Five major foundation models released in a single week!…
Google pledges to fix Gemini’s inaccurate and biased image generation
Google’s Gemini model has come under fire for its production of historically-inaccurate and racially-skewed images, reigniting concerns about bias in AI systems. The controversy arose as users on social media platforms flooded feeds with examples of Gemini generating pictures depicting racially-diverse Nazis, black medieval English kings,…
Stability AI previews Stable Diffusion 3 text-to-image model
London-based AI lab Stability AI has announced an early preview of its new text-to-image model, Stable Diffusion 3. The advanced generative AI model aims to create high-quality images from text prompts with improved performance across several key areas. The announcement comes just days after Stability AI’s…
New model identifies drugs that shouldn’t be taken together
Any drug that is taken orally must pass through the lining of the digestive tract. Transporter proteins found on cells that line the GI tract help with this process, but for many drugs, it’s unknown which of those transporters they use to exit the digestive tract.
Identifying the transporters used by specific drugs could help to improve patient treatment because if two drugs rely on the same transporter, they can interfere with each other and should not be prescribed together.
Researchers at MIT, Brigham and Women’s Hospital, and Duke University have now developed a multipronged strategy to identify the transporters used by different drugs. Their approach, which makes use of both tissue models and machine-learning algorithms, has already revealed that a commonly prescribed antibiotic and a blood thinner can interfere with each other.
“One of the challenges in modeling absorption is that drugs are subject to different transporters. This study is all about how we can model those interactions, which could help us make drugs safer and more efficacious, and predict potential toxicities that may have been difficult to predict until now,” says Giovanni Traverso, an associate professor of mechanical engineering at MIT, a gastroenterologist at Brigham and Women’s Hospital, and the senior author of the study.
Learning more about which transporters help drugs pass through the digestive tract could also help drug developers improve the absorbability of new drugs by adding excipients that enhance their interactions with transporters.
Former MIT postdocs Yunhua Shi and Daniel Reker are the lead authors of the study, which appears today in Nature Biomedical Engineering.
Drug transport
Previous studies have identified several transporters in the GI tract that help drugs pass through the intestinal lining. Three of the most commonly used, which were the focus of the new study, are BCRP, MRP2, and PgP.
For this study, Traverso and his colleagues adapted a tissue model they had developed in 2020 to measure a given drug’s absorbability. This experimental setup, based on pig intestinal tissue grown in the laboratory, can be used to systematically expose tissue to different drug formulations and measure how well they are absorbed.
To study the role of individual transporters within the tissue, the researchers used short strands of RNA called siRNA to knock down the expression of each transporter. In each section of tissue, they knocked down different combinations of transporters, which enabled them to study how each transporter interacts with many different drugs.
“There are a few roads that drugs can take through tissue, but you don’t know which road. We can close the roads separately to figure out, if we close this road, does the drug still go through? If the answer is yes, then it’s not using that road,” Traverso says.
The researchers tested 23 commonly used drugs using this system, allowing them to identify transporters used by each of those drugs. Then, they trained a machine-learning model on that data, as well as data from several drug databases. The model learned to make predictions of which drugs would interact with which transporters, based on similarities between the chemical structures of the drugs.
Using this model, the researchers analyzed a new set of 28 currently used drugs, as well as 1,595 experimental drugs. This screen yielded nearly 2 million predictions of potential drug interactions. Among them was the prediction that doxycycline, an antibiotic, could interact with warfarin, a commonly prescribed blood-thinner. Doxycycline was also predicted to interact with digoxin, which is used to treat heart failure, levetiracetam, an antiseizure medication, and tacrolimus, an immunosuppressant.
Identifying interactions
To test those predictions, the researchers looked at data from about 50 patients who had been taking one of those three drugs when they were prescribed doxycycline. This data, which came from a patient database at Massachusetts General Hospital and Brigham and Women’s Hospital, showed that when doxycycline was given to patients already taking warfarin, the level of warfarin in the patients’ bloodstream went up, then went back down again after they stopped taking doxycycline.
That data also confirmed the model’s predictions that the absorption of doxycycline is affected by digoxin, levetiracetam, and tacrolimus. Only one of those drugs, tacrolimus, had been previously suspected to interact with doxycycline.
“These are drugs that are commonly used, and we are the first to predict this interaction using this accelerated in silico and in vitro model,” Traverso says. “This kind of approach gives you the ability to understand the potential safety implications of giving these drugs together.”
