This week’s list of data news highlights covers June 17-22, 2018, and includes articles about IBM’s new AI system that can debate humans and a robot that can help firefighters navigate through smoke.
Consumer genomics company 23andMe will provide its genetic analysis kits to help reunite hundreds of migrant families separated at the U.S. border. In recent weeks, U.S. immigration authorities separated thousands of children from their families to send to different shelters and detention facilities, making reuniting children with their parents difficult. 23andMe’s analysis will help match children with their parents now that the Trump administration has halted its family separation policy.
IBM has developed an AI system called Project Debater capable of having debates with humans on specific topics. Project Debater can formulate an initial argument about the issue, a rebuttal to a human response, and a closing argument. Rather than craft arguments based on an understanding of the topic, Project Debater scans text of arguments people have already made about the topic and Wikipedia to piece together new arguments.
Portland, Oregon has launched an initiative called the Traffic Sensor Safety Project to deploy 200 sensors on street lights on streets with the highest rates of fatal traffic accidents. The sensors will log and differentiate between cars, pedestrians, and cyclists that pass by, as well as log their speed. This data could allow Portland to make more informed decisions about traffic management and road safety.
Nvidia has developed an AI system that can convert regular videos into realistic-looking slow-motion video. Making slow-motion video typically requires capturing a large number of frames per second (fps) and slowing the footage down, but standard video equipment such as smartphone cameras do not have high enough frame rates, making the videos choppy when slowed down. The AI system analyzes sequential frames from 30 fps videos and generates intermediate frames by estimating what would come between two frames, creating 240 fps slow-motion video.
Researchers at Google have developed an AI system that analyzes electronic health records and can accurately predict patient outcomes, including how long a patient is likely to stay in a hospital, likelihood of readmission, and when patients are likely to die. The system can also analyze unstructured data from patient records, such as handwritten comments, to help it make predictions.
Conservation technology nonprofit Wild Me is developing open-source software called Wildbook to serve as a social-media style resource for wildlife researchers. Wildbook allows researchers to share images and data about wildlife and uses AI to identify the individual animals that appear in multiple images so that conservationists can track animals throughout their lifetimes and better estimate population sizes. Researchers can populate Wildbook with additional data about animals, such as where particular images were taken, what other animals were nearby, and environmental conditions.
Facebook has announced that it is using machine learning to automatically identify pages that deliberately spread hoaxes and fake news to make money, similar to how pages run by Macedonians shared fake news about the 2016 U.S. presidential election to drive clicks and collect ad revenue. While Facebook relies primarily on human fact checkers to identify fake news, its machine learning system can identify duplicated and modified versions of hoaxes that its fact checkers already debunked. For example, a fact-checker in France flagged a fake story about how pricking someone’s finger with a needle could save them from a stroke, and its machine learning identified 20 other domains and over 1,400 links repeating that hoax.
A team at Örebro University in Sweden has developed a robot named SmokeBot that relies on a system of sensors to navigate smoky areas and guide firefighters and search-and-rescue teams. SmokeBot uses gas sensors, radar, a thermal camera, and other technology to map and steer itself through smoke-filled environments.
Engineers at the Massachusetts Institute of Technology have developed a smart power outlet that uses an artificial neural network to track electricity usage and automatically switch off power in the event of an arc fault—a high-powered electrical discharge between conductors. Arc-fault detectors are common and help prevent electrical fires, however even when they use algorithms to differentiate between normal and abnormal electrical current patterns, known as arc signatures, they often trigger even when there is no safety risk. The engineers’ smart outlets instead use machine learning to analyze arc signatures in real time to differentiate between dangerous and safe arc faults with 99.95 percent accuracy. The outlets can also share data with each other to improve their ability to identify different arc signatures.
Adobe is developing machine learning tools that can automatically identify if an image has been digitally edited. The tools can identify three common types of image manipulation, including combining parts from two different images, duplicating an object already in the picture, and editing objects out of an image.
Image: Steve Morgan.