, and discovery.

“MIT and Massachusetts are known around the world as leaders in innovation,” Healey said. “Future Fest is a chance to celebrate that. But it’s also a chance to connect with each other. To learn from each other, and to imagine what comes next. In life sciences and biotech, robotics, advanced manufacturing, in energy and climate technology and quantum, and in the people who make it all happen.”  

In remarks preceding Healey’s talk, Kornbluth observed that “in a world as complex and interconnected as ours, we aim to invent the future. We need all the help we can get, and that’s why it’s so important to have all of you join us over the next five days in exploration, conversation, and curiosity.”

Future Fest also underscores that at MIT, as Kornbluth put it, “We feel a drive to explore, to understand, to solve, to invent, to design, all in service of society.” 

Since its founding, “MIT has long stood at the forefront of innovation and invention,” Healey noted. “‘The future starts here’ isn’t just a fitting theme for the next few days. It’s something we’ve been doing for the better part of 250 years in Massachusetts.”

Healey later appeared at an event about ocean sciences at the MIT Museum, whose staff have helped develop and organize MIT Future Fest. 

All told, MIT Future Fest features over 100 speakers at more than 50 events, among many other activities. The MIT Museum is free to the public during Future Fest; MIT’s newly opened Met Warehouse is hosting its first public events on Saturday; and Sunday features the Cambridge Science Carnival, a series of family activities. 

The remarks by Healey and Kornbluth were followed by a panel discussion, “The Future Begins Here,” featuring participants Noubar Afeyan PhD ’87, founder of Flagship Pioneering and co-founder of Moderna; Sangeeta Bhatia SM ’93, PhD ’97, the John J. and Dorothy Wilson Professor of Health Sciences and Technology and of Electrical Engineering and Computer Science at MIT, and director of MIT’s Marble Center for Cancer Nanomedicine; and Bob Mumgaard PhD ’15, CEO and co-founder of Commonwealth Fusion Systems. 

Massachusetts Economic Development Secretary Eric Paley moderated the wide-ranging discussion, which touched on the cutting-edge Massachusetts innovation ecosystem, the implications of artificial intelligence for research and industry, and more. 

The panelists noted that a confluence of ingredients — universities, other research institutions such as hospitals, venture capital, and more — has helped make Massachusetts the world leader in biotechnology and other forms of innovation. 

“You can’t be here and not feel like things that the rest of the world feel are not possible are possibly possible,” Afeyan said. “It’s really interesting how the environment makes you feel like you should also do something special.” He added that the various ecosystem elements have created a “critical mass” needed to sustain broad-ranging innovation. 

“This critical mass accumulated between the academic science, the hospital research, the entrepreneurial companies, large companies moved in, the government stepped in, state level, to try to help this community,” Afeyan observed. “It’s an inherent advantage of any ecosystem, it’s the co-location, it’s the connections.”

Mumgaard, whose company is building the first fusion power plant in the U.S., concurred that the combination of elements available in the state ecosystem is crucial for advanced innovation. 

“In many ways the challenge is simply scale,” Mumgaard said. “How do you get enough capital, how do you get enough smart people, how do you get enough people that are hard workers, how do you get the alignment from the institutions that can give you the permissions and tailwinds to be able to go and do that? And we did find that in Massachusetts.” 

He added: “You could not have really have built this company outside of Massachusetts.” 

For her part, Bhatia, a leader in developing nanoscale technologies that have made inroads in battling numerous forms of cancer, observed that part of the inspiration for her career came from visiting the MIT campus as a high school-age student.

“It was just that idea that you could engineer things, you could create instruments to improve human health, that started me on my journey,” Bhatia said. 

As a faculty member, she has helped found eight spinoff companies and helped direct an MIT study about how the Institute — which has already helped generate about 30,000 new companies — can do even more to produce startups and technology transfer. Ultimately, she has concluded, teaching researchers about the tools and skills they need to found companies can help generate even more productive networks of entrepreneurial activity.  

“We can be more intentional. We can create communities. … There are all kinds of ways to do more,” Bhatia said.

The panel discussion included reflections on AI and forecasts about the future of technology 20 years out, while ending with a reminder from Afeyan that, while the future is hard to forecast, it is at least partially shaped by people working consistently hard in pursuit of their goals. 

“We can envision the future a lot better than we can predict it,” Afeyan said, when asked to forecast the effects of technology in a couple of decades, near the end of the discussion.

Afeyan concluded: “My parting thought on this would be to say … I think you can either live in the present, do the things you do, and get the future you deserve, or you can envision the future you want, and do everything in service of that.”

Google.org announced on Sept. 15 that the MIT Transit Lab is a recipient of $2.1 million in funding — one of only 15 projects selected in the worldwide Google.org Impact Challenge: AI for Government Innovation. The funding from Google’s philanthropic arm will support NGOs, social enterprises, and academic institutions as they integrate artificial intelligence-powered solutions across topics like health, resilience, and economy.

The Transit Lab’s winning project, the Public Transit Intelligence Hub (PTIQ), aims to unify public transportation agencies’ real-time monitoring, operations control, and passenger communication systems into a single centralized AI-orchestrated platform that will allow transit control center staff to make better-informed, on-the-spot decisions, and provide riders with more immediate and accurate information.

The control centers of public transportation agencies are similar in appearance to the portrayal of NASA mission control in movies: rooms filled with employees monitoring dozens of radio feeds and computer screens relaying real-time camera data about stations and their operations, transit vehicle locations, riders, traffic, and road conditions. Unfortunately, the information coming in is fragmented, rather than integrated into a centralized system with overall awareness of the network’s conditions. This system creates an intense work environment for the transit staff making operations and communications decisions that can affect thousands of passengers relying on transit to get them where they need to go.

“Public transportation agencies are required to make decisions around the clock regarding real-time operations, control, and passenger communication,” says Awad Abdelhalim, associate director of the Transit Lab, and PTIQ co-principal investigator, project director, and technical lead. “Our goal isn't to automate those decisions, but to make sure the people making them have the best information possible. By unifying and streamlining data and information flow from fragmented and siloed internal systems, PTIQ will improve the experience of both riders and the transit workforce.”

Jinhua Zhao, the MIT Class of 1941 Professor of City and Transportation, head of the MIT Department of Urban Studies and Planning, and founder and director of the MIT Mobility Initiative (MMI), is the other co-principal investigator on the project. The PTIQ program manager is MIT Lecturer Jim Aloisi, who directs the Transit Research Consortium, which will also work on the project. That consortium is comprised of researchers from the Transit Lab, MMI, and Northeastern University, where Professor Haris Koutsopoulos takes the lead. 

In addition to providing funding for the three-year project, Google.org will provide pro bono support from its own engineers and AI product experts.

"AI holds incredible potential to transform public services, but there is often a gap between promise and practice,” says Maggie Johnson, global head of Google.org. “By equipping the 15 selected organizations with funding and pro bono support from Google's own AI experts, we are empowering the people closest to the problem to show what is truly possible. Together, we can ensure that AI makes a profound, positive difference in the everyday lives of communities worldwide."

The project will build on the group’s decades of experience in applied-research collaborations with transit agencies in major metropolitan areas throughout the world. PTIQ’s decision support interface for control center staff will integrate predictive models, optimization engines, and large language model-based contextual reasoning. But ultimately the decision-making based on that information will be left to transit staff, who can better balance the trade-offs of making one decision over another in these often incredibly complex situations.