In addition to identifying potential interactions between drugs that are already in use, this approach could also be applied to drugs now in development. Using this technology, drug developers could tune the formulation of new drug molecules to prevent interactions with other drugs or improve their absorbability. Vivtex, a biotech company co-founded in 2018 by former MIT postdoc Thomas von Erlach, MIT Institute Professor Robert Langer, and Traverso to develop new oral drug delivery systems, is now pursuing that kind of drug-tuning.
The research was funded, in part, by the U.S. National Institutes of Health, the Department of Mechanical Engineering at MIT, and the Division of Gastroenterology at Brigham and Women’s Hospital.
Other authors of the paper include Langer, von Erlach, James Byrne, Ameya Kirtane, Kaitlyn Hess Jimenez, Zhuyi Wang, Natsuda Navamajiti, Cameron Young, Zachary Fralish, Zilu Zhang, Aaron Lopes, Vance Soares, Jacob Wainer, and Lei Miao.
MIT Solve announces 2024 Global Challenges and Indigenous Communities Fellowship
The driving mission of MIT Solve is inviting new voices and proposed solutions to world problems as a way to achieve a more sustainable and equitable future for all. To that end, Solve recently announced the 2024 Global Challenges and the Indigenous Communities Fellowship to help find and scale the best.
Solve invites anyone from anywhere in the world to submit a solution to this year’s Global Challenges by April 18. Solve is seeking solutions that use technology in innovative and equitable ways to make learning more inclusive, mitigate and adapt to the climate crisis, improve access to quality health care, build peaceful and prosperous economies, and strengthen Indigenous communities.
Selected innovators will form the 2024 Solver Class, pitch their solutions during U.N. General Assembly Week, and share over $1 million of available funding. Innovators also take part in a nine-month support program that includes capital, leadership, and community support to scale their solutions.
“MIT Solve is on a quest to find the amazing innovators solving the pressing challenges of their communities and the world. And once we select the best, we mobilize the Solve community to help them scale,” says Hala Hanna, executive director of MIT Solve. “We can’t do this without our generous and foresighted supporters.”
Funding available for selected Solvers and fellows includes:
- MIT Solve funding: $10,000 to each Solver and fellow selected;
- GM Prize (supported by General Motors) for solutions that help create smart, safe, and sustainable communities around the world, selected from the 2024 Global Learning Challenge, the 2024 Global Climate Challenge, and the 2024 Indigenous Communities Fellowship;
- GSR Foundation Prize (supported by GSR Foundation) for solutions that use technology in an innovative way to address pressing issues in their communities, especially solutions that remove barriers to financial inclusion and place a strong emphasis on learning, selected from any 2024 Global Challenge;
- Morgridge Family Foundation AI Innovation Prize (supported by Morgridge Family Foundation) for solutions that use AI to boldly spark change through innovation, disruption, and transformation, selected from any 2024 Global Challenge or from any Solver class;
- AI for Humanity Prize (supported by The Patrick J. McGovern Foundation) for solutions that leverage data science, artificial intelligence, and/or machine learning to benefit humanity, selected from any 2024 Global Challenge; and
- Prince Albert II of Monaco Ocean Innovation Prize (supported by Prince Albert II of Monaco Foundation) for a solution that supports innovation for coasts, oceans, and the broader blue economy, selected from the 2024 Global Climate Challenge.
Additional prizes will also be announced.
The Solve community will convene on MIT’s campus for its flagship event, Solve at MIT, May 22-23 to celebrate the past 2023 Solver Class. Members of the public may request an invitation, while press interested in attending the event should contact maya.bingaman@solve.mit.edu.
The Friday Roundup – YouTube Tips and VideoStudio
How to Get Views on a Zero YouTube Channel I wanted to include a couple of YouTube specific video tutorials this week as we are still at the beginning of 2024. Getting started on YouTube is not too hard these days and there are a ton…
Play it again, Spirio
Seated at the grand piano in MIT’s Killian Hall last fall, first-year student Jacqueline Wang played through the lively opening of Mozart’s “Sonata in B-flat major, K.333.” When she’d finished, Mi-Eun Kim, pianist and lecturer in MIT’s Music and Theater Arts Section (MTA), asked her to move to the rear of the hall. Kim tapped at an iPad. Suddenly, the sonata she’d just played poured forth again from the piano — its keys dipping and rising just as they had with Wang’s fingers on them, the resonance of its strings filling the room. Wang stood among a row of empty seats with a slightly bemused expression, taking in a repeat of her own performance.
“That was a little strange,” Wang admitted when the playback concluded, then added thoughtfully: “It sounds different from what I imagine I’m playing.”
