This week’s list of data news highlights covers November 23-November 29, 2019, and includes articles about using AI to detect and respond to hate speech online and using machine learning to detect malware.
Researchers from the University of California, Santa Barbara, and Intel have developed an Internet bot that uses AI to detect and respond to hate speech posted online. The researchers trained the system on thousands of hateful posts on Reddit and Gab and moderators’ responses. The system can detect hate speech with 80 to 90 percent accuracy. Its responses are effective roughly 40 percent of the time, compared to 70 percent for human moderators.
Heliogen, a clean energy firm based in the United States, has used AI and mirrors to generate heat above 1,000 degrees Celsius. At this extreme temperature, solar energy can help make cement, steel, and glass. Heliogen used computer vision software and other technology to train the mirrors to reflect solar beams to a single point, creating the extreme heat.
The UK’s National Health Service is piloting the use of an AI system that provides medical patients a complexity score, which is similar to a credit rating for a patient’s health. A patient’s underlying health conditions, such as high blood pressure, affect the score, and a high score indicates a patient will likely need hospital admittance. Roughly 100 patients are taking part in the pilot, which NHS hopes can help doctors identify patients that need to reduce their health risks.
Anheuser-Busch InBev, the maker of Budweiser, is using machine learning to decide which small merchants in Brazil to provide short-term financing to for beer purchases. Many of the merchants struggle to receive financing and often pay Inbev with cash. The system decides which merchants will likely repay the loan by analyzing data such as the type of business, location, and local demographics.
Researchers from Neurobotics, a Russian research firm, and the Moscow Institute of Physics and Technology have developed an AI system that can accurately guess the type of video an individual is watching based on their brainwaves. The researchers trained the system on data concerning the brainwaves of people watching different types of video clips. The system correctly categorizes the type of video with nearly 90 percent accuracy, providing tags such as waterfalls, extreme sports, or human faces.
Researchers from Tianjin University and Beijing University of Posts and Telecommunications in China have developed an AI system that can retrieve images of an object or concept based on a sketch. This task has been difficult to automate because human sketches are often deformed or abstract. The researchers trained the system on thousands of natural images and sketches.
Researchers led by an individual from the California Institute of Technology have developed a sensor that can detect the level of metabolites and nutrients in an individual’s body by analyzing their sweat. The sensor can detect levels of uric acid, which is associated with painful joint problems, and tyrosine, an indicator of liver disease and eating disorders, as well as respiratory rate and heart rate. The sensor could help doctors continuously monitor the health of patients with illnesses such as cardiovascular disease, kidney disease, and diabetes.
Researchers led by an individual from Case Western Reserve University in the United States have developed an AI system that can automatically identify changes in lung cancer, which can help doctors determine the success of their treatment approach. The researchers trained the system on the CT scans of fifty patients, finding the system could identify changes in the size, texture, volume, and shape of a cancer lesion. Some doctors only use size to track cancerous growths.
Researchers from Brunel University in London have developed a soil-monitoring system that can help farmers efficiently use their resources. Farmers can plant the system’s pods into the ground, and the pods’ sensors will collect data on soil and air temperature, soil moisture, and air humidity. The system’s algorithms will use this data to predict future soil temperature and moisture levels during the day, which can help farmers best decide when to water crops.
Microsoft used machine learning to detect and protect devices from malware called Dexphot that infected devices to mine cryptocurrency. Dexphot used encryption, randomized file names, and other obfuscation techniques to hide its installation and activity. Microsoft’s behavior-based machine learning algorithms analyzed a device’s activity, instead of scanning for known infected files. This process helped Microsoft’s system identify and block malicious files at the beginning of the attempted infection.