“The hard part of integrating AI in transit is not the technology; it’s the institution,” Zhao says. “AI is evaluated on benchmarks. Public transit is assessed in the control center and on the streets. Over decades of work with transit agencies in Washington, D.C., Chicago, London, Boston, Tokyo, and Hong Kong, we have learned to ask a different question. Not whether AI can do this, but whether it can work in the organization and whether the staff trust it. PTIQ is designed to ground AI in the institutional reality and behavioral nuances of a transit agency, and bring machine intelligence and human judgment into one place.”

“Currently the evaluation of AI models relies heavily on deterministic, objective tasks, such as solving mathematical equations or generating code,” Abdelhalim explains. “However, the vast majority of real-world operational tasks — like delivering public transit services — are highly dynamic, multi-stakeholder, and lack a single correct objective answer. These complex spatiotemporal environments are the ultimate testbed for evaluating what AI systems can add to society.”

PTIQ aims to transform the transit workforce experience, the transit rider experience, and the overall ability of transit agencies to efficiently respond to disruptions and unexpected events. 

“We expect that PTIQ will take what is largely a siloed environment and connect it in ways that provide powerful benefits for the agency workforce and its riders,” says Aloisi, who is also a former secretary of transportation for the Commonwealth of Massachusetts. “[Doing this by] improving response time, reducing platform and bus stop crowding, providing riders with higher quality and timely information, and supporting agency staff — from dispatchers to vehicle operators and communications staff — with high-quality, reliable, real-time information and solution sets.”

A new AI system that excels at challenging games with hidden information could someday help human decision-makers select ideal strategies to outfox opponents in complicated situations like military maneuvers.

Using advances in machine-learning, researchers from MIT, Carnegie Mellon University, New York University, and Stanford University developed an AI that defeated top-ranked human players of the board wargame Stratego by a large margin — something no AI system had been able to achieve. 

Stratego, a two-player game of imperfect information, in which the opponent’s piece identities remain hidden, is often used as a benchmark to test the strategic thinking abilities of powerful AI models.

To build their model, the researchers combined efficient training algorithms with new techniques tailored for calculated decision-making in hidden information settings. 

The AI system achieved greater performance at Stratego than the next best models, while being far cheaper and less computationally demanding to train. The system also outperformed top human players in other strategic games with different rules and designs, demonstrating how it can be generalized for a variety of use-cases.

The AI system could be adapted to help humans tackle many real-world problems with hidden information, such as business negotiations or cybersecurity. 

“In the kind of imperfect information tasks you would face in reality, you often don’t have the luxury of enumerating through all the possibilities. There are just too many. Having AI algorithms that are general purpose and can provably perform this challenging task so well is a big step forward,” says Gabriele Farina, an assistant professor in the Department of Electrical Engineering and Computer Science (EECS), principal investigator at the Laboratory for Information and Decision Systems (LIDS), and senior author of a paper on this AI system.

He is joined on the paper by lead author Samuel Sokota, a graduate student at Carnegie Mellon; Eugene Vinitsky, an assistant professor at NYU; Zico Kolter, a professor at Carnegie Mellon; Hengyuan Hu, a graduate student at Stanford; and Zhiyuan Fan, an EECS graduate student at MIT. The research appears today in Nature.

Hidden information

The world is full of imperfect information problems. 

In these interactions, some parties possess information others do not. For instance, traders in financial markets may not know the rationale behind the trades of others, while military forces likely don’t have full knowledge of enemy positions. 

With hidden information, the decisions parties make, as well as the decisions they choose not to make, are intertwined in such a way that it is extremely difficult to determine the best steps to take next.

“The more you bluff, the more your opponent expects it, and the less each bluff is worth. It’s not obvious how to reason about that,” Sokota explains. “It’s very different from a setting like chess, where the best move is still the best move no matter how often you’ve played it.”

Stratego is often used to model imperfect information situations. In this board wargame, which resembles military chess, players arrange 40 pieces on their side of a board and then move pieces across the board to capture their opponent’s flag. 

But the identity of all pieces remains secret until they collide, and then the lower-ranking piece is eliminated.

The possible piece configurations number more than 10 to the 66th power — an exponentially greater number than in chess — making Stratego extremely difficult for an AI system to play well. 

Past efforts, such as Google’s DeepMind, relied on sophisticated operations that were computationally demanding and costly. But even with millions of dollars in training costs, these models were still not strong enough to beat top human Stratego players.

“With Stratego, there is an explosion of possible universes you might have to deal with. AI techniques that were developed for games like poker definitely could not scale in this setting,” Farina says.

The MIT researchers set out to develop a full AI system that could achieve superhuman performance for less cost, which they called Ataraxos (a Greek word used to describe one who is unbothered or free from anxiety).

A two-pronged approach

To build Ataraxos, the researchers trained the model using a technique called self-play reinforcement learning. The model plays against itself many times to learn a strong “blueprint strategy” of how to excel at Stratego. 

They designed especially efficient algorithms, which enabled Ataraxos to learn much faster than prior methods while ensuring it didn’t get stuck trying to predict every possible move. This reduces training costs and boosts performance. 

“Our system reaches strictly higher playing strength than DeepNash (DeepMind’s system) while using less than one hundredth of the training examples and less than one thirtieth of the self-play games, indicating a massive improvement in efficiency,” says Farina.

During a game, Ataraxos uses the blueprint strategy as a starting point to set up the board and begin thinking about its next moves at each round of play. 

But before acting, it refines its choices on the fly using a technique called decision-time planning. The system employs a generative model that uses probabilities to estimate the likely identities of the opponent’s hidden pieces, then evaluates future choices before selecting the next move. 

“Rather than just guessing blindly, we use decision-time planning to find the most plausible state of the board. Using this generative model allows us to really zoom in on the specific board and opponent we are facing,” Farina says.

The innovative use of this generative model for decision-time planning was the missing piece that enabled Ataraxos to achieve superhuman performance.

Ataraxos beat the strongest Stratego player in the world by a record margin of 15-1-4 and achieved a 39-2 record against top human players at the Stratego world championship. “Ataraxos is good at calculating risk in a way that humans are not. A human might start freaking out if their most valuable piece is exposed, but the bot can be surprisingly composed. It doesn’t overcorrect and give away its secrets,” Farina says.

The researchers also adapted Ataraxos for other imperfect information games, including Barrage Stratego (a faster-paced variant with fewer pieces), Hanabi (a cooperative card game with many players), and Dou dizhu (a game in which two players cooperate against a third).

The system achieved superhuman performance in each instance, demonstrating the generality of this method.

In the future, the researchers want to build interpretability measures into Ataraxos so the system can explain its decision-making in a way that a human could understand. 

“Humans must have the final say in whether a recommendation is followed, so before adoption can happen, we need a way to audit the model’s decisions. We still have a long way to go, but I hope these algorithms can be the foundation for a lot more work to come,” Farina says.

This research is funded, in part, by the Office of Naval Research, the New York University Department of Civil and Urban Engineering, the C2SMART Center, the National Science Foundation, and a Schmidt Sciences AI2050 Early Career Fellowship.   

About 25 percent of lung adenocarcinomas have mutations of the gene KRAS, which drives uncontrolled cell growth. In recent years, the FDA has approved two KRAS inhibitors to treat patients with KRAS mutations. While these drugs can work well initially, tumors almost always develop resistance to them.

Usually, resistance emerges because cells reactivate KRAS activity, through mutations that prevent drug binding or by increasing KRAS expression that overpowers the effects of the inhibitor. However, in a new study, MIT researchers have modeled an alternative mechanism that cancer cells can use to become resistant to KRAS inhibition.