This unusual lesson took place during a nearly three-week residency at MIT of the Steinway Spirio | r, a piano embedded with technology for live performance capture and playback. “The residency offered students, faculty, staff, and campus visitors the opportunity to engage with this new technology through a series of workshops that focused on such topics as the historical analysis of piano design, an examination of the hardware and software used by the Spirio | r, and step-by-step guidance of how to use the features,” explains Keeril Makan, head of MIT Music and Theater Arts and associate dean of the School of Humanities, Arts, and Social Sciences.
Wang was one of several residency participants to have the out-of-body experience of hearing herself play from a different vantage point, while watching the data of her performance scroll across a screen: color-coded rectangles indicating the velocity and duration of each note, an undulating line charting her use of the damper pedal. Wang was even able to edit her own performance, as she discovered when Kim suggested her rhythmic use of the pedal might be superfluous. Using the iPad interface to erase the pedaling entirely, they listened to the playback again, the notes gaining new clarity.
“See? We don’t need it,” Kim confirmed with a smile.
“When MIT’s new music building (W18) opens in spring 2025, we hope it will include this type of advanced technology. It would add value not just to Wang’s cohort of 19 piano students in the Emerson/Harris Program, which provides a total of 71 scholars and fellows with support for conservatory-level instruction in classical, jazz, and world music. But could also offer educational opportunities to a much wider swath of the MIT community,” says Makan. “Music is the fifth-most popular minor at MIT; 1,700 students enroll in music and theater arts classes each semester, and the Institute is brimming with vocalists, composers, instrumentalists, and music history students.”
According to Kim, the Spirio enables insights beyond what musicians could learn from a conventional recording; hearing playback directly from the instrument reveals sonic dimensions an MP3 can’t capture. “Speaker systems sort of crunch everything down — the highs and the lows, they all kind of sound the same. But piano solo music is very dynamic. It’s supposed to be experienced in a room,” she says.
During the Spirio | r residency, students found they could review their playing at half speed, adjust the volume of certain notes to emphasize a melody, transpose a piece to another key, or layer their performance — prerecording one hand, for example, then accompanying it live with the other.
“It helps the student be part of the learning and the teaching process,” Kim says. “If there’s a gap between what they imagined and what they hear and then they come to me and say, ‘How do I fix this?’ they’re definitely more engaged. It’s an honest representation of their playing, and the students who are humbled by it will become better pianists.”
For Wang, reflecting on her lesson with Kim, the session introduced an element she’d never experienced since beginning her piano studies at age 5. “The visual display of how long each key was played and with what velocity gave me a more precise demonstration of the ideas of voicing and evenness,” Wang says. “Playing the piano is usually dependent solely on the ears, but this combines with the auditory experience a visual experience and statistics, which helped me get a more holistic view of my playing.”
As a first-year undergraduate considering a Course 6 major (electrical engineering and computer science, or EECS), Wang was also fascinated to watch Patrick Elisha, a representative from Steinway dealer M. Steinert & Sons, disassemble the piano action to point out the optical sensors that measure the velocity of each hammer strike at 1,020 levels of sensitivity, sampled 800 times per second.
“I was amazed by the precision of the laser sensors and inductors,” says Wang. “I have just begun to take introductory-level courses in EECS and am just coming across these concepts, and this certainly made me more excited to learn more about these electrical devices and their applications. I was also intrigued that the electrical system was added onto the piano without interfering with the mechanical structure, so that when we play the Spirio, our experience with the touch and finger control was just like that of playing a usual Steinway.”
Another Emerson/Harris scholar, Víctor Quintas-Martínez, a PhD candidate in economics who resumed his lapsed piano studies during the Covid-19 pandemic, visited Killian Hall during the residency to rehearse a Fauré piano quartet with a cellist, violist, and violinist. “We did a run of certain passages and recorded the piano part. Then I listened to the strings play with the recording from the back of the hall. That gave me an idea of what I needed to adjust in terms of volume, texture, pedal, etc., to achieve a better balance. Normally, when you’re playing, because you’re sitting behind the strings and close to the piano, your perception of balance may be somewhat distorted,” he notes.
Kim cites another campus demographic ripe for exploring these types of instruments like the Spirio | r and its software: future participants in MIT’s relatively new Music Technology Master’s Program, along with others across the Institute whose work intersects with the wealth of data the instrument captures. Among them is Praneeth Namburi, a research scientist at the MIT.nano Immersion Lab. Typically, Namburi focuses his neuroscience expertise on the biomechanics of dancing and expert movement. For two days during the MTA/Spirio residency, he used the sensors at the Immersion Lab, along with those of the Spirio, to analyze how pianists use their bodies.
