This week’s list of data news highlights covers October 27 – November 2, 2018, and includes articles about Google’s challenge to use AI for social good and a legal AI application better than human lawyers at reviewing non-disclosure agreements.
1. Using AI to Detect False Robbery Reports
Researchers at Cardiff University in Wales and Charles III University of Madrid have created a tool called VeriPol that uses machine learning to detect false robbery reports. The researchers trained the machine-learning model on more than 1,000 police robbery reports from the Spanish National Police to identify several factors that could indicate whether a report is false, such as short length and focus on property lost instead of details concerning the robbery itself. A pilot study of VeriPol found that it detected significantly more false reports than human officers.
Automakers are increasingly experimenting with connected cars and traffic signals that can communicate with each other to reduce congestion, emissions, and accidents. For example, Volkswagen has begun testing a smart traffic light system in Germany that transmits data to connected cars about when traffic lights will change colors to reduce the amount of unnecessary starting and stopping by cars. In addition, Honda has tested a “smart intersection” in Ohio where cameras provide a 360-degree-birds-eye view of cars and pedestrians. The system sent audio and visual alerts to connected cars about potential issues occurring out of their view, such as pedestrians crossing the street or an oncoming emergency vehicle.
3. Spurring AI For Social Good
Google has launched a global competition called the Google AI Impact Challenge to find the best proposals for how AI could address the world’s biggest social, humanitarian, and environmental problems. Winning proposals will receive a total of $25 million to help develop their ideas, access to Google’s cloud resources, and technical support from data science nonprofit DataKind.
4. Waymo to Test Fully Driverless Cars in California
California has given Waymo, the autonomous vehicle subsidiary of Google parent Alphabet, permission to test fully driverless cars without humans in the driver’s seat. The company will be able test its vehicles on streets, rural roads, and highways with up to 65 miles-per-hour speed limits. Waymo will also restrict its testing to cities within Silicon Valley, where its vehicles have logged a significant portion of the over 8 million miles they have driven on public roads.
5. AI Outperforms Lawyers in Legal Analysis
LawGeex, a firm that uses AI to automate the review and approval of contracts, created an AI system that outperformed 20 human lawyers in identifying risks in non-disclosure agreements (NDAs). During a test, the lawyers and LawGeex’s AI were each given five NDAs to review. The lawyers took an average of 92 minutes to review the contracts and also had a mean accuracy score of 85 percent. LawGeex’s AI, however, achieved 94 percent accuracy and only took 26 seconds to review all the contracts.
University of Western Australia researchers have trained an AI system initially designed by Facebook for facial recognition to detect galaxies in deep space. The new system, Classifying Radio sources Automatically with Neural networks (CLARAN), scans images taken by radio telescopes to spot radio galaxies, which are galaxies that emit strong radio waves from black holes at their centers. These waves emitted by the galaxies can stray far from their source, however, making it difficult for traditional computer programs to detect the location of the galaxy. CLARAN accurately detects 90 percent of the galaxies.
7. Using AI to Secure Border Crossing Points
Hungary, Latvia, and Greece are using an AI system called iBorderCtrl as part of a six-month pilot to increase the efficiency and accuracy of border checks. The system analyzes 38 different subtle gestures travelers’ faces can make while asking them questions such as “What’s in your suitcase?” Passengers then receiver a QR code to cross the border or the system refers them to a human border patrol agent if they fail the test. The pilot will not, however, prevent anyone from crossing the border in its current state.
Researchers at the Massachusetts Institute of Technology have created an AI system called AI Physicist that can determine the laws of physics in artificial worlds. For example, AI Physicist can detect the force of gravity in a simulated but complex world by analyzing the variables that may cause a ball to fall. The researchers trained the AI system to be able to perform such a task by having it produce multiple theories for why the ball fell and having it prioritize simpler theories. AI Physicist also remembers its past predictions for the laws that govern a certain world, which helps it learn faster and use less data to detect the relevant law of physics in a world.
Startups are increasingly using algorithms to detect fake images. For example, Serelay, a U.K. startup, evaluates approximately 100 mathematical values about each image its users upload to create a digital fingerprint that it claims can be used to determine if even a single pixel was edited. U.S.-based startup Truepic uses a similar approach, but also stores its own copies of users’ images and metadata on a blockchain that it can use to authenticate potentially altered photos.
Researchers at the University of Glasgow have developed a machine learning system to detect the origins of different viruses. The researchers trained their system on data about 500 RNA viruses and used machine learning to spot genetic signals that can indicate where a virus originated, such as in bats, and how it spreads, including through mosquitoes, sandflies, or ticks. The system can predict a virus’ original host with 72 percent accuracy and determine how it spreads with 90 percent accuracy, which could help health authorities better respond to dangerous new diseases.