This week’s list of data news highlights covers March 4 – 10, 2017 and includes articles about an algorithm that could help judges make better decisions about a defendant’s flight risk, and DeepMind’s new plan to track health data with blockchain-style technology.
Irish Internet of Things company Moocall has developed a wearable device for cows that constantly monitors a pregnant cow’s activity and helps farmers be prepared to provide medical attention when it is ready to give birth. The device detects and transmits the direction a cow’s tail moves. Normally, cows move their tails back and forth to swat flies, but during pre-birth contractions, cows move their tails up and down with varying speed depending on the frequency of contractions. With this data, farmers can learn as soon as a cow is about to give birth and seek a veterinarian if the cow is likely to experience complications, which kill 110,000 calves and 50,000 cows every year.
Researchers at the Fraunhofer Institute for Telecommunications in Berlin have developed a method that can reverse-engineer decisions made by an artificial neural network to determine the factors that influenced its decision-making process. The method works by feeding images to a neural network and assigning numerical scores to individual pixels based on how important a neural network deemed them to be when trying to determine the contents of the image. This approach could help determine what went wrong if a neural network makes a mistake as well as improve neural network training by reducing the amount of data necessary to teach a network to identify specific objects.
Baidu has developed a text-to-speech AI system called Deep Voice capable of learning to produce speech in just a few hours without human input. Traditional text-to-speech systems rely on large pre-recorded databases of spoken words, whereas AI-based systems such as Deep Voice synthesize speech from text in real-time, which can be more versatile, though it can require a lot of training and be computationally demanding. Deep Voice uses deep learning to break down text into perceptually distinct units of sound called phenomes and then assembles these phenomes into spoken words.
Researchers from the U.S. National Bureau of Economic Research have developed an algorithm that can predict the flight risk of defendants more accurately than judges. The algorithm analyzes a defendant’s prior criminal history and historical data from hundreds of thousands of New York City court records to determine the likelihood a defendant would flee before his or her trial. The algorithm deliberately excludes demographic data from its risk predictions to limit potential racial bias from influencing its decisions. The researchers estimate that by applying their algorithm in New York City, it could reduce the population of people in jail awaiting trial by over 40 percent without increasing crime rates.
Google has developed an application programming interface (API) called Video Intelligence API that can recognize the contents of non-annotated videos. Image recognition algorithms already exist, but because videos are unstructured data, they typically require videos to be manually tagged with information about their contents for a computer to effectively interpret them. The Video Intelligence API allows users to search through collections of videos simply by searching for keywords of their contents.
DeepMind has announced it is developing a tool called Verifiable Data Audit that uses block-chain-style technology to track how patient data is used in a health system. Verifiable Data Audit will create a digital ledger that records every time a patient’s health data is used, allowing a health-care provider to monitor patient data use in real-time, automatically flag suspicious usage patterns, and better track health records across fragmented systems. DeepMind will test Verifiable Health Audit in London’s Royal Free Hospital.
Stanford student Joshua Browder has modified a chatbot he developed for a service called DoNotPay, which helped people overturn unjust parking fines, to provide legal advice to refugees seeking asylum in the United States and Canada. Browder worked with lawyers to help train DoNotPay to walk users through the asylum-claiming process by having a conversation with it over Facebook Messenger, and as more people interact with the chatbot, it will improve.
Electronic health records firm Epic has partnered with speech-recognition company Nuance to develop an AI-powered virtual assistant named Florence to help veterans make health-care appointments with the U.S. Department of Veterans Affairs (VA). Veterans can converse with Florence to schedule appointments and Florence will automatically flag a user’s health records for the VA to begin processing. By using AI to support natural voice communication, Florence can be particularly useful for users with impaired vision or motor skills.
IBM researchers have successfully created single-atom magnets capable of storing data. A hard drive consists of magnetized areas that can indicate a 1 or a 0 based on their polarity to represent data. The smaller a magnet is, the less stable its polarity is, so it can be challenging to make these magnetized areas smaller and thus making the hard drive have a larger capacity while ensuring they are reliable. The researchers were able to create a stable magnetized atom capable of storing two bits of data with a stable polarity, which is a dramatic improvement from commercially available storage which required 1 million atoms per bit, as well as the latest experimental storage, which requires 12 atoms per bit.
The Chinese government has announced that it is developing a national plan to accelerate the adoption of artificial intelligence. The plan will focus on spurring AI adoption in a broad array of sectors, including social welfare and environmental protection. Government funding will support AI deployment to promote economic growth and strengthen national security, as well as fundamental research.