When cloud storage firm Dropbox decided to close its offices with the outbreak of the COVID-19 pandemic, co-founder and CEO Drew Houston ’05 had to send nearly 3,000 of the company’s employees home and tell them they could not work. But not coming back. anytime soon. “It felt like I was declaring a snow day or something.”

In the early days of the pandemic, Houston says Dropbox reacted as many others did to make sure employees were safe and customers were taken care of. “It is real, there is no playbook on Zoom for running a global company in a pandemic. For a lot of this we were taking it as we go. ,

Houston talked about her experience leading Dropbox through a public health crisis and how COVID-19 has hit MIT’s Stephen A. In a fireside chat with Dan Huttenlocher, dean of Schwarzman College of Computing, distributed work has accelerated.

During the discussion, Houston also mentioned its $10 million gift to MIT, which would provide the first shared professorship between the MIT Schwarzman College of Computing and the MIT Sloan School of Management, as well as providing a catalyst startup fund for the college. .

“The goal is to find ways to unlock more of our brainpower through a multidisciplinary approach between computing and management,” Houston says. “It’s often at the intersection of these disciplines where you can bring people together from different perspectives, where you can really unlock big. I think academia has a big role to play. [here], and I think MIT is in a very good position to lead. So, I want to do anything I can to help with that.”

virtual first

While the sudden swing to remote work was unexpected, Houston says it was pretty clear that the whole way of working as we knew it was going to change indefinitely for knowledge workers. “There’s a silver lining to every crisis,” says Houston, noting that people have been using Dropbox for years to do things more flexibly, so it’s time for the company to lean on and be early adopters of the distributed work paradigm. It makes sense to become one in which employees work in different physical locations.

Dropbox redesigned the work experience across the company in October 2020, unveiling a “virtual first” working model, with remote work being the primary experience for all employees. Personal work spaces went by the wayside and offices located in areas with high concentrations of employees were converted into combination and collaborative spaces called Dropbox studios for working individually with teammates.

“There is a lot we can say about Covid, but for me, the most important thing is that we will look back in 2020 as the year when we are going to be working out of offices permanently, mainly working out of screens. Were were It’s a transition that’s been going on for a while, but Covid has completely ended the swing,” says Houston.

Video thumbnails

play video

Imagining a Future Workplace: A Fireside Chat with Dropbox’s Drew Houston

Designing for the workplace of the future

Houston says the pandemic prompted Dropbox to reevaluate its product line and think about ways to improve. “We have this whole new way of working that’s forced upon us. Nobody designed it; it just happened. Even tools like Zoom, Slack and Dropbox are old world and designed for that.” were done.”

Going through that process helped Dropbox gain clarity on where they could add value and led to the realization that they needed to go back to their roots. “In many ways, what people in theory need today is exactly what they needed in the beginning — a place for all their stuff,” Houston says.

Dropbox revamped its product roadmap to revamp efforts from syncing files to organizing cloud content. The company is focusing on moving towards this new direction with the release of new automation features that users can easily implement to better organize their uploaded content and find it quickly. Dropbox also recently announced the acquisition of Command E, a universal search and productivity company to help accelerate its efforts in this area.

Houston sees Dropbox as still evolving and sees many opportunities ahead in this new era of distributed work. “We need to design better tools and smart systems. It’s not just the individual parts, but how they are woven together.” He is surprised at how little intelligence is actually integrated into current systems and He believes that rapid advances in AI and machine learning will soon lead to a new generation of smart tools that will eventually reshape the nature of work – “in the same way that we had with the new generation of cloud tools.” It has revolutionized the way we work and we have all the advantages we could not have imagined.”

founding roots

Houston famously turned its frustrations on carrying USB drives and emailing the files to itself in a demo by becoming Dropbox.

After graduating from MIT in 2005 with bachelor’s degrees in electrical engineering and computer science, he, along with fellow classmate Arash Ferdowsi, co-founded Dropbox in 2007 and created a platform for a service used by 700 million people worldwide. Led the company’s development from simple thought. Today.

Houston credits MIT with preparing her well for her entrepreneurial journey, recalling that what surprised her most about her student experience was how much she learned outside of the classroom. At the event, he emphasized the importance of developing both sides of the brain to a select group of computer science and management students who were in attendance, and a wider live stream audience. “One thing you learn about starting a company is that the hardest problems usually aren’t technical problems; they’re people problems.” He says he didn’t realize it at the time, but that some of his first lessons in management were gained by taking on responsibilities in his fraternity and in various student organizations, creating a sense of being “on the hook”.

As CEO, Houston has had the opportunity to see behind the scenes how things happen and appreciate that problems don’t solve themselves. While individual people can make a big difference, he points out that many of the challenges facing the world at the moment are inherently multidisciplinary, which sparked his interest in the MIT Schwarzman College of Computing.

He says the college’s mindset of connecting computing to other disciplines resonated and inspired him to launch his biggest philanthropic effort as soon as possible because “we don’t have that much time to address these problems.”

Read More

Sometimes, the software is just like us. It can be bloated, slow and messy. One can see a doctor if these symptoms persist (perhaps not for a glitch), but rarely do we push a flawed software program to see its developer over and over again.

The answer to why our software is flawed lies in a web of reliance on flashy hardware, the limitations of a “code-and-fix” approach, and inadequate design. MIT Professor Daniel Jackson, associate director of MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL), looked to existing limitations to create a new framework to improve the way our programs function. His theory of software design takes a human-centered approach that views an app as a collection of interacting concepts. Jackson’s new book, “The Essence of Software,” draws on his many years of software research, including designing an open source language and analyzer alloy for software modeling.

