A clever shield against photo fakery

Remember that selfie you posted last week? There’s currently nothing stopping someone from taking it and editing it with AI—and it might be impossible to prove that the resulting image is fake. 

The good news is that a new tool created by researchers at MIT could prevent this.

The tool, called PhotoGuard, works like a protective shield by altering photos in tiny ways that are invisible to the human eye but prevent them from being manipulated. If someone tries to use an editing app based on a generative AI model such as Stable Diffusion to manipulate an image that has been “immunized” by PhotoGuard, the result will look unrealistic or warped. 

Right now, “anyone can take our image, modify it however they want, put us in very bad-looking situations, and blackmail us,” says Hadi Salman, a PhD student at MIT who contributed to the research. PhotoGuard is “an attempt to solve the problem of our images being manipulated maliciously by these models,” says Salman. The tool could, for example, help prevent women’s selfies from being made into nonconsensual deepfake pornography.

The MIT team used two different techniques to stop images from being edited using Stable Diffusion. In the first, PhotoGuard adds imperceptible signals to the image so that the AI model interprets it as something else, such as a block of pure gray. In the second, it disrupts the way the AI models generate images, essentially by encoding them with secret signals that alter how they’re processed by the model, so any edited image looks like that gray block. For now, the technique works reliably only on Stable Diffusion, an open-source image generation model. 

In theory, people could apply this protective shield to their images before they upload them online, says Aleksander Madry, SM ’09, PhD ’11, a professor of electrical engineering and computer science who contributed to the research. But a more effective approach, he adds, would be for tech companies to add it to images that people upload into their platforms automatically—though it’s an arms race, because new AI models that might be able to override any new protections are always coming out.

Do-it-yourself breast ultrasound

Early detection is key to surviving breast cancer, but tumors that develop in between routine mammograms—known as interval cancers—tend to be especially aggressive. A wearable ultrasound device devised by MIT researchers could help detect such tumors when they are still in early stages. 

The device can be attached to a specialized bra to let an ultrasound tracker image the breast tissue from different angles. In their study, the researchers showed that they could obtain images comparable in resolution to those done at medical imaging centers.

“We changed the form factor of the ultrasound technology so that it can be used in your home. It’s portable and easy to use, and provides real-time, user-friendly monitoring of breast tissue,” says Canan Dagdeviren, an associate professor in MIT’s Media Lab and the senior author of the study. Dagdeviren drew up the first rough schematic of the device as an MIT postdoc at the bedside of her aunt Fatma Caliskanoglu, who (despite regular screenings) died of breast cancer six months after receiving a diagnosis at age 49. 

“My goal is to target the people who are most likely to develop interval cancer,” says Dagdeviren, whose research group specializes in developing wearable electronic devices. “With more frequent screening, our goal is to increase the survival rate to up to 98%.”

To make her vision of a diagnostic bra a reality, Dagdeviren designed a scanner that’s based on the same kind of technology used in medical imaging centers but can be much smaller thanks to the use of a novel piezoelectric material.

To make the device wearable, the researchers designed a flexible, 3D-printed patch, which has honeycomb-like openings. Using magnets, this patch can be attached to a bra with openings that allow the ultrasound scanner to contact the skin. The scanner fits inside a small tracker that can be moved to six different positions, and it can be rotated to take images from different angles. It does not require any special expertise to operate.

Working with the MIT Center for Clinical and Translational Research, the researchers tested their device on a 71-year-old woman with a history of breast cysts and succeeded in detecting cysts as small as 0.3 centimeters in diameter—the size of early-stage tumors. The patch can be used over and over, and the researchers envision that it could be used at home by people at high risk for breast cancer. It could also help diagnose cancer in people who don’t have regular access to screening.

Today, the researchers have to connect the device to a traditional ultrasound machine to view the images. But they are working to develop a smartphone-size version of the system used to read ultrasound scans, so that patients wouldn’t have to visit an imaging center.

Eventually, artificial intelligence might be used to analyze how the images change over time, which could be more accurate than relying on the assessment of a radiologist comparing images taken years apart. The researchers also plan to explore adapting the ultrasound technology to other parts of the body. 

Low-power underwater communication

MIT researchers have demonstrated a technology that can transmit underwater signals much farther than existing methods, using only about a millionth as much power. 

The system is based on backscatter communication, a method of encoding data in sound waves that are reflected from the sound source, or interrogator, back to a receiver in the same location. The underwater backscatter device uses nodes made from piezoelectric materials, which produce an electrical signal when a mechanical force—including sound waves—is applied. The nodes use that charge to scatter some of the acoustic energy back to the receiver.

To make the system more efficient, the researchers used a 70-year-old technology called a Van Atta array, in which symmetric pairs of antennas are connected so that energy is reflected back in the direction it came from, and placed a transformer between pairs of connected nodes. It can be used with data-collecting sensors and send data to a ship or onshore station.

In tests, the device achieved ranges of 300 meters, more than 15 times longer than previously demonstrated—and a model suggests that kilometer-scale ranges are possible. That could make it suitable for things like coastal hurricane prediction and climate modeling.

“There are still a few interesting technical challenges to address, but there is a clear path from where we are now to deployment,” says Fadel Adib, director of the Signal Kinetics group in the MIT Media Lab and the senior author of two papers on the work.

Energy-storing concrete

A supercapacitor made from cement and carbon black (a conductive material resembling fine charcoal) could form the basis for a low-cost way to store energy from renewable sources, according to MIT researchers.

The amount of power a capacitor can store depends on the total surface area of its conductive plates. Professors Franz-Josef Ulm, Admir Masic, and Yang Shao-Horn and colleagues found that if carbon black is introduced into a mixture with cement powder and water, the water naturally forms a branching network of openings when the resulting concrete cures—and the carbon migrates into that network to make wire-like structures, yielding a conductive material with an extremely large internal surface area. 

Two electrodes made by soaking this material in a standard electrolyte, separated by a thin space or an insulating layer, form a very powerful supercapacitor, the researchers found. A cube about 3.5 meters across could store about 10 kilowatt-hours.

The simple technology could eventually be incorporated into the concrete foundation of a house, where it could store a day’s worth of energy. The researchers also envision a roadway that could provide contactless recharging for electric cars as they travel.

It’s “a new way of looking toward the future of concrete as part of the energy transition,” Ulm says. 

AI-tocracy

It’s often believed that authoritarian governments resist technical innovation in a way that ultimately weakens them both politically and economically. But a more complicated story emerges from a new study on how China has embraced AI-driven facial recognition as a tool of repression. 

“What we found is that in regions of China where there is more unrest, that leads to greater government procurement of facial-recognition AI,” says coauthor Martin Beraja, an MIT economist. Not only has use of the technology apparently worked to suppress dissent, but it has spurred software development. The scholars call this mutually reinforcing situation an “AI-tocracy.” 

In fact, they found, firms that were granted a government contract for facial-recognition technologies produce about 49% more software products in the two years after gaining the contract than before. “We examine if this leads to greater innovation by facial-recognition AI firms, and indeed it does,” Beraja says.

Adding it all up, the case of China indicates how autocratic governments can potentially find their political power enhanced, rather than upended, when they harness technological advances—and even generate more economic growth than they would have otherwise.

The scholars are now studying the extent to which China is exporting facial-recognition tech around the world—highlighting a mechanism through which government repression could grow globally.