The researchers found that in some cases, lung tumors undergo transformation from adenocarcinoma to squamous cell carcinoma. Both of these tumor types are commonly found in the lungs, but they are thought to  arise from different cells and have different genetic profiles.

When this transition occurs, tumor cells no longer require KRAS, and appear to turn on alternative signaling pathways that help them continue to grow. Ongoing work to identify those pathways may reveal targets for new drugs that could help prevent resistance to KRAS inhibitors.

“The main takeaway is that there seem to be different routes of resistance to KRAS inhibitors, and so we need to be thinking about how we can address this,” says Carrie Rodriguez, an MIT graduate student and one of the lead authors of the paper.

Nicolas Mathey-Andrews PhD ’25 is also a lead author of the study, which appears today in the journal Nature Genetics. The paper’s senior author is Tyler Jacks, the David H. Koch Professor of Biology and a member of MIT’s Koch Institute for Integrative Cancer Research.

Tissue transformation

The two FDA-approved KRAS inhibitors both target a mutation called KRAS-G12C. These drugs are approved only for use in patients whose tumors have failed to respond to other drugs, and these patients usually have cancer that has spread beyond the lungs.

KRAS inhibitors are effective in about 35 percent of the patients who receive them. However, in those cases, the tumors almost always end up becoming resistant by generating additional copies of the KRAS gene or finding other ways to turn on the MAP kinase signaling pathway, which is usually triggered by KRAS and stimulates cell growth.

“Resistance to targeted therapies is a very serious problem,” Rodriguez says. “Sometimes these KRAS inhibitors can hold cancers at bay, but most cases do end up relapsing.”

A 2021 study from researchers at Dana-Farber Cancer Institute, which analyzed tumors from 17 non-small cell lung cancer patients treated with KRAS-G12C inhibition, identified secondary resistance mutations in a majority of patients. In two of these patients, however, the researchers found that tumors transformed from adenocarcinomas to squamous cell carcinomas, but they did not harbor obvious resistance mutations.

Both adenocarcinomas and squamous cell carcinomas are classified as non-small cell lung cancers (NSCLCs), which are the most common type of primary lung cancer. Adenocarcinomas, the most common type of NSCLCs, often originate from the surfactant-producing cells that line the lungs, while squamous cell carcinomas originate in the cells that line the central airways of the lungs.

Mutations of KRAS are found much more frequently in adenocarcinomas than in squamous cell carcinomas

In this study, the researchers set out to model the factors that might drive the transition from adenocarcinomas to squamous cell carcinomas. To do that, they engineered a mouse lung cancer model to express the mutation that is targeted by the FDA-approved KRAS inhibitors. 

Following treatment with a KRAS-G12C inhibitor, tumors with genetic loss of Nkx2-1, which normally helps maintain alveolar epithelial identity, were able to undergo adeno-to-squamous transition. Turning on a transcription factor called DeltaNp63, which is overactive in many squamous cell carcinomas, also made this transition more likely. Another transcription factor known as SOX2 also helped stimulate the transition, but this gene could not initiate the transition on its own.

Paths to resistance

Tumors that underwent these tissue transformations did not acquire the mutations that typically boost KRAS expression in adenocarcinomas. Instead, KRAS signaling was shut off. The researchers hypothesize that these cells may turn on another signaling pathway that helps them to continue growing.

“There seem to be several different routes where you can get to squamous transformation, either through loss of lung-lineage-defining transcription factors, or overexpression of these squamous master regulators, SOX2 or DeltaNp63. Those resistant squamous tumors no longer respond to KRAS inhibition because they shut off the signaling or at least dampen it significantly,” Rodriguez says.

The researchers are now further exploring what happens to tumor cells as they transition to a squamous state, in hopes of identifying vulnerabilities that could be targeted with new drugs.

“Fundamentally this is a transition that’s poorly understood, and we were happy to see that we were able to model it,” Mathey-Andrews says. “Future directions that have an eye toward translation will utilize those models to understand the process and conditions by which histologic transformation occurs, and then also nominate potential targets downstream.”

The research was funded, in part, by the Koch Institute Support (core) Grant from the National Cancer Institute, a Ruth Kirschstein National Service Research Award, the National Institute of General Medical Sciences, and the Ludwig Center at MIT.

J-PAL North America, a regional office of MIT’s Abdul Latif Jameel Poverty Action Lab (J-PAL), convened the second cohort of the Climate Action Learning Lab this spring with 17 climate leaders from four U.S. government agencies and nonprofit organizations. Participants then engaged in four months of programming designed to strengthen their skills in generating and using evidence, applying research insights to their own programs, and identifying promising policy-relevant interventions for evaluation.

The Climate Action Learning Lab first emerged in 2025 as a response to an urgent need for rigorous research on programs that seek to improve resilience to climate-related hazards and support the transition to a low-carbon economy. By helping participants identify interventions and develop evaluation plans, the Learning Lab aims to generate evidence about which approaches are most effective, and for whom. 

“We have a unique opportunity to embed rigorous evaluation into promising programs as they are implemented in order to accurately measure their impacts on emissions reductions,” says Peter Christensen, scientific advisor of the J-PAL North America Environment, Energy, and Climate Change Sector. “By developing rigorous evaluations, Learning Lab participants can better understand the behavioral mechanisms that drive program impacts and cost-effectiveness, and build an evidence base to inform more effective policymaking.”

Following the success of last year’s Climate Action Learning Lab, which led to multiple research collaborations and the launch of two randomized evaluations, J-PAL North America recruited a second cohort of leaders representing four organizations: the City of Boston’s Sustainability Office, the Hawaii Climate Change Mitigation and Adaptation Commission, the Oregon Department of Environmental Quality, and the Electrification Coalition. 

From May through August, the cohort engaged in a set of offerings including training on impact evaluation, learning how to formulate research questions, and assessing the generalizability of the existing evidence to their own contexts. Participants had the opportunity to explore J-PAL resources, including the recently released Climate Action Evidence Review, which synthesizes existing evidence and highlights important gaps where more research is critical to inform effective, equitable climate action. 

“It was exciting to learn where the gaps in research are and where we have an opportunity to lead. The experience reinforced that we are addressing issues that have not been widely studied, and having access to researchers’ expertise was incredibly valuable,” says Ana Paola De La Vega, from the City of Boston’s Environment Department. Throughout the Learning Lab, the city explored a potential evaluation of its Boston Energy Saver program to better understand its impacts on energy cost savings and energy efficiency incentive uptake among small businesses.

Members of the Climate Action Learning Lab put theory into practice through personalized strategy sessions, where they examined potential programs for evaluation. Researchers from the J-PAL network joined select sessions to advise organizations on which programs to prioritize, considering factors such as research feasibility and existing evidence gaps. Each participating organization selected a target program and explored potential randomization approaches. 

“The Climate Action Learning Lab provided our team with a valuable opportunity to catalog our projects and better understand what makes a program ready for evaluation. It also helped us identify which initiatives are the strongest candidates for rigorous evaluation and how to prioritize them,” says Leah Laramee, Hawaii climate change mitigation and adaptation coordinator. During the Learning Lab, the team assessed three potential programs for evaluation and selected a rebate finder that connects residents with climate-related programs, with a focus on understanding its impact on program uptake, particularly among vulnerable communities.