Why: Worms. Security flaws. Design flaws. Has software always been bad?

a: The software is actually better than ever. It’s just that the power and functionality of software has grown so fast that we haven’t always been able to keep up with it. And there are some software products (for example Apple Keynote) that are close to perfect – easy to use, flexible, almost no bugs. My book provides a vision that will empower everyone to make software good.

Why: In your new book, “The Essence of Software,” you introduce a theory of software design that shows how a software system can be viewed as “a collection of interaction concepts.” How does this overturn conventional wisdom?

a: First, conventional wisdom sees the user experience primarily in the user interface – its layout, colors, labels, etc. The concept design goes deeper, addressing the fundamental mechanisms and user experiences created by the programmer.

Second, most apps have large areas of overlapping functionality, but current approaches do not recognize this, and developers repeatedly create similar pieces of functionality as if they were new, without taking advantage of the fact that they were built multiple times. Huh. before this. For example, just think about how many social media apps have implemented up-voting or comments or favorites. Concepts allow you to identify these reuse opportunities and take advantage of accumulated design knowledge.

Why: The year 2021 was one of the worst years for data breaches and cyber attacks – we have seen fragility in everything from electronic medical records to social media and big tech companies. Can your approach help with security loopholes?

a: A high proportion of safety and security issues come from a lack of clarity in the design. Concepts can help with this. More directly, concepts can ensure that users truly understand the implications of their actions, and we know that many disasters happen because users do the wrong thing. In the field of security, allowing the user to do something wrong (such as granting access to someone who shouldn’t have access) is usually the easiest way to control a system. So, if you can design an app that makes it harder for users to do things they will regret, you can reduce this problem.

Read More

Shardul Chiplunkar, a senior in Course 18C (Maths with Computer Science), entered MIT with an interest in computers, but was soon trying to do everything from extinguishing fires to building firewalls. He worked in audio engineering and glass blowing, had a stint for the MIT/Wellesley Toons a cappella group, and learned to sail.

“When I was entering MIT, I thought I would be interested in math and computer science, academics, and research.” “Now what I appreciate most is diversity of people and ideas.”

Academically, his focus is on the interface between people and programming. But his extracurriculars have helped him realize his secondary goal, to be a translator of sorts between the tech world and professional users of software.

“I want to create a better conceptual framework to explain and understand complex software systems, and to develop better tools and methodologies for professional software development at large through fundamental research in programming languages ​​and the theory of human-computer interaction.” “

It’s a role he was practically born to play. Growing up in Silicon Valley, at the height of the dot-com bubble, he was drawn to computers at an early age. He was 8 years old when his family moved to Pune, India for his father’s job as a networking software engineer. In Pune, his mother also worked as a translator, editor and radio newscaster. Chiplunkar could eventually speak English, Hindi, French and his native Marathi.

At school, he was active in math and coding competitions, and a friend introduced him to linguistic puzzles, which he recalls were “like math.” He excelled at the Linguistics Olympiad, where secondary school students solve problems based on linguistics – the scientific study of languages.

Chiplunkar came to MIT to study what he calls the “Perfect Major,” course 18c. But as the child of a tech dad and a translator mom, it was perhaps inevitable that Chiplunkar would figure out how to combine the two disciplines into a unique career trajectory.

While he was a natural at human languages, it was a Computer Science and Artificial Intelligence Laboratory undergraduate Research Opportunities program that solidified his interest in researching programming languages. Under Professor Adam Hiddle, he developed a specification language for Internet firewalls, and a formally verified compiler to convert such specifications into executable code, using correct-by-construction software synthesis and proof-of-construction techniques. developed by

“Let’s say you want to block a certain website,” Chiplunkar explains. “You open your firewall and enter the website address, how long you want it to block, and so on. You have some parameters in a built in language that tells the firewall what code to run. But you How to know that a firewall will code that language without a mistake? That was the gist of the project. I was trying to create a language to specify the behavior of firewalls mathematically, and to convert it into code and to prove that the code will do what you want it to do. The software will come with mathematically proven guarantees.”

He has also explored proximate interests in probabilistic programming languages ​​and program inference through cognitive science research, working under Professor Tobias Gerstenberg at Stanford University and later Joshua Rule in the Tenenbaum Laboratory in MIT’s Department of Brain and Cognitive Sciences. working under.

Chiplunkar says, “In regular programming languages, the basic data you deal with are atomic, fixed numbers. But in probabilistic programming languages, you deal with probability distributions. Instead of a constant five, you can have one random variable.” which has an average value of five, but every time you run the program it is somewhere between zero and 10. It turns out that you can calculate with these probabilities as well – and that some aspects of human cognition are computerized. There’s a more powerful way to model. Language lets you express concepts that you can’t express otherwise.”

“There are many reasons I love computational cognitive science, the same reasons I love programming and human languages,” he explains. “Human cognition can often be expressed in a representation that is like a programming language. It is more of an abstract representation. We do not know what actually happens in the brain, but the hypothesis is that some level of abstraction But, it is a good model for how cognition works.

Chiplunkar also hopes to bring a better understanding of modern software systems to the public domain, to empower techno-curious communities such as lawyers, policy makers, doctors and educators. To aid in this pursuit, he has taken courses at MIT on Internet policy and copyright law, and follows the work of digital rights and freedom activists. He believes that talking about the architecture of computer systems for broader social purposes requires a fundamentally new language and concepts for programmers.

“I want us to be able to explain why a surgeon should rely on a robotic surgery assistant, or how a law about data storage needs to be updated for modern systems.” “I think creating better conceptual languages ​​for complex software is just as important as creating better practical tools. Because complex software is so important in the world now, I want the computing industry – and I – to engage with a wider audience. be better able.

Read More

As her topic for the 2021 Mildred S Dresselhaus lecture, Stanford University professor Jelena Vuskovic posed a question: Are computers better than humans at designing photonics?

