U.S. Intelligence Building System to Track Mass Movement of People Around the World
The Pentagon's intelligence branch is developing new tech to help it track the mass movement of people around the globe and flag anomalies."
The project is called the Hidden Activity Signal and Trajectory Anomaly Characterization (HAYSTAC) program and it aims to establish normal' movement models across times, locations, and populations and determine what makes an activity atypical," according to a press release from the Office of the Director of National Intelligence (DNI).
HAYSTAC will be run by the DNI's Intelligence Advanced Research Projects Activity (IARPA). It's kind of like DARPA, the Pentagon's blue-sky research department, but with a focus on intelligence projects. According to the agency, the project will analyze data from internet-connected devices and smart city" sensors using AI.
An ever-increasing amount of geospatial data is created every day," Jack Cooper, HAYSTAC's program manager, said in a press release about the project. With HAYSTAC, we have the opportunity to leverage machine learning and advances in artificial intelligence to understand mobility patterns with exceptional clarity. The more robustly we can model normal movements, the more sharply we can identify what is out of the ordinary and foresee a possible emergency."
In a talk about HAYSTAC on the program's website, Cooper explained it a little more. He gave the example of watching traffic patterns to help predict a terrorist attack. When you're going down the highway during rush hour you expect there to be lots of vehicles there because you've made that trip many times," he said in the video. When you're going down the highway at rush hour and there's nobody there, you have a sense of this doesn't make sense, this is anomalous.' We can use this type of information to predict how things will likely go."
Cooper also mentioned privacy, or rather a lack of it, as a motivation for thinking about human movement. Today you might think that privacy means going to live off the grid in the middle of nowhere," he said. That's just not realistic in today's environment. Sensors are cheap. Everyobodys got one. There's no such thing as living off the grid."
HAYSTAC's landing page on IARPA's website includes several proposals from companies looking to be part of the project as well as a March 22 briefing detailing existing Pentagon projects from defense contractor Assured Information Security (AIS) that HAYSTAC might be interested in.
In one project, AIS simulated a cyber attack with 104 individuals and watched the way they moved. Devices included traditional desktop systems, laptops, tablets, and mobile platforms. Modalities included accelerometer and gyroscope, keystroke data, mouse data, touchscreen interactions, and other information," the firm said. The technology tracks users through biometric features, including keystroke biometrics, mouse movement behavior, and gait detection."
In another project, called GANSpoofer, AIS used an AI model called a generative adversarial network (GAN) to make fake users that could defeat a biometric scanner. GANs have been used to create hyper realistic photos of people and animals that don't exist. We've shown that we can both detect the unique anomalies associated with an individual's biometric behaviors and use this information to transform data into, not only realistic patterns at a population level, but patterns specific to that individual," AIS said.
The defense contractor also claimed to have developed a learning model that could detect symptoms of a traumatic brain injury in a soldier just by watching how they moved their smartphone. According to its research, the placement of the phone in and around the body and the accelerometer and gyroscope data from the device could help it predict certain diseases and injuries with more than 90 percent of accuracy.
HAYSTAC will pursue these, and other similar projects, over the next four years. As the HAYSTAC systems mature, they will be evaluated based on probability of detection and false alarm performance in creating relevant alerts, ultimately seeking to identify 80% of anomalous activity while generating normal activity that is only 10% detectable," the press release said.