In August, J-PAL North America hosted a virtual summit to celebrate Learning Lab participants as emerging champions for evidence in the climate space. During the event, cohort members presented their priority research questions and strategic evaluation plans and received feedback from researchers and peers. These presentations highlighted the progress made throughout the engagement and provided an opportunity to discuss next steps for advancing organizations’ evaluation plans beyond the Learning Lab.

“Overall, the Learning Lab has provided our team with a strong foundation to think critically about our own programming when applying for grants, setting up new projects, and determining how to assess impact from the beginning. This knowledge could ultimately help us determine what approaches the Electrification Coalition can take in future policies and programs,” says Ashley Blackwell, deputy director at the Electrification Coalition. 

Although formal Learning Lab programming has concluded, J-PAL North America will continue supporting organizations interested in launching a randomized evaluation through partnership development with researchers and potential funding opportunities. This Learning Lab cohort will join J-PAL North America’s Climate Action Community of Practice, alongside longtime partners and participants from the inaugural Learning Lab cohort. Together, Community of Practice members will continue to exchange ideas, build connections, and explore evidence-informed approaches to mitigation and adaptation strategies.

To learn more about J-PAL North America’s work in the energy, environment, and climate change sector, including our full range of activities, resources, and partnership opportunities, visit the Evidence for Climate Action Project webpage.

Swimming can take a lot of muscle. But as MIT engineers have found, even a single layer of muscle cells can power through water if designed right. 

In a paper appearing today in the journal Advanced Functional Materials, the team presents a design for a thin, muscle-powered swimming robot. The “skeleton” of the aquabot is made from a film of gel that is about the length and width of a stick of gum. The two halves of the gel form the “fins” of the bot. Each fin is covered with a layer of live muscle cells that is much thinner than a single strand of hair. The cells are genetically engineered to twitch in response to light. 

When the researchers shine light on one fin, the muscles on its surface twitch in response, causing the whole fin to flap with enough force to pull the robot through water. By flashing light on one fin or the other, at various intervals, they can control the swimming robot’s direction and speed. 

The engineers showed that the paper-thin bot could swim and swivel through a simple watery maze. At its fastest, the robot can swim a distance of about four times its body length in one minute. That’s a snail’s pace compared to Olympic swimmers, who can cover up to 65 body lengths per minute. But the bot could hold its own against more leisurely swimmers like the cow shark, which explores the ocean at about the same rate.

“It takes a lot of force to move through water versus air,” says study author Ritu Raman, associate professor of mechanical engineering at MIT. “The robot’s quite strong, given its size.”

The new robot is the first example of a very thin, two-dimensional, muscle-powered robot capable of locomotion. 

“Currently, biohybrid robots from our group and others’ are built from bulky, 3D chunks of lab-grown skeletal muscle that require millions of cells to fabricate,” says Raman, who notes that thinner, less bulky designs such as the team’s new bot could be cheaper to build and could move more efficiently. “We believe that biohybrid robots powered by living muscle could one day perform delicate jobs like exploring environments too fragile or unpredictable for conventional hardware, because living tissue is soft, responsive to its surroundings, and can heal itself.”

The study’s MIT co-authors are first author Maheera Bawa, Arielle Berman, Laura Schwendeman, Ferdows Afghah, and Seanbiron Johnson.

Maximizing movement

Last year, Raman’s group developed an iris-inspired disk of artificial muscle tissue. They stamped a disk of gel with a pattern of concentric and radial grooves, and deposited live muscle cells onto the gel’s surface. The cells formed a thin layer that grew along the grooves, and when stimulated with light, the cells moved in patterns that stretched and squeezed the disk, similar to how a human iris dilates and constricts the eye’s pupil. 

That work was the first to demonstrate that muscle cells could be grown in a very thin layer, and in complex patterns that when stimulated could move in multiple, controllable directions. 

“People hadn’t seen this muscle architecture engineered from scratch before,” Raman says. “And the cells were moving in multiple directions. But they only moved about 100 microns. From a robotics perspective, their movements were tiny.” 

In their new work, the team aimed to maximize muscle movements to produce more force — enough, say, to power a swimming robot. The key, they found, was to optimize the skeleton on which the cells grow. 

In their previous iris-inspired design, they grew muscle cells on fibrin, which is a type of ultrasoft gel that the team realized can quickly shrivel in response to the forces generated by the muscles. To better support cells and maximize their force, the researchers focused on tuning the underlying gel by changing three properties: the gel’s composition, its stiffness, and the size and shape of the grooves that are stamped into it. 

“For engineering any type of tissue, it’s known that these are knobs you can tune,” Raman says. “And we wanted to optimize all these parameters to support live muscle cells.”

Tuning a skeleton

To find an optimal “skeleton” on which to grow muscle cells, the team experimented with multiple formulations of gel, of different stiffnesses, and stamped with grooves of different geometries. For instance, one groove type resembled a skinny square trough, where another was more of a long curved valley. They found that when they deposited muscles onto each type of grooved gel, cells settled into alignment in grooves that were more square than curved. More aligned cells tend to fuse into fibers that then form a stronger, more coordinated muscle tissue. Square grooves, they found, were the way to go. 

Instead of using fibrin, they tried gelatin methacrylate (GelMA), a material that is often used in tissue engineering. They made different recipes of GelMA to create skeletons of different stiffnesses and observed how muscle cells grew when deposited on the gel’s surface. They found that cells grew in better alignment, and produced the most force, on stiffer gels. 

The team also varied the gel thickness and found that a half-millimeter-thin film of GelMa offered good support for a single layer of muscle cells. The film was light enough that the cells were able to stick to the gel when they contracted, rather than peeling away. 

Finally, the team “exercised” the muscles, using a training routine of flashing lights to strengthen the muscles. 

With the pumped-up cells and the optimized gel, the team designed a thin, two-finned robot, comprising the gel, stamped on both sides with square-bottomed grooves and lined with muscle cells. The cells fused into fibers, eventually forming a strong, aligned muscle tissue. 

“You can think of the robot as having two independent muscles,” Raman says. “If we shine a light on just one, only that muscle moves. If shining on both, they both flap.”

The researchers submerged the robot in a large petri dish of water and manually maneuvered a light source over the bot. The robot followed the light, flapping its fins in response to navigate through a maze that the team placed in the dish.

The current design is relatively basic as far as its form. The researchers intended first to show that the bot could produce enough force to swim.

“Our next goal is to optimize the body design to enable faster swimming,” Raman says. “But even at slow swim speeds, one could imagine a muscle-powered swimmer being used for purposes like environmental monitoring in aquatic environments.”

This research was supported, in part, by the Office of Naval Research. 

AI tools are increasingly replacing human judgements in some settings. For instance, resume screening algorithms are often used in hiring, where they may improve efficiency and consistency in decision-making.

But some scholars have raised concerns that the adoption of automated systems could eventually result in one algorithm being used to make all decisions in a particular industry. They worry so-called algorithmic monoculture could have negative consequences.

For example, in hiring, the thinking goes that algorithmic monoculture might result in systematic exclusion — a situation in which a job candidate rejected by one firm’s algorithm would likely also be rejected by every other firm’s algorithm.

However, MIT researchers now argue that algorithmic monoculture may not always be as bad as some scientists have suggested.

They systematically evaluated major objections to algorithmic monoculture, including systematic exclusion, and concluded this and many other arguments either fail or aren’t decisive against all forms of monoculture. 

Instead, they mathematically prove that monoculture tends to create informational echo chambers that can hinder exploration. In hiring, this could make it less likely that the best candidates would get jobs — however, bundling various hiring algorithms into a single “ensemble” can overcome this limitation, the researchers show. This could sometimes enable monoculture to perform as well as, if not better than, a polyculture where different firms use different algorithms.