Throughout his speech, presented in a hybrid format to more than 500 attendees on November 15, the Jensen Huang Professor in Global Leadership at the Stanford School of Engineering presented several examples arguing that yes, computer software Can help identify better solutions than traditional methods. Leading to smaller, more efficient devices as well as completely new functionalities.

Photonics, the science of guiding and manipulating light, is used in many applications such as optical interconnects, optical computing platforms for AI or quantum computing, augmented reality glasses, biosensors, medical imaging systems, and sensors in autonomous vehicles.

For all of these applications, Vuskovic said, multiple optical components must be integrated onto a single chip that can fit the footprint of your glasses or mobile device. Unfortunately, there are several problems with high-density photonic integration. Conventional photonic components are large, sensitive to environmental factors such as fabrication errors and temperature changes, and are designed by manual tuning with few parameters. So, Vuskovic and his team asked, “How can we design better photonics?”

His answer: Photonics inverse design. In this process, scientists rely on sophisticated computational tools and modern computing platforms to find the optimal photonic solution or device design for a particular function. In this inverse process, the researcher first considers how he or she would like the photonic block to operate, then uses computer software to find the entire parameter space of possible solutions that is optimal, within construction restrictions.

From guiding light around corners to dividing colors of light in a compact footprint, Vuskovic presented several examples to prove this process – using computer software to conduct physics-guided searches of many possibilities. Unconventional solutions are generated that increase the efficiency and/or reduce the footprint of photonic devices.

Video thumbnails

play video

2021 Mildred S. Dresselhaus Lecture: Jelena Vuskovic, Stanford University

Enabling New Functionalities – High-Energy Physics

State-of-the-art particle accelerators, which use microwave or radio frequency waves to propel charged particles, can be the size of a full city block; For example, Stanford’s SLAC National Accelerator Lab is two miles long. Low-energy accelerators, such as those used in medical radiation facilities, are not that large but still occupy an entire room, are expensive, and are not very accessible. “If we can use a different spectrum of electromagnetic waves with shorter wavelengths to do the same thing as accelerating particles,” Vuskovic said, “we should, in principle, reduce the size of an accelerator.” should be able to.” The solution is not as simple as reducing the size of all parts, as electromagnetic building blocks will not work for optical waves. Instead, Vuskovic and his team used the inverse design process to create new building blocks, and built a single-stage on-chip integrated laser-driver particle accelerator that is only 30 micrometers in length.

Micrograph of a micrometer-long piece of fabricated silicon.

A few-micrometer-long piece of fabricated silicon that acts as a compact stage of a particle accelerator and accelerates electrons by interacting with a coupled laser field. This structure can shrink linear accelerators on a silicon chip from miles to an inch.

Image courtesy of Jelena Vukovic.

Applying inversely designed photonics to practical environments

Autonomous vehicles have a large lidar system on the roof housing mechanics that enables the rotation of the beam to scan the environment. Vuskovic considers how it can be improved. “Could you build this system inside the footprint of a chip that would be like another sensor in your car, and could it be cheaper?” Through inverse design, his research group found optimal photonic structures to enable beams to be driven with lasers cheaper than with state-of-the-art systems, and gained 5 degrees of additional beam steering.

Next up: Scaling up a superconducting quantum processor on a diamond or silicon carbide chip. In this example, Vuskovic retracts the 2020 Dresselhaus lecture given by Harvard Professor Evelyn Hu on taking advantage of defects at the nanoscale. By relying on impurities at low concentrations in these materials, naturally trapped atoms can be very useful for quantum applications. Vuskovic’s group is working on material development and fabrication techniques that allow them to place these trapped atoms in the desired state with minimal defects.

“For many applications, letting computer software search for an optimal solution leads to better solutions than you designed or anticipated based on your intuition. And the process is material-agnostic, in line with commercial foundry.” compatible, and enables new functionalities,” Vuskovic said. “Even if you try to build something better than traditional solutions – one that is smaller in footprint or higher in efficiency – we can come up with many solutions that are equally as good or less efficient than before. better. We’re re-learning photonics and electromagnetics in the process.”

Mildred S. Honoring Dresselhaus and Jean Dresselhaus

Vuskovic was the third speaker to deliver the Dresselhaus Lecture, established in 2019 to honor the late MIT physics and electrical engineering professor Mildred Dresselhaus. This year, the lecture was also dedicated to Jean Dresselhaus, the famous physicist and husband of Millie, who passed away at the end of September 2021.

Jing Kang, a professor of electrical engineering and computer science at MIT, opened the lecture by reflecting on Dresselhaus’s scientific achievements. Kong highlights the American Physical Society’s Oliver E. Buckley Condensed-Matter Physics Prize—considered the most prestigious award given in the field of condensed-matter physics—given to both Millie (2008) and Jean (2022). “Although they worked together on many important topics,” Kong said, “it is remarkable that they received this award for separate research work. It is our privilege to follow in their footsteps.”

Read More

While standing in the kitchen, you push a few metal bowls across the counter into the sink with a clang, and drape a towel over the back of a chair. In the other room, it looks like some precariously stacked wooden blocks fell off, and there’s an epic toy car accident. These interactions with our environment are something that humans experience on a daily basis at home, but this world may seem real, but it is not.

A new study from researchers at MIT, the MIT-IBM Watson AI Lab, Harvard University and Stanford University is enabling a rich virtual world, which is like stepping into “The Matrix.” Their platform, called Thredworld (TDW), simulates high-fidelity audio and visual environments, both indoor and outdoor, and allows users, objects and mobile agents to interact in real life and according to the laws of physics. allows. Object orientations, physical characteristics and velocities are calculated and calculated for liquids, soft bodies and hard objects, as interactions occur, producing precise collision and impact sounds.