“We used motion capture that can help us contrast the motion paths of experts such as Mi-Eun from those of students, potentially aiding in music education,” Namburi recounts, “force plates that can give scientific insights into how movement timing is organized, and ultrasound to visualize the forearm tissues during playing, which can potentially help us understand musicianship-related injuries.”
“The encounter between MTA and MIT.nano was something unique to MIT,” Kim believes. “Not only is this super useful for the music world, but it’s also very exciting for movement researchers, because playing piano is one of the most complex activities that humans do with our hands.”
In Kim’s view, that quintessentially human complexity is complemented by these kinds of technical possibilities. “Some people might think oh, it’s going to replace the pianist,” she says. “But in the end it is a tool. It doesn’t replace all of the things that go into learning music. I think it’s going to be an invaluable third partner: the student, the teacher, and the Spirio — or the musician, the researcher, and the Spirio. It’s going to play an integral role in a lot of musical endeavors.”
3 Questions: Shaping the future of work in an age of AI
The MIT Shaping the Future of Work Initiative, co-directed by MIT professors Daron Acemoglu, David Autor, and Simon Johnson, celebrated its official launch on Jan. 22. The new initiative’s mission is to analyze the forces that are eroding job quality and labor market opportunities for non-college workers and identify innovative ways to move the economy onto a more equitable trajectory. Here, Acemoglu, Autor, and Johnson speak about the origins, goals, and plans for their new initiative.
Q: What was the impetus for creating the MIT Shaping the Future of Work Initiative?
David Autor: The last 40 years have been increasingly difficult for the 65 percent of U.S. workers who do not have a four-year college degree. Globalization, automation, deindustrialization, de-unionization, and changes in policy and ideology have led to fewer jobs, declining wages, and lower job quality, resulting in widening inequality and shrinking opportunities.
The prevailing economic view has been that this erosion is inevitable — that the best we can do is focus on the supply side, educating workers to meet market demands, or perhaps providing some offsetting transfers to those who have lost employment opportunities.
Underpinning this fatalism is a paradigm which says that the factors shaping demand for work, such as technological change, are immutable: workers must adapt to these forces or be left behind. This assumption is false. The direction of technology is something we choose, and the institutions that shape how these forces play out (e.g., minimum wage laws, regulations, collective bargaining, public investments, social norms) are also endogenous.
To challenge a prevailing narrative, it is not enough to simply say that it is wrong — to truly change a paradigm we must lead by showing a viable alternative pathway. We must answer what sort of work we want and how we can make policies and shape technology that builds that future.
Q: What are your goals for the initiative?
Daron Acemoglu: The initiative’s ambition is not modest. Simon, David, and I are hoping to make advances in new empirical work to interpret what has happened in the recent past and understand how different types of technologies could be impacting prosperity and inequality. We want to contribute to the emergence of a coherent framework that can inform us about how institutions and social forces shape the trajectory of technology, and that helps us to identify, empirically and conceptually, the inefficiencies and the misdirections of technology. And on this basis, we are hoping to contribute to policy discussions in which policy, institutions, and norms are part of what shapes the future of technology in a more beneficial direction. Last but not least, our mission is not just to do our own research, but to help build an ecosystem in which other, especially younger, researchers are inspired to explore these issues.
Q: What are your next steps?
Simon Johnson: David, Daron, and I plan for this initiative to move beyond producing insightful and groundbreaking research — our aim is to identify innovative pro-worker ideas that policymakers, the private sector, and civil society can use. We will continue to translate research into practice by regularly convening students, scholars, policymakers, and practitioners who are shaping the future of work — to include fortifying and diversifying the pipeline of emerging scholars who produce policy-relevant research around our core themes.
We will also produce a range of resources to bring our work to wider audiences. Last fall, David, Daron, and I wrote the initiative’s inaugural policy memo, entitled “Can we Have Pro-Worker AI? Choosing a path of machines in service of minds.” Our thesis is that, instead of focusing on replacing workers by automating job tasks as quickly as possible, the best path forward is to focus on developing worker-augmenting AI tools that enable less-educated or less-skilled workers to perform more expert tasks — as well as creating work, in the form of new productive tasks, for workers across skill and education levels.
As we move forward, we will also look for opportunities to engage globally with a wide range of scholars working on related issues.