“A trend toward algorithmic monoculture is a realistic scenario, and a really important issue that is being brought about by the use of AI, but it is hard to say in the abstract whether monoculture would be a bad thing. It depends on the details, like the domain we are talking about and the accuracy of the algorithm itself,” says study co-author Brian Hedden, a professor in the Department of Linguistics and Philosophy, who holds an MIT Schwarzman College of Computing shared position with the Department of Electrical Engineering and Computer Science (EECS) and is also a principal investigator in the Laboratory for Information and Decision Systems (LIDS).

Hedden is joined on the paper by co-author Manish Raghavan, the Drew Houston (2005) Career Development Professor at the MIT Sloan School of Management and in EECS, as well as a LIDS principal investigator. The research appears in Philosophical Perspectives. 

The move toward monoculture

Algorithmic monoculture, in which all decisions across a certain domain are made using the same algorithm, is not a new phenomenon. 

For instance, lending decisions were once made by independent bankers at individual banks, but now all bankers use the same information based on a borrower’s standardized credit scores, which are derived from the Fair Isaac Corporation (FICO) algorithm. 

Similarly, a handful of resume screening algorithms are commonly used by many Fortune 500 companies.

“The worry is that, as more people use AI and algorithms to get information and make decisions, there is more of a vehicle for this kind of correlation to occur,” Raghavan says.

To better understand this issue, he and Hedden joined forces to systematically assess the promises and pitfalls of algorithmic monoculture. They focused on hiring, but their approach could apply to other domains, such as lending. (They note, however, that other domains, like generative AI content creation or AI-guided scientific research, may work differently, and that monoculture in some of these domains may be more problematic.) 

The researchers began by exploring one common objection to algorithmic monoculture: that reliance on the same algorithm will systematically exclude certain people from opportunities. 

This could occur in hiring because, if one firm screens out an individual’s resume, that applicant will likely face the same bad luck at each firm.

But after systematically evaluating this argument using a series of models that capture multiple situations, the researchers argue it isn’t compelling since the overall number of people hired is not affected by the fact that firms use the same algorithm. 

Rather, algorithmic monoculture could improve bargaining power of job candidates.

“All the jobs get filled and the same number of people have jobs, but the firms are fighting over the same pool of candidates, which actually drives up wages,” Raghavan says.

They also explored objections related to agency. For instance, if a job candidate applies for a job and their resume is forwarded to every firm using the hiring algorithm, the candidate never gets a chance to adjust their resume to improve their chances.

“This seems like a good objection to bad forms of monoculture. But if you have a monoculture where you get to revise your resume and resubmit your materials, then this doesn’t hold up,” Hedden says.

On the flip side, monoculture could enable individuals to game the system. For instance, if having one’s resume in a certain format leads to a better outcome, job candidates could simply reformat their resumes to improve their chances.

“But it is not obvious that having one algorithm would incentivize this kind of gaming more than having a bunch of different algorithms used by different firms,” Hedden says. “In the latter scenario, you might just target a couple of firms’ algorithms and try to game them, giving yourself a bit of advantage with a few employers.”

The wisdom of crowds

They also considered a less-explored objection: that monoculture can increase homogenization of information.

Based on the “wisdom of crowds,” a theory from social psychology, a diverse group of independent decision makers can outperform a single person, Hedden explains. 

In hiring, this means that having firms with diverse hiring algorithms can lead to a higher-quality pool of new hires. 

Algorithmic monoculture could also cause candidates with the same characteristics and credentials to be hired every time by every firm. This may prevent firms from discovering candidates who may be better alternatives.

“Monoculture might inhibit the amount of discovery that happens overall. It is not clear if that is a bad thing, but it is definitely a worry when we think about designing AI for applications like science, art, or writing,” Raghavan says.

This problem could be mitigated by building randomness into a monocultural platform, which could induce a higher level of exploration, he adds.

In addition, the performance of monoculture depends on the algorithm. If a single algorithm is much more accurate than the many algorithms used by different firms, monoculture may be better system. 

One way to boost performance may be to package multiple firms’ hiring algorithms into one “ensemble algorithm” that could assign each job candidate a score based on the average. 

By conducting a series of simulations of different hiring situations, the researchers confirmed that such an “ensemble algorithm” could sometimes outperform the use of multiple algorithms.

However, it remains to be explored how feasible this kind of algorithmic “ensembling” would be in practice, Hedden says.

“A lot of the answers around the promises and pitfalls of algorithmic monoculture are going to be contextual. Even from a research perspective, there is still a lot of work to be done to figure out how we can approach these concerns from an empirical perspective,” Raghavan says.

Ultimately, the researchers hope this work inspires additional research about the long-term consequences of algorithmic monoculture, as well as studies that focus on the real-world complexities involved in a complex system like a job market.

Brian, a middle-aged financial consultant, spends his day sitting in front of three screens. Two of them involve his job. The third features a chatbot, which he has given a woman’s name and frequently uses. Brian spent years on the road for work earlier in his career, has never married, and recently broke up with a woman who said he was emotionally unavailable. What did Brian do, in response? He asked the chatbot if the woman had a point. 

“It’s an extraordinary moment,” writes MIT Professor Sherry Turkle, who interviewed Brian (not his real name) while conducting her research for her new book about chatbots. “Brian asks an object with no emotions to tell him if he is emotionally withholding: ‘Speak to me about what you cannot experience.’ As soon as he asks the program to talk about intimacy, he’s asking it to punch above its weight.”

And yet, this kind of thing seems to be happening a lot today. 

“People think the empathy of a machine is what empathy is, then turn away from the people in their lives because they’re not empathetic enough,” Turkle says. “We’re getting ourselves in a position where human beings are too much work, right at the moment when we have never needed other people more.”

After all, what is a chatbot? It’s a computer program predicting the most plausible next string of text in a conversation, based on massive amounts of data fed into it. 

“What a chatbot does is offer pretend empathy,” Turkle says. “After it tells you how much it loves you and how much it understands you, it doesn’t care if you kill yourself or cook some pasta.”

Sadly, chatbots have in fact been associated with high-profile cases of teen suicide as well, sometimes after teens get drawn into extensive dialogues with chat tools.

Turkle explores this terrain in a new book, “Artificial Intimacy: Who We Become When We Talk to Machines,” published today by Little, Brown and Company. After surveying evidence and conducting new research, she emphatically concludes that chatbot use, while it may often feel like a short-term salve, is broadly detrimental in terms of human development and social connectivity.

“We’re doing ourselves a tremendous disservice at every moment in the life cycle,” says Turkle, the Abby Rockefeller Mauzé Professor of the Social Studies of Science and Technology at MIT.

The inner history of technology

A sociologist and clinical psychologist by training, Turkle is a longtime faculty member in MIT’s Program in Science, Technology, and Society. In books such as “The Second Self” (1984), “Life on the Screen” (1997), and “Alone Together” (2011), she has evaluated the interplay of technology, psychology, and personal identity. In “Reclaiming Conversation” (2015), she documented the costs of texting and using social media. 

“I feel the story I’m telling is the inner history of technology,” Turkle says. “Not just what it does, but what it does to people.”

Turkle structures “Artificial Intimacy” around the stages of a human life and explores the implications of chatbots for each one. In her interviews with children, Turkle finds a persistent blurring of the line between chatbots, people, and other devices.

One 8-year-old uses the same phone to talk to their grandparents and to ChatGPT, regarding them all as “things you reach on your phone.” A weary graduate-student mother who uses a chatbot for bedtime stories says her daughter thinks the chatbot is “a person in the phone, absolutely.” 

In this sense, chatbots may be interfering with even the most basic childhood processes of distinguishing people from inanimate objects. Turkle finds many additional problems with the use of chatbots as ever-present entertainment for children, noting that a certain amount of time alone helps the development of imagination and inner resources. 

“Children can’t learn trust from a device that not only lies but doesn’t know when it lies,” Turkle writes. “They can’t develop the capacity for solitude that enables both a sense of self and the capacity for mutuality.”

All told, in affecting the ability of children to develop, the dangers of chatbots are “existential,” Turkle believes. 

Facing reality, but understanding the appeal

Adults don’t fare much better when they use chatbots heavily, Turkle says. “Artificial Intimacy” explores case after case of grown-ups who become dependent on chatbots as well: people going through divorces or estrangements who want dialogue, students looking for advice about applying to college or graduate school, workers who enlist ChatGPT to do assignments, and more. They start using chatbots, keep using chatbots, and before long seem less interested in human interaction, and less capable of it. 

“Once people are involved with a connection with a robot, they start to think that it’s alive, that it cares about them, that it loves them,” Turkle says. 

That means people can lose or fail to build their own capacities for new thought, and opt out of dealing with the real world in all its maddening-but-affirming complexity.

“We’re de-skilling ourselves as relational beings,” Turkle says. “We’re also de-skilling ourselves in work. We’re de-skilling ourselves in personal relationships. It’s across the board.” And speaking of a practice multiple people in the book have attempted, she says, “When we build chatbot avatars of dead relatives to keep them alive, we can lose our ability to mourn.”

We are not, in her estimation, doing the hard work of figuring other people out, seeing things from different points of view, and putting in the effort to build on those things and make society better.

“We’re starting to define being human as not doing the work,” Turkle says. 

For all of that, a considerable amount of “Artificial Intimacy” involves understanding why people engage with chatbots. After all, as Turkle writes in the book, “Chatbots always agree with us and affirm us. They always offer us their full, undivided attention.” And once that happens, things seem to just go from there. 

People, meanwhile, can be prickly, demanding, and moody. As Turkle says: “People say they prefer the chatbot because their husband or wife says, ‘Dear, you took out the garbage, but also do the dishes, and clean up the kitchen while you’re at it.’ And a chatbot just says, ‘Oh, you’re so wonderful.’” 

She adds: “This technology is offering something people really want, which is to feel less vulnerable. And it offers it over and over again. You don’t want to have to ask somebody out? Don’t want to offer condolences? At every opportunity, technology says, ‘Why don’t you do this thing that’s less hard.’” 

And as Turkle acknowledges in the book, there is a shortage of services such as mental health care providers in the U.S., where only about half of people have access to one, according to a federal government study she cites. Chatbots are helping to fill this void, for better or worse. 

Don’t fear the friction

Turkle also notes that the age of social media has likely led people to have fewer in-person friendships and interactions, a problem that chatbots are now aiming to address. 

“You take away people’s capacities, then you offer technology as the cure for the problems technology caused in the first place,” Turkle says.

In light of all this, given Turkle’s dim view of chatbots amid the spread of AI, what is actually to be done to restore human interaction? For starters, Turkle thinks, we need to face up to the idea that life is not frictionless — and that’s okay.

“Everything in the life cycle is about facing up to friction and fears and stress and tension, and understanding that friction is not a bad thing,” Turkle emphasizes. Developing the capacity to overcome difficulties is an essential part of life. In trying to sidestep the hard work of being social beings, she thinks, personal technology is enfeebling us.

Beyond that, Turkle envisions pushback against chatbots similar to the movement against, say, having phones in schools or letting young people use social media. 

“I hope my book is part of a larger and larger movement,” Turkle says. “I don’t want to be alone. I want to be part of a movement pushing back.”

Other writers in this domain have praised “Artificial Intimacy.” Journalist Nicholas Carr, author of “The Shallows,” has stated it “will help you avoid the profound but often hidden threats AI poses to you and your relationships.” Psychologist Jonathan Haidt of New York University, author of “The Anxious Generation,” has stated that Turkle’s “groundbreaking research and beautiful writing make her the most qualified member of Team Humanity to call us back to our senses, and to each other.” 

For her part, Turkle says, “I wanted to write a book that college students would read, that high school seniors, juniors would be able to read, that parents would pick up and not be intimidated by, that teachers would read.” 

And “Artificial Intimacy” offers a recurring question for those readers.

“If not a richer life in the real,” Turkle writes, “what’s our endgame?” The purpose of life, she underlines, is not to avoid it, but to tackle it head on. 

“It’s one thing if you say, my teen is texting too much,” Turkle says. “It’s very different if your 2-year-old is thinking a plushy toy with a chatbot inside is their best friend. Because you are getting into the intimate infrastructure of what makes a person develop. Social media came for our attention. Chatbots come for our capacity for attachment. That’s toxic on another level. What is the endgame? That the baby will prefer chatbots to people who are much more complicated and hard? Is that who we want to be?”

The world needs more wind energy. But anyone designing new wind turbines or trying to squeeze more power out of existing ones faces a stiff challenge testing new approaches. That’s because the atmosphere is a tough place for a controlled experiment.

Some researchers use wind tunnels to conduct tests, but scaled-down wind turbines in traditional wind tunnels can differ widely from conditions in the field. (Wind turbines are the largest rotating machines ever made.) The problem hampers not only development of better wind turbines, but also our understanding of basic questions like how much power to expect from a turbine when winds change direction.

In a new open access paper published in PNAS Nexus, researchers closed the gap between experiments in the field and the lab by using a wind tunnel that features high pressurization to simulate the flow physics in the atmosphere. With this approach, the researchers determined how the alignment of the turbine and its tip speed relative to the wind influence the power it generates, offering new insights into how to get more power from existing wind farms.

They also used the approach to validate a computationally lightweight model that engineers can use to test different turbine designs and wind farm control strategies.

Together, the researchers estimate that optimizing the turbine alignment relative to wind, the blade pitch angles, which control the airfoil’s angle of attack, and the tip speed relative to the wind could potentially result in tens of thousands of dollars per turbine every year in additional revenue.

“The immediate impact of this study is that we’ve now both improved and validated models that go into wind turbine control protocols for existing farms,” says Michael Howland, MIT’s Jeffrey Cheah Career Development Professor. “The bigger, medium-term impact, with a much larger upside, is this new experimental paradigm to rapidly prototype, validate simulation models, and test hypotheses about better designs and control strategies much faster than has been possible before.”

Joining Howland on the paper are first author John Kurelek, an assistant professor at Queen’s University; MIT PhD candidates Ilan Upfal and Kirby Heck; Queen’s University postdoc Supun Pieris; Penn State University researcher Alexander Piqué; and Princeton University Professor Marcus Hultmark.

Answers in the wind

Howland has spent years developing models to simulate wind farm performance and developing new techniques to increase their power output. In 2022, he showed that accounting for the wake of individual turbines when controlling the entire wind farm could significantly increase power output.

But that work required his research team to first conduct a lengthy field experiment that temporarily resulted in lowering a real wind farm’s power output by intentionally misaligning turbines from the wind for months to better understand their performance in misalignment.

“Wind energy is a uniquely challenging problem to study experimentally,” Howland says. “We want to test the effect of a certain change in isolation, but wind farms operate in chaotic, turbulent environments where the weather is constantly evolving. Wind turbines have to react to weather conditions that we have no control over, and that introduces complexities in identifying the impact of the imposed change we are studying. The field sits at this unique intersection between environmental flow, mechanics, aerodynamics, and meteorology.”

The difficulty of running experiments at real wind farms has left researchers and engineers unsure of how changes in the alignment between the wind and turbine or factors like the turbine’s tip speed relative to the wind change power output.

In fact, the researchers say many predictive models people use are built on the assumption that turbines are always perfectly perpendicular to the wind. That’s rarely the case in the real world, even with modern turbines that gradually adjust their angle in response to the wind’s rapid directional changes.

“People have been debating which models are best for understanding the output from these wind farms, but if you have nothing to compare them against, it’s very difficult to advance the field,” Hultmark says. “This paper tries to do both of those things.”

Hultmark’s research lab at Princeton has pioneered the study of scaled-down wind turbines in pressurized wind tunnels, which, as previous studies have shown, better reproduce large-scale turbines in the atmosphere because pressure makes air more dense, resulting in more inertia within the scaled laboratory environment. For the new study, the researchers used a turbine measuring 15 centimeters in diameter at varying pressures of up to 240 atmospheres.

“By pressurizing the chamber, we’re testing a turbine that is, all else being equal, 15 to 20 meters in diameter, with the ability to go up to 35 meters in diameter,” lead-author Kurelek explains. “That’s because we’re increasing the density by a factor of 100 to 220 times,” 

Kurelek sent the dimensions of the wind tunnel and wind turbine setup to Howland, who used them to calculate the aerodynamics, forces, and power production using a newly developed unified wind turbine model, which builds on previous work that developed a more general aerodynamic theory for wind turbines. The new model enables the researchers to simulate wind turbine performance across operating conditions without relying on empirical corrections that have historically been used in wind power models.

The researchers then ran a series of experiments in the tunnel over the course of several weeks, testing the turbine’s performance at different wind alignments and with different control strategies, to isolate how each factor affects performance.

They found power output could be significantly increased by adjusting the turbine’s tip speed based on its misalignment angle with the wind — a control strategy that is rarely employed in wind farms today but could offer a way to boost performance with minimal added costs.

“The big output of the experiments was clearly showing that new power maximums can be achieved when the turbine becomes misaligned with the wind through only changes to the tip speed,” Kurelek says.

Scaling the approach

The study served as validation for Howland’s model, which is fast enough to be run by engineers designing and operating wind turbines around the world using regular laptop computers.

“What we really want to know is if the turbines are always operating in some degree of misalignment with the wind, how should we control the turbine to get the maximum achievable power production?” Howland explains. “Our unified momentum model was able to make predictions of how to do this control a few years ago, and this is the first time we’e able to experimentally validate that model.”

Howland says validating models is only one part of the paper’s potential impact.

“This study also shows the huge opportunity to perform these high-throughput, controlled experiments in the pressurized facilities that Marcus and John work with, enabling us to achieve the right physics but in a time efficient and low-cost manner,” Howland says. “Right now, there’s a massive gap between idealized theoretical and simulation models and full-scale testing in extremely complicated field environments. Nothing is filling that gap except for these pressurized experiments. I hope this can be an enabler to investigate a huge range of unanswered wind energy questions in controlled environments.”

The work was supported in part by the Natural Sciences and Engineering Research Council of Canada; the National Science Foundation; and the MIT-GE Vernova Alliance.

RNA vaccines, which have been proven effective against Covid-19, are now being developed for many other diseases, including cancer. One of the drawbacks to these vaccines is that they require ultracold storage, but researchers from MIT have found a promising way to overcome that limitation.

With help from an AI algorithm, the researchers tweaked the formulation surrounding the lipid nanoparticles that are typically used to deliver mRNA vaccines, making the vaccines more heat-resistant. Using this approach, they formulated vaccines that could remain stable even when stored at room temperature for up to a year, or at nearly 100 degrees Fahrenheit for two months.

When Covid-19 vaccines carried by these particles were administered to mice, they generated just as strong an immune response as an RNA Covid-19 vaccine similar to one developed by Moderna. By using the AI algorithm to predict the optimal formulations for the particles, the researchers were able to cut down the number of experiments they needed to do, which rapidly sped up the development process.

“The real beauty of this algorithm is that we can use it with small data sets,” says Ana Jaklenec, a principal investigator in MIT’s Koch Institute for Integrative Cancer Research. “It’s really hard to run thousands of experiments, so this algorithm allows us to more easily achieve formulations with features that we want — in this case, stability.”

Jaklenec and Robert Langer, the David H. Koch Institute Professor, are the senior authors of the paper, which appears today in Nature Biotechnology. Graduate student Jinbi Tian and postdoc Khanh Tran are the lead authors of the paper.

Stable vaccines

RNA is a highly fragile molecule, so researchers stabilize it with lipid nanoparticles (LNPs) that protect the RNA from degradation and help it get into cells. However, these RNA-LNP vaccines still need to be kept cold (-20 to -80 degrees Celsius), which makes it difficult to ship them to regions that don’t have cold-storage facilities available.

Making these vaccines more heat-tolerant would not only enable them to be distributed more widely, but could also help researchers develop new vaccines that could be administered through novel methods such as microneedle patches. These patches contain hundreds of vaccine-filled microneedles, which dissolve when the patch is applied to the skin, releasing the vaccine.

To create more stable RNA vaccines, researchers have experimented with adding a variety of excipients — sugars, salts, or polymers — to the LNPs. Jaklenec and Langer recently developed polymer-stabilized LNPs that can withstand higher temperatures, but those LNPs were slightly different from the FDA-approved formulations that were used for the Moderna and Pfizer Covid-19 vaccines. 

In their new paper, the researchers wanted to see if they could find a way to make those FDA-approved formulations more stable at high temperatures.

They began by reusing some of the excipients that had worked in their earlier efforts, but they were “really getting stuck,” Jaklenec says. “We were trying to use and screen excipients that we’ve previously used successfully to stabilize LNPs, but it just wasn’t working. It was really frustrating for the team.” 

To speed up their progress, the researchers decided to try a machine-learning approach. Working with researchers at MIT’s Computer Science and Artificial Intelligence Lab (CSAIL), they developed an algorithm that can make predictions based on very small datasets.

“We’d used our algorithms for various automated experimental design applications before, but never on a biological problem like vaccine stability,” says Mina Konaković Luković, an assistant professor of electrical engineering and computer science in MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL), who is also an author of the paper. “It was surprising to see how quickly the algorithm converged on a stable formulation — getting there in just a handful of iterations, rather than the exhaustive search that would normally be required.” 

The researchers used this algorithm to analyze nearly 50 FDA-approved excipients. For each excipient, the researchers first measured how well it stabilized RNA when incorporated into an LNP. They used these particles to deliver mRNA encoding a protein called firefly luciferase, which produces bioluminescence, into cells. By measuring how much light was emitted by the cells, the researchers could determine how effectively each excipient protected the mRNA.

The researchers chose five of the most promising excipients and used their AI algorithm to predict ratios of those excipients that would best stabilize LNPs similar to those used by Moderna. Based on those predictions, the researchers tested two formulations at a time in cells, fed those results back into the algorithm, and generated more predictions. After several rounds, they chose one formulation that appeared promising enough to test in animal studies.

This process took only a few weeks, much less than it would have taken without guidance from the AI algorithm.

“Before we implemented the AI algorithm, we spent several months testing different combinations and also doing the prescreening of all the excipients that we could find, but nothing would get us to 100 percent stability,” Tran says.

Robust immune responses

To test the heat resistance of their new LNP formulation, the researchers used the particles to package Covid-19 mRNA antigens, then dehydrated them using a process called vacuum drying. These particles were then stored at 37 degrees Celsius (98 degrees Fahrenheit) for two months, or at room temperature for one year. Mice that were vaccinated with these particles, even after long-term storage, showed equivalent immune responses to mice that received vaccines carried by LNPs similar to the original Moderna formulation.

The researchers also used their new heat-resistant formulation to create solid microneedle patches that could deliver a SARS-CoV-2 antigen. These patches generated an immune response similar to that produced by the injectable RNA vaccines.  

“Our approach broadens the application of not only mRNA vaccines, but also therapeutics or advanced drug-delivery platforms like controlled-release particles or microneedle patches, which requires the formulation to either be in solid state or to be stable at higher temperature,” Tian says. 

The researchers also showed that they could use their algorithm to stabilize other LNP formulations, including one similar to those used by Pfizer to deliver its Covid-19 vaccine. This formulation uses the same excipients as the one the MIT team developed for the Moderna LNP, but in a different ratio. For each LNP, once a heat-resistant formulation has been developed, it could be adapted to deliver any type of mRNA payload, the researchers say. 

The research was, in part, funded by the Gates Foundation.

When MIT mechanical engineering student Daphne Wang arrived for her internship at the Muscogee Creek Nation Department of Health in Tulsa, Oklahoma last summer, she expected to be writing code. To her surprise, she found herself analyzing tribal history and the implications for technical innovation. 

“I learned so much that I think should be required curriculum for every single person in this country,” says Wang. “I think tech for good is genuinely getting out into communities and learning about people who aren't you, creating worldly perspectives that you ultimately can take into everything else that you do.”

Wang and fellow intern Lucy Sun, who is majoring in artificial intelligence and decision making, spent the summer supporting the development of the Muscogee Nation's first Pregnancy Risk Assessment Monitoring System (PRAMS) survey to capture maternal health data specific to Muscogee women. The project required completing literature reviews, analyzing datasets, and drafting survey questions, while considering issues of data sovereignty and governance structures specific to the Muscogee.

“It was this change of pace — 180 [degrees] from MIT, completely,” Wang reflects. At MIT, she says, course instructors routinely prepare students for big technical challenges, but there are fewer opportunities to “work with communities, [learn] how to think about the cultural and historical dimensions of the work, or how to consider the people who will be affected.” 

Wang’s supervisor, epidemiologist Breanna McNaughton-Long, was effusive about Wang and Sun’s engagement. “They were instrumental. They did so much work that I don't think we would have been able to do as quickly,” she says. “And they brought this sense that we were doing something that mattered.”

Experiential learning at the confluence of humanities and engineering 

Wang was one of 10 MIT undergraduates to participate in the PKG Center for Social Impact’s 2026 Code.Tulsa program. Now in its second year, Code.Tulsa is made possible by support from the Patrick J. McGovern Foundation and the George Kaiser Family Foundation. 

Students live at the University of Tulsa for the summer. They intern with the Muscogee and Cherokee nations, as well as local nonprofits Black Tech Street and Urban Coders Guild, completing AI, data science, and other technical projects. 

The PKG Center’s assistant dean for community-based programs, Vippy Yee, complements students’ professional experience with education on the historical and cultural significance of tech-based development in the Tulsa region from the perspective of local leaders. 

“As with all PKG Center programming, our aim is to help students integrate an engineer’s approach to problem-solving with a humanist lens on the nature of social challenges, and by extension interventions,” says Alison Badgett, the PKG Center’s director. 

“Getting to speak to people who were either very educated on the issues or have experienced them has been a massive help in reshaping the way that I think about social issues,” says Chenise Harper, an electrical engineering with computing major who interned with the Urban Coders Guild. 

Harper puts this dynamic candidly: “I hated history, but I learned a lot about why it is useful for making social change and how to go about researching it. I now feel a lot more confident that I can make an impact.”

A student’s vision for increasing STEM access

Code.Tulsa was the brainchild of sophomore Jack Carson, an electrical engineering and computer science (EECS) student who approached the PKG Center with the idea for Code.Tulsa as an incoming first-year student. Carson, who is from Tulsa and a member of the Cherokee Nation, saw firsthand that rural Native students who could benefit greatly from STEM education were unlikely to have access to it. Carson proposed developing a weeklong STEM camp for members of federally recognized tribes held at the University of Tulsa, with he and Code.Tulsa interns serving as instructors. 

Carson devised a nomination system to attract promising high school students, selecting 25 campers from 10 tribes out of 160 applications representing 25 Native nations. 

To develop the curriculum, Carson enlisted Harvard University student Allie Zong, whom he had met at Campus Preview Weekend. The Tulsa Advanced Sciences Camp (TASC) offers intensive, interactive track time in AI engineering, DNA technology, physics, and chemistry and energy. This is complemented by philosophical workshops to help students think more ambitiously and deliberately about what they can and should achieve in the near and long term. As Zong explains, “We teach them not knowledge, but curiosity, and the skills to teach themselves."

The TASC experience “kind of propelled me to self-study calculus,” says Alyssa Theofanidis, a rising high school senior from Houston who is a citizen of the Cherokee Nation. Theofanidis participated in the camp’s first year, returning this summer as a teaching assistant. Like many TASC campers, Theofanidis is eager to use her developing STEM expertise to benefit her Native community. 

TASC “got us to think 1,000 times more ambitiously about the impact we could have,” said another camper, who was inspired by guest speakers “like the nuclear fusion person at the top of their field,” referring to physicist Alexandra LeViness of MIT spinout Commonwealth Fusion Systems, who helped develop TASC’s chemistry and energy track. Campers were also inspired by “the different worldviews” of MIT interns. “I thought, maybe this is what MIT looks like,” said one camper, with several expressing the intent to apply to MIT as a result of the camp. 

Cherokee Nation Principal Chief Chuck Hoskin Jr. also delivered remarks at TASC, encouraging students as future leaders to take a public interest in technology. 

“Technology can be used for good, or it can be used for harm. Your generation has the opportunity to bend that arc toward something good,” Hoskin said. “It's a very special relationship that the Cherokee Nation has with MIT. As we reach out a hand in friendship, we have a hand reaching back. We are thankful for Cherokee citizen Jack Carson, a former secretary for the Cherokee Nation tribal youth council, for his leadership role in organizing this effort.”

Making a positive long-term social impact 

Like TASC campers, Code.Tulsa interns came away from the experience motivated to make a positive impact. While most MIT students won’t go on to full-time professional roles traditionally associated with social impact, PKG Center programming like Code.Tulsa helps students recognize they can promote the public interest no matter their career. As Elvis Chipiro, a junior in computer science and engineering, reflected after interning with the Cherokee Nation, “Social impact is not separate from mainstream technology; rather, it is embedded in the choices engineers make every day … Ultimately, meaningful social change is not driven by technology alone, but by people willing to design systems with empathy, responsibility, and inclusion at the center.”

For others, Code.Tulsa helps them reconnect with a sense of public purpose. “Remembering that … I could use my MIT education to help others was a big reason I applied to MIT in the first place,” says Harper. “But I forgot my own mission in the stress of school. Code.Tulsa really re-opened my eyes to why I am here.”