10 Bits: the Data News Hotlist
This week’s list of data news highlights covers August 6-12, 2016 and includes articles about how IBM’s Watson saved a life by catching a misdiagnosis and the U.S. governments new policy to increase the use of open source code.
The U.S. Defense Advanced Research Project Agency (DARPA) has launched an initiative called Explainable AI (XAI) to develop machine learning technologies that can create easy-to-interpret models of how they function so that humans can more reliably understand, trust, and manage AI systems. XAI will focus on two challenges common to a wide array of potential use cases: using machine learning to classify interesting events in heterogenous multimedia data, and using machine learning to develop policies governing how an autonomous system makes decisions in simulated missions. After XAI concludes, DARPA will release the XAI toolkit library for additional research and for integration into commercial applications.
Google has partnered with Stanford University School of Medicine to advance genomics research by creating the Clinical Genomics Service, which will improve how Stanford Medicine integrates genomic data into routine care for patients with cancer. Google will provide cloud hosting for the school’s genomic data stores so doctors can make better use of it to improve cancer diagnostics and treatment, and try to prevent patients from getting sick in the first place. Stanford Medicine will also work with Google to identify opportunities to use machine learning to assist radiologists in analyzing medical scans.
Doctors at the University of Tokyo’s Institute of Medical Science used IBM’s cognitive computing platform, Watson, to correctly diagnose and save a woman suffering from a rare form of leukemia after the doctors misdiagnosed her. Doctors had originally diagnosed the 60-year-old patient with acute myeloid leukemia, but she failed to respond to treatments and her condition worsened. The doctors then used Watson to analyze the patient’s health records and sift through 20 million clinical oncology studies to identify what was wrong. Watson deduced that the woman had a very rare form of leukemia and recommended a change in treatment, marking the first time AI has saved someone’s life in Japan.
The White House has finalized the Federal Source Code policy, requiring that when federal agencies build or contract out custom source code, they must also make that code freely available to other agencies for modification and reuse. Federal agencies must also now evaluate whether or not there is existing federal code that they can modify to meet their needs before purchasing new software. The Federal Source Code policy also establishes a pilot program that will make a portion of custom-developed federal agency code freely available to the public.
Researchers at University College London have developed an algorithm that can analyze a person’s handwriting and produce original text in the same style with a high level of accuracy. Handwriting-style fonts already exist, but they lack the irregularities and messy details that every human exhibits in their handwriting. The researchers’ algorithm classifies unique aspects of a person’s handwriting based on a short written sample and then applies those aspects to new text, producing almost indistinguishably similar writing. This technique could be used to give standardized documents a more personal feel or make computer-generated messages appear more genuine, while still leaving enough clues for close examination to reveal that the writing style is computer-generated, thereby preventing fraud.
Facebook has partnered with Summit Public Schools, a nonprofit charter school network based in California, to implement a student-directed learning management system this fall in nearly 120 schools. The Summit Personalized Learning Platform will provide students with a comprehensive view of all of their responsibilities for their academic year and create customizable lesson plans for each student to progress at their own pace based on their individual skill levels. Students will be able to complete specific assignments in a variety of ways, as well as work through their lesson plans independently of other students. This will free up teachers to focus on one-on-one instruction for students as they need it, rather than teaching to the whole class simultaneously. In a 19-school pilot last year, the personalized platform resulted in a significant improvement in the percentage of students reading at or above their grade level compared to classes that did not use the platform.
Researchers at Harvard University, the Massachusetts Institute of Technology, and Samsung have developed a machine-learning system called the Molecular Space Shuttle that can quickly analyze molecules and evaluate how effective they would be in organic light-emitting diode (OLED) screens, which are used in televisions, smartphones, and other electronics. OLED screens produce images by manipulating particles that can emit green, red, and blue light, however particles that can emit blue light efficiently are very rare and workarounds can be expensive. The researchers built a database of 1.6 million particles that could potentially emit blue light and applied machine learning algorithms to predict which particles would be best-suited for the job. The system identified 2,500 promising candidates that researchers can now more closely scrutinize and potentially use to build more efficient and cost-effective OLED screens.
The Aircraft Owners and Pilots Association, a U.S.-based aviation association, has reported that the number of fatal crashes in small private planes has fallen to the lowest level in decades thanks largely to the proliferation of data-driven apps and tools. Collision avoidance systems, equipment sensors that can warn pilots they are about to lose control, and mobile apps that provide pilots with granular and up-to-date weather data have helped make small planes safer than ever, with only 1.03 fatal crashes per 100,000 flight hours in 2015.
The U.S General Services Administration (GSA) has completed its first auction for IT procurement by using automated bots that competed against each other to make the most competitive bid. The GSA team focused on improving the government’s digital services, called 18F, developed an application programming interface (API) to allow companies to develop automated bidding tools for contracts. Normally, 18F would create an auction for a contract and a small number of bidders—three to five, on average—would make bids manually (often more than one bid per bidder), totalling an average of five to seven bids for an auction, before GSA selected a winner. Using the API, this most recent auction received 70 bids from seven unique bidders, with the majority of bids coming from automated bots. By making it easier for any company to bid on a contract, 18F hopes to make the bidding process more competitive and cost-effective.
An increasing number of companies are using the Internet of Things to measure how happy and productive their employees are, and they are using this data to improve management practices. For example, Hitachi since last year has been testing an algorithm that can infer employee happiness by analyzing data from sensors in employee badges that can indicate how much time during the work day an employee spends sitting, talking, walking, and performing other activities. Additionally, Bank of America piloted sensor-embedded badges on call center employees and analyzed this data to find that by allowing groups of employees to take breaks together, employee productivity increased by 10 percent.
Image: Matthew Rollings.