TDW is unique in that it is designed to be flexible and generic, generate synthetic photo-realistic visual and audio renderings in real time, which can be compiled into audio-visual datasets, interacting within the scene Can be modified through, and adapted to human and nervous. Network learning and prediction testing. A variety of robotic agents and avatars can also be created within controlled simulations to perform, say, action planning and execution. And using virtual reality (VR), human attention and play behavior within space can provide real-world data, for example.

Study lead author Chuang Gan, MIT-IBM Watson AI Lab research scientist, says, “We are trying to build a general-purpose simulation platform that mimics the interactive richness of the real world for a variety of AI applications. “

Creating realistic virtual worlds with which to investigate human behavior and train robots has been a dream of AI and cognitive science researchers. “Most AI right now is based on supervised learning, which relies on huge datasets of human-annotated images or sounds,” says Josh McDermott, an associate professor in the Department of Brain and Cognitive Sciences (BCS) and an MIT-IBM Watson AI. Lab Project Lead. These descriptions are expensive to compile, creating a bottleneck for research. And for physical properties of objects, such as mass, which are not always readily apparent to human observers, labels may not be available at all. A simulator like TDW overcomes this problem by generating a view where all parameters and annotations are known. Many competing simulations were inspired by this concern but designed for specific applications; Through its flexibility, TDW aims to enable many applications that are poorly suited for other platforms.

Another advantage of TDW, McDermott notes, is that it provides a controlled setting to understand the learning process and facilitate the improvement of AI robots. Robotic systems that rely on trial and error can be taught in an environment where they cannot cause physical harm. Furthermore, “many of us are excited about the doors that open for experiments on humans to understand human perception and cognition in this type of virtual world. These very rich sensory landscapes have the potential to create , where you still have total control and complete knowledge of what is happening in the environment.”

McDermott, Gan, and their colleagues are presenting this research at the Conference on Neural Information Processing Systems (NeurIPS) in December.

behind the structure

The work began as a collaboration between a group of MIT professors, along with Stanford and IBM researchers, who linked individual research interests to hearing, vision, cognition and perceptual intelligence. TDW brought these together on one platform. “We were all interested in the idea of ​​building a virtual world for the purpose of training AI systems that we could actually use as brain models,” says McDermott, who studies human and machine hearing. does. “So, we thought this kind of environment, where you could have objects that would interact with each other and then present realistic sensory data from them, would be a valuable way to start studying this.”

To achieve this, the researchers built TDW on a video game platform called the Unity3D Engine and is committed to incorporating both visual and auditory data rendering without any animation. The simulation consists of two components: the build, which renders the images, synthesizes the audio, and runs the physics simulation; and Controller, which is a Python-based interface where the user sends commands to the build. Researchers build and populate a scene by dragging furniture pieces, animals and vehicles from an extensive 3D model library of objects. These models accurately respond to light changes, and their physical structure and orientation in the scene dictate their physical behavior in space. Dynamic lighting models accurately simulate visible illumination, creating shadows and blurring that correspond to the appropriate time of day and sun angle. The team has also created a well-equipped virtual floor plan that the researchers can fill with agents and avatars. To synthesize true-to-life audio, TDW uses generative models of effects sounds that are triggered by collisions or other object interactions within the simulation. TDW also simulates noise attenuation and reverberation according to the geometry of space and the objects in it.

Reactions between two physics engines and interacting objects in TDW power distortions—one for rigid bodies, and the other for soft objects and liquids. TDW calculates instantaneousness with respect to mass, volume and density as well as any frictional or other forces acting on the material. This allows machine learning models to learn how objects with different physical properties will behave together.

Users, agents and avatars can bring scenes to life in a number of ways. A researcher can apply force directly to an object through controller commands, which can virtually set a virtual ball in motion. Avatars may be empowered to act or behave in a certain way within space – for example, with articulated organs capable of performing work experiments. Finally, VR heads and handsets could allow users to interact with virtual environments, potentially to generate human behavioral data that machine learning models can learn.

rich AI experience

To test and demonstrate TDW’s unique features, capabilities, and applications, the team ran a battery of tests comparing datasets generated by TDW and other virtual simulations. The team found that neural networks trained on scene image snapshots with randomly placed camera angles from TDW outperformed snapshots of other simulations in image classification tests and are closer to systems trained on real-world images. The researchers also designed and trained a material classification model in TDW on audio clips of small objects falling on surfaces and asked it to identify the types of interacting materials. They found that TDW earned a significant advantage over its competitors. Additional object-drop testing with neural networks trained on TDW showed that the combination of audio and vision is the best way to identify physical properties of objects, prompting further studies of audio-visual integration.

TDW is proving particularly useful for designing and testing systems that understand how physical phenomena in a scene will evolve over time. This includes facilitating a benchmark of how well a model or algorithm makes physical predictions, for example, the stability of a stack of objects, or the motion of objects after a collision – humans learn many of these concepts as children. learn, but many machines need to demonstrate this ability to be useful in the real world. TDW has also enabled comparisons of human curiosity and prediction against machine agents designed to evaluate social interactions in different scenarios.

Gan points out that these applications are only the tip of the iceberg. By expanding TDW’s physical simulation capabilities to more accurately depict the real world, “we are seeking to create new benchmarks for advancing AI technologies, and use these benchmarks to open up many new problems.” which have been difficult to study until now.”

The research team on the paper also includes MIT engineers Jeremy Schwartz and Seth Alter, who are critical to the operation of TDW; BCS Professors James DiCarlo and Joshua Tenenbaum; graduate students Aidan Curtis and Martin Shrimpf; and former postdocs James Traer (now an adjunct professor at the University of Iowa) and Jonas Kubilius PhD ’08. His collaborators are IBM director of the MIT-IBM Watson AI Lab David Cox; Research Software Engineer Abhishek Bhandardar; and Dan Gutfreund, IBM’s research staff member. Additional researcher co-authors are Harvard University assistant professor Julian de Freitas; and from Stanford University, assistant professors Daniel LK Yamins (a TDW founder) and Nick Haber, postdoc Daniel M. Bear, and graduate students Megumi Sano, Kuno Kim, Elias Wang, Damien Moroca, Kevin Feigelis and Michael Lingelbach.

This research was supported by the MIT-IBM Watson AI Lab.

Read More

Every holiday season, a popular new video game causes a disproportionate amount of hype, anticipation, and last-minute purchases. But some of them offer a completely new way of playing the game. Even less has ripple effects that reach far beyond the gaming universe.

When Guitar Hero was released in 2005, challenging players to hit notes for classic rock songs on guitar-like controllers, it grew from a holiday hit to a cultural phenomenon that inspired a new generation of rock. Taught to love ‘n’ roll music. Along the way, it showed the video game industry the power of innovative, music-based games.

Guitar Hero and the related Rock Band franchise were developed by Harmonix Music Systems, formed at MIT’s Media Lab more than 25 years ago after a pair of friends set out to help people interact with music. technology began to be used. Since then, it has released over a dozen games that have helped millions of people experience the thrill of making music.

“The thing we’ve always tried to accomplish is to innovate in musical gameplay,” says Aron Egozzi ’93, SM ’95, a professor of practice in music and theater arts at MIT, who worked in the company with Alex Rigopulos. was co-founded. ’92, SM ’94. “That’s what the company is constantly trying to do – creating new types of compelling music experiences.”

To further that mission, Harmonix became part of industry giant Epic Games last month. It’s a major milestone for a company that has seen its game go from small passion projects to ubiquitous sources of expression and fun.

Egozi has seen harmonics games on the tour buses of famous bands, at the offices of tech giants like Google, in bars hosting “Rock Band Nights” and being featured in popular TV shows. Most importantly, they are heard from music teachers who say that the games inspired children to play real instruments.

In fact, Egozi just heard from the principal of his son’s school that the reason he plays drums is a rock band.

“This is probably the most gratifying part,” says Igozzi, who plays the clarinet professionally. “Of course, we had high hopes and aspirations when we started the company, but we didn’t think we’d really make such a big impact. We’re completely surprised.”

mission driven launch

As an undergraduate at MIT, Egozy majored in electrical engineering and computer science and majored in music, But he didn’t even think about combining computers and music until he attended an undergraduate research opportunity program in the Media Lab under then-graduate student Michael Hawley.

The experience inspired Egozzi to earn his master’s degree at Media Lab’s Opera of the Future group, led by Todd Machovar, where he began building software that produced music based on intuitive controls. He also met Rigopulos at the Media Lab, who quickly became a friend and colleague.

“Alex had this idea: Wouldn’t it be cool if we took a joystick that had a more friendly interface and used that to play the parameters of our generative music system?” The ego remembers.

The joystick-based system immediately became one of the most popular demonstrations at the Media Lab, prompting the pair to participate in the MIT $10K Entrepreneurship Contest (today MIT $100K).

“I think MIT has filled me with the feeling that there’s no point in trying to do something that someone has already done,” Egozi says. “If you’re going to work on something, try to do something inventive. It’s a widespread attitude around MIT, not just in the Media Lab.”

After graduation, Egozzi and Rigopulos knew they wanted to continue working on the system, but doubted they could find a company that would pay them to do so. Harmonics was born out of that simple logic.

The founders spent the next four years working on the technology, which led to a product called the X, which Egozzi describes as a “total flop”. They also built a system for Disney at Epcot Amusement Park and attempted to integrate their software with karaoke machines in Japan.

“We endured many failures trying to figure out what our business was all about, and it took us a long time to find a way to fulfill our mission, which is to make everyone in the world experience the joy of making music. As it turns out, that was through video games,” Egozi says.

Many of the company’s first video games weren’t huge hits, but by iterating on the core platform, Harmonix was able to continually improve the design and gameplay.

As a result, when it came time to create Guitar Hero around 2005, the founders had music, graphics and design systems they knew could work with unique controllers.

Igozzi has described Guitar Hero as a relatively low-budget project within Harmonix. The company had two games in development at the time, and the Guitar Hero team was small. It was also a quick turnaround: He finished Guitar Hero in about nine months.

Through its other releases, the Harmonix team was trained to expect most of its sales to come in the weeks leading up to the Christmas holiday, and then essentially close the sale. With Guitar Hero, the game sold out incredibly quickly – so quickly that retailers immediately wanted more, and the company that made the guitar controllers had to multiply their orders with manufacturers.

But what really surprised the founders was that January sales surpassed those of December. …then February surpassed January. In fact, month after month, the sales graph looked like nothing Harmonix’s team of 45 people had seen before.

“It was mostly shock and disbelief within Harmonix,” says Egozi. “We just love making Guitar Hero. It was the game we’ve always wanted to make. Everyone at Harmonix was involved in some sort of music. The company had a band room so people could go and jam. And so the fact that it also sold really well was extremely gratifying – and very unexpected.”

After that things went fast for Harmonix. Work on Guitar Hero 2 began immediately. Guitar Hero was taken over by Activision, and Harmonix was acquired by MTV Networks for several years. Harmonix developed the Rock Band franchise, which brought players together for lead guitar, bass, keyboards, drums and vocals of popular songs.

“It was really wonderful because it was about a group effort,” says Egozzi. “Rock Band was social in the sense that everyone was playing music together in the same room, not competing, but working toward a common goal.”

an ongoing legacy

Over the past decade, Harmonix has continued to explore new ways of musical gameplay with releases such as SingSpace, which offers a social karaoke experience, and Fuser, a DJ-inspired game that lets users mix and match different tracks. lets you. The company also released Rock Band VR, which makes players feel like they’re on stage in front of a live audience.

These days Egozzi, who has been on the board since becoming a full-time professor at MIT in 2014, teaches 21M.385/6.185 (Interactive Music Systems), a class that combines computer science, interaction design, and music. “This is the class I wish I was in here at MIT as an undergrad,” says Egozzi.

And every semester, the class visits the Harmonix office. They are often told that it is the students’ favorite part of the class.

“I’m really proud of what we were able to do, and I’m still amazed and humbled by our cultural influence,” Igozzi says. “There’s a generation of kids who grew up playing these games that have learned about all this music from the ’70s and ’80s. I’m really glad we were able to introduce kids to that great music.”

Read More

The spread of misinformation on social media is a serious societal problem that tech companies and policy makers grapple with, yet those who study the issue still lack a deep understanding of why and how false news spreads. .

To shed some light on this obscure topic, researchers at MIT developed a theoretical model of a Twitter-like social network to study how news is shared and detect situations where a non-verbal A credible news item will spread more widely than the truth. Agents in the model are motivated by a desire to persuade others to have their say: The main assumption in the model is that people are bothered to share something with their followers if they feel it is persuasive and encourages others to speak their mind. What is likely to take you closer is the mindset. Otherwise they won’t share.

The researchers found that in such a setting, when a network is highly connected or the views of its members are increasingly polarized, news that is likely to be false will spread more widely and tend to be more reliable than news with high credibility. will travel deeper into the network in comparison.

This theoretical work could inform empirical studies of the relationship between the credibility of news and the size of its dissemination, which could help social media companies optimize networks to limit the spread of false information.

“We show that, even if people are rational in how they decide to share news, it can lead to the amplification of information with less credibility. With this persuasion motive, I believe No matter how extreme – given that the more extreme they are, the more I gain from pushing others’ opinions – there is always someone who elevates it [the information]says senior author Ali Jadbaby, professor and head of the Department of Civil and Environmental Engineering and a core faculty member at the Institute for Data, Systems and Society (IDSS) and a principal investigator in the Laboratory for Information and Decision Systems. Cap).

The first authors to join forces on the paper are Chin-Chia Soo, a graduate student in the Social and Engineering Systems Program at IDSS, and Amir Azorlu, a LIDS research scientist. The research will be presented this week at the IEEE Conference on Decision and Control.

consider persuasion

The research builds on a 2018 study by Sinan Aral, the David Austin Professor of Management at the MIT Sloan School of Management; Deb Roy, Professor of Media Arts and Sciences in the Media Lab; and former postdoc Soroush Vosofi (now assistant professor of computer science at Dartmouth University). Their empirical study of Twitter’s data found that false news spreads wider, faster and deeper than real news.

Jadbai and her colleagues wanted to dig deeper into why this happens.

They hypothesized that persuasion may be a strong motive for sharing news – perhaps agents in the network want to persuade others to obey them – and decided to build a theoretical model that would let them explore this possibility.

In their model, agents have some prior beliefs about the policy, and their goal is to persuade followers to move their beliefs closer to the agent side of the spectrum.

A news item is initially issued to a small, random subset of agents, who must decide whether or not to share the news with their followers. An agent weighs the newsworthiness and credibility of the item, and updates its belief based on how surprising or reassuring the news is.

“They will do a cost-benefit analysis to see whether, on average, this news will drive people closer to what they think or drive them away. And we include a marginal cost to share. For example, taking some action. Well, if you’re scrolling through social media, you’ll have to stop to do so. Think of it as a cost. Or if I share something that’s embarrassing it may cost reputation. Everyone It costs money, so the more extreme and interesting the news is, the more likely you are to share it,” says Jadbai.

If the news confirms the agent’s point of view and has a driving force that exceeds the nominal cost, the agent will always share the news. But if an agent feels that the news is something that other people have already seen, the agent is discouraged from sharing it.

Since an agent’s willingness to share news is a product of his perspective and how persuasive the news is, the more extreme the agent’s approach or the more surprising the news, the more likely the agent will share it.

The researchers used this model to study how information spreads during the news cascade, an unbroken sharing chain that rapidly penetrates networks.

Connectivity and Polarization

The team found that when a network has high connectivity and the news is surprising, the reliability threshold for initiating a news cascade is low. High connectivity means that there are many connections between many users in the network.

Similarly, when the network becomes substantially polarized, there are many agents who are highly opinionated who want to share the news item, starting a news cascade. In both of these cases, news with low credibility forms the largest cascade.

“For any news, there is a natural network speed limit, a limit of connectivity, which facilitates good transmission of information where the size of the cascade is maximized by true news. If you cross the line, you will end up in situations where there is a lot of false news or news with low credibility,” says Jadbaby.

If the views of users in the network become more diverse, there is less chance that a poorly credible news will spread more widely than the truth.

Jadabi and his colleagues designed the agents in the network to behave rationally, so the model would better capture the actions that real humans might perform if they wanted to persuade others.

“One might say that’s why people don’t share, and that’s valid. Why people do certain things is a subject of intense debate in cognitive science, social psychology, neuroscience, economics, and political science,” he says. “Depending on your assumptions, you get different results. But it seems to me that this notion of persuasion being the motive is a natural assumption.”

Their model also shows how costs can be manipulated to reduce the spread of false information. Agents conduct a cost-benefit analysis and will not share the news if the cost of doing so exceeds the benefit of sharing.

“We don’t make any policy prescriptions, but one thing this work shows is that, perhaps, having some cost associated with news sharing isn’t a bad idea. The reason you get a lot of these cascades is that news sharing The cost of doing it is actually very low,” he says.

“The role of social networks in shaping thoughts and influencing behavior has been widely noted. Empirical research conducted by Sinan Aral among his colleagues at MIT shows that false news is more widespread than true news.” In their new paper, Ali Jadbaby and colleagues help us an elegant model for this puzzle, says Sanjeev Goel, professor of economics at the University of Cambridge, who was not involved in this research. provide an explanation”

This work was supported by an Army Research Office Multidisciplinary University Research Initiative grant and a Vannevar Bush fellowship from the Office of the Secretary of Defense.

Read More

For all that neural networks can accomplish, we still don’t really understand how they work. Sure, we can program them to learn, but understanding a machine’s decision-making process remains like a fancy puzzle with a dizzying, complicated pattern where too many integral pieces don’t yet fit.

For example, if a model was trying to classify the image of the said puzzle, it would encounter well-known but annoying adversarial attacks, or even more run-of-the-mill data or processing problems. could. But a new, more subtle type of failure recently identified by MIT scientists is another cause for concern: “overinterpretation,” where algorithms make confident predictions based on details that are incomprehensible to humans, such as random patterns or image borders.

This can be particularly worrisome for high-stakes environments, such as the split-second decision for self-driving cars, and medical diagnoses for diseases that require more immediate attention. Autonomous vehicles in particular rely heavily on systems that can accurately sense the surroundings and then make quick, safe decisions. The network used specific backgrounds, edges, or particular patterns of the sky to classify traffic lights and road signs – regardless of what else is in the image.

The team found that neural networks trained on popular datasets such as CIFAR-10 and ImageNet suffered more interpretation. For example, models trained on CIFAR-10 made confident predictions even when 95 percent of the input images were missing, and the rest are insensitive to humans.

“Exaggeration is a dataset problem that is caused by these redundant signals in the dataset. Not only are these high-confidence images unrecognizable, but insignificant areas such as borders make up less than 10 percent of the original image. We found these The images were meaningless to humans, yet models could still classify them with high confidence,” says Brandon Carter, MIT Computer Science and Artificial Intelligence Laboratory PhD student and lead author on a paper about the research.

Deep-image classifiers are widely used. In addition to medical diagnostics and the promotion of autonomous vehicle technology, there are use cases in security, gaming and even an app that tells you if something is a hot dog, because sometimes we need reassurance. . The technique discussed works by processing individual pixels from tons of pre-labeled images for the network to “learn”.

Image classification is difficult, because machine-learning models have the ability to capture these redundant subtle signals. Then, when image classifiers are trained on datasets such as ImageNet, they can reliably make reliable predictions based on those signals.

Although these redundant signals can lead to model fragility in the real world, the signals are actually valid in the dataset, meaning that overexpression cannot be diagnosed using specific evaluation methods based on that accuracy.

In order to find the rationale for predicting the model on a particular input, the methods in the current study start with the whole image and repeatedly ask, what can I remove from this image? Essentially, it keeps covering the image until you have the tiniest piece left that still makes a convincing decision.

For this, it may also be possible to use these methods as a kind of validation criteria. For example, if you have an autonomous driving car that uses a trained machine-learning method to recognize stop signs, you can test that method by identifying the smallest input subset that constitutes a stop sign. can. If it contains a tree branch, a particular time of day, or something that isn’t a stop sign, you may be concerned that the car may stop at a place it’s not supposed to.

While it may seem that the model is the likely culprit here, the dataset is more likely to blame. “The question is how can we modify the dataset in a way that allows the model to be trained to more closely mimic how a human would think of classifying images and, therefore, hopefully enable these real Make better generalizations to world scenarios, such as autonomous driving and medical diagnostics, so that this redundant behavior does not occur in the model,” Carter says.

This may mean creating datasets in a more controlled environment. Currently, it is only images extracted from the public domain that are then classified. But if you want to do object recognition, for example, it may be necessary to train the model with objects with a non-informative background.

This work was supported by Schmidt Futures and the National Institutes of Health. Carter co-authored the paper with Siddharth Jain and Jonas Muller, scientists at Amazon, and David Gifford, a professor at MIT. They are presenting the work at the 2021 conference on Neural Information Processing Systems.

Read More

As artificial intelligence makes it easier to create hyper-realistic digital characters, much of the conversation around these tools has focused on deceptive and potentially dangerous deepfake content. But technology can also be used for positive purposes—to revive Albert Einstein to teach a physics class, talk through a career change with your old self, or talk to people while preserving face-to-face communication. to anonymity.

To encourage the technology’s positive potential, MIT Media Lab researchers at the University of California, Santa Barbara and Osaka University, and their colleagues have compiled an open-source, easy-to-use character generation pipeline that integrates facial expressions, AI for voice Combines models. and can be used to speed up and create a variety of audio and video outputs.

The pipeline marks the resulting output with a traceable, as well as human-readable, watermark to distinguish it from authentic video content and to show how it originated – to help prevent its malicious use. an extra for .

By making this pipeline readily available, the researchers hope to inspire teachers, students and health care workers to explore how such tools can help them in their respective fields. If more students, teachers, health care workers and physicians have the chance to create and use these characters, the results could improve health and wellness and contribute to personalized learning, the researchers write. nature machine intelligence,

Video thumbnails

play video

AI-generated characters can be used for positive purposes such as enhancing educational content, maintaining confidentiality in sensitive conversations without erasing non-verbal cues, and allowing users to interact with animated characters adaptable in potentially stressful situations . Video: Jimmy Day / MIT Media Lab

“It will be a truly strange world when AI and humans start sharing identities. This paper does an incredible job of thought leadership, exploring what is possible with AI-generated characters in domains ranging from education to health to close relationships.” Mapping its place provides a solid roadmap on how to avoid the ethical challenges surrounding privacy and misrepresentation, says Jeremy Belenson, founding director of the Stanford Virtual Human Interaction Lab, who was not involved with the study. .

Although the world mostly knows the technology from deepfakes, “we see its potential as a tool for creative expression,” says the paper’s first author Pat Patranutaporn, a PhD student in media technology professor Patty Mays’ Fluid Interfaces Research Group. .

Other authors on the paper include Mays; Fluid Interfaces Master’s student Waldemar Daenerys and PhD student Joan Leong; Media Lab research scientist Dan Novi; Osaka University assistant professor Parinya Punapongsanan; and University of California at Santa Barbara assistant professor Misha Sara.

deep truth and deep learning

Generative Adversarial Networks, or GANs, a combination of two neural networks that compete against each other, have made it easy to create photorealistic images, clone voices, and animate faces. Pataranutaporn, along with Danry, first explored his possibilities in a project called Machinoia, where he produced several alternative representations of himself – as a child, as an old man, as a woman – from different perspectives. To self-diagnose life choices. He says the unusually deep experience made him aware of his “journey as a person.” “It was deeply true—using your own data to uncover something about yourself that you’ve never thought of before.”

Researchers say self-exploration is one of the positive applications of AI-generated characters. For example, experiments show that these characters can make students more enthusiastic about learning and improve cognitive task performance. As a complement to traditional instruction, Pataranutaporn explains that the technique provides a way for instruction to be “personalized to your interest, your idols, your context, and can be changed over time”.

For example, researchers at MIT used their pipeline to create a synthetic version of Johann Sebastian Bach, in a live conversation with renowned cellist Yo Yo Ma in Media Lab Professor Todd Machovar’s Music Interfaces class – students and students. For the happiness of both.

Other applications may include characters that help provide therapy to reduce a growing shortage of mental health professionals and reach the estimated 44 percent of Americans with mental health issues who never receive counseling, or AI-generated content that provides exposure therapy to people with social anxiety. , In a related use case, the technology can be used to anonymize faces in video while preserving facial expressions and emotions, which can be useful for sessions where people can personally access sensitive information such as health and Want to share trauma experiences, or for whistleblower and witness accounts.

But there are also more artistic and playful use cases. In this fall’s experiments in a deepfake class, led by Mays and research associate Roy Shilkrot, students used the technique to animate figures in a historical Chinese painting and create a dating “breakup simulator,” among other projects.

Legal and ethical challenges

The many applications of AI-generated characters raise legal and ethical issues that should be discussed as the technology develops, the researchers note in their paper. For example, how do we decide who has the rights to digitally recreate a historical character? Who will be held legally liable if an AI clone of a celebrity promotes harmful behavior online? And is there any danger that we would prefer to interact with synthetic characters over humans?

“One of our goals with this research is to raise awareness of what is possible, ask questions, and start a public conversation about how this technology can be used ethically for social benefit. Potential for harm What technical, legal, policy and educational actions can we take to promote positive use cases while reducing Mess tells.

By sharing the technology widely, explicitly labeling it as synthesized, Pataranutaporn says, “We hope to encourage more creative and positive use cases, as well as the potential benefits of the technology to people.” and educate about the pitfalls.

Read More

The annual MIT SHASS Research Fund supports research in the institute’s humanities, arts, and social sciences areas that show promise of making a significant contribution to the proposed field of activity. Congratulations to the six recipients for 2022:

Dwai Banerjee, associate professor in the Program in Science, Technology and Society, will use his award funding for work on the book project, “A Counter History of Computing in India”. Although India supplies cheap technical labor to the rest of the world, the country lags behind in basic computing education, research and development. Banerjee will trace major changes in the relationship between the Indian state and computer science since the 1950s.

Tristan Brown, assistant professor of history, will use the prize to collect data on the spread of Islam in China during the Ming and Qing dynasties. This research could potentially reveal that the Chinese state was often heavily involved in building and supporting Islamic institutions during these periods. The final product will be a website where scholars can examine and engage with data-driven maps showing the historical location of mosques across China and the individuals associated with their construction.

The award will enable history professor Eric Goldberg to take a research trip to Berlin in the summer of 2022 for his new book project, “The Carolingians and the Vikings: Contact, Conflict, and Accommodation, 751–987”. The book will work against the stereotypes of Scandinavian raiders to provide a valuable new perspective on the history of the ninth and tenth centuries in Europe.

Funding by Professor Nick Montfort in the Comparative Media Studies/Writing Program will support two projects. The first is 101 Basic Poems, a literary and media effort that will feature miniature computer programs for classics that will re-work and comment on the poetry and art of the past century. The second is the development of Curveship, a programming platform for creating variable narratives with potential creative, learning and research uses.

Tanalis Padilla, professor of history, will use her prize fund to organize short-term exploratory research trips to Chile, Bolivia and Mexico for her new book on the effects of Cuban medical internationalism in Latin America. Since the 1960s, thousands of Cuban medical professionals were sent to Asia, Africa and Latin America, upset by the island’s long-standing symbolism of anti-imperialistism. Padilla will examine the local effects of these international political transactions.

The award will assist Ken Urban, Senior Lecturer in the Music and Theater Arts Section, in creating a new multimedia play, “The Conquered”, directed by Jay Scheib, a 1949 professor of theater arts. The production will bring six actors to campus along with a design team for a two-week intensive workshop in the new year using MIT’s technical resources to produce the video and audio elements for the play.

Read More