This week’s list of data news highlights covers October 21-27, 2017, and includes articles about a new partnership to accelerate the discovery of cancer drugs and a smartphone-powered ultrasound machine that can detect cancer.
The Greater London Authority is working with public sector innovation-focused charity Nesta and data science firm ASI Data Science to use data analytics to better police houses in multiple occupation (HMOs)—rental units with at least three unrelated tenants. Landlords of HMOs are required to keep them properly maintained and pay for regular safety inspections, but many do not disclose that their properties are HMOs to avoid these costs, which puts tenants at risk. ASI Data Science used housing data from the City of Westminster to build an algorithm that can identify an HMO with 500 percent greater accuracy than picking properties at random, which can help housing authorities investigate these cases much more productively.
Two U.S. national labs, GlaxoSmithKline, and the University of California, San Francisco, have launched a consortium called Accelerating Therapeutics for Opportunities in Medicine (ATOM) to use data sharing and high-performance computing to reduce the time it takes for a potential new cancer drug to become a clinical candidate from six years to one year. ATOM will develop an open platform to share biological data from public and private sources and GlaxoSmithKline will provide data about 2 million compounds that could be useful for drug discovery. ATOM will also use the supercomputing resources of the national labs to develop predictive modeling techniques that could help identify new drug targets.
AI firm Vicarious has developed an AI system capable of solving Completely Automated Public Turing Test To Tell Computers and Humans Apart (CAPTCHAs)—online tests that require people to interpret jumbled letters and numbers to prove they are human. CAPTCHAs are easy for humans to solve but challenge even advanced AI systems because slight variations in characters and extraneous information can make them difficult to interpret. Vicarious used a machine learning technique called a recursive cortical network that builds models of every character it sees and tries to identify them in images allowing for variation, and can solve common varieties of CAPTCHAs with between 57 and 66 percent accuracy.
Médecins Sans Frontières (MSF), also known as Doctors Without Borders, is piloting software called Dharma Platform to help aid workers responding to the crisis in Syria quickly collect, analyze, and share health data to boost relief efforts. Dharma Platform allows aid workers without data science training to easily compile information, create surveys, data visualizations, and other tools, and securely share data even without an Internet connection using Bluetooth. MSF has tested the software in several other crises with promising results.
The Information Technology Industry Council, a technology trade association