10 Bits: the Data News Hotlist
This week’s list of data news highlights covers June 24-30, 2017, and includes articles about how Unilever is using algorithms in its hiring process and an AI system that can predict a neighborhoods wealth from space.
London-based startup RAVN, working with the U.K. Serious Fraud Office (SFO), has developed an AI system capable of analyzing unstructured documents dramatically faster than humans to help SFO better detect fraud. In an investigation into potential corruption at Rolls-Royce, SFO had amassed 30 million potentially relevant documents that human barristers could sort through at a rate of 3,000 a day. SFO began using RAVN’s AI system in January 2016, and it was able to process 600,000 documents per day, with fewer errors and for less money than barristers, and resolving the investigation by January 2017.
Unilever has replaced several steps in its hiring process with algorithms designed to weed out between 60 and 80 percent of applicants before they ever interact with a human. First, algorithms analyze applicants’ resumes for entry-level jobs and internships and identifies candidates that meet the role’s requirements. Then, algorithms evaluate candidates as they complete a series of online games designed to gauge traits such as short-term memory and concentration. Top performing applicants after these two rounds then submit a video interview through HireVue, which uses AI to evaluate factors such as word choice and attitude in interviews. Unilever says that this process has made the hiring process more accurate, with 80 percent of applicants who pass these rounds of algorithmic evaluation receiving job offers.
Microsoft has developed an analytics program called the Sports Performance Platform to aggregate and analyze data about athlete performance and provide recommendations for coaches, such as identifying if a player is at risk of getting injured. The Sports Performance Platform incorporates fitness tracker data, player recovery times, training regimens, and other factors and uses machine learning to identify relationships between this data and player performance. Microsoft has partnered with teams such as Seattle Reign FC in the United States and Real Sociedad in Spain to pilot the platform.
Pittsburgh startup Marinus Analytics has developed a machine learning facial recognition tool called FaceSearch that can help law enforcement track victims advertised on the deep web by human traffickers. Marinus Analytics’ software uses AI to sort through posts on the deep web and analyze publicly available data to identify patterns that can help catch human traffickers, such as cell phone numbers listed in online sex ads. With FaceSearch, authorities can upload a photograph of a missing person and the system will flag posts with photographs that match his or her likeness.
The Wimbledon tennis tournament has partnered with IBM to use its Watson cognitive computing platform to help fans navigate the tournament, generate video highlights, and with a new tool called SlamTracker with Cognitive Keys to the Match, flag matches it predicts will be the most exciting. SlamTracker works by analyzing a metric called competitive margin, which compares players’ ratios of forced and unforced errors—the smaller the competitive margin, the more likely Watson will predict the matchup to be exciting.
Researchers at Carnegie Mellon University and data visualization firm Stamen Design have developed an AI system called Penny that can analyze satellite imagery of a neighborhood and predict the level of median income in the area. Penny combines income data from the U.S. Census Bureau with satellite imagery of New York City and St. Louis and identifies characteristics that typify low, medium, and high median income areas, such as proximity to freeways, the amount of green space, and the density of solar panels. Users can also observe how the addition of different characteristics, such as a helipad or a baseball diamond, would alter Penny’s assessment of an area’s income levels.
Automatic teller machines (ATMs) for financial services company China UnionPay in Macau are now using facial recognition to authenticate and track transactions and curb money laundering. Macau’s large casino industry and its separate financial system from China makes it a popular location for money laundering, as people can withdraw money from Chinese bank accounts for use in betting but then exchange this money for foreign currency to funnel it out of the country instead. China UnionPay customers now must scan their faces at ATMs in Macau to authenticate a transaction, which can prevent people from making unauthorized withdrawals from the accounts of friends and family.
Researchers at Rutgers University and Facebook’s AI Research lab (FAIR) have developed an AI system that can generate artistic images in entirely new styles. The system consists of one neural network generating an artificial image and another “discriminating” neural network evaluating it to determine if it has desired qualities, such as characteristics that would classify the image as cubism or impressionism. The researchers trained their system on 81,500 paintings to teach it to distinguish between artistic styles and replicate them to varying degrees. By having the discriminating network desire results that did not comport with any existing styles of art, the system was able to generate images in novel styles.
Instagram has developed an AI tool called DeepText that allows users to automatically block offensive comments. DeepText uses a language classification system based on a concept called word embeddings which allows it to interpret words in context, similar to how the human brain processes language. This enables it detect objectionable content that uses benign words. Instagram originally developed DeepText to fight spam, but the company has now trained it to identify and block offensive comments in nine different languages.
Researchers at the Massachusetts Institute of Technology have developed a method to allow an AI system to link what it interprets from one sense to another, such as linking the sound of an ambulance to the sight of one. This allows an AI system to gain a more holistic understanding of objects and language and allow a system to learn to associate information across different mediums without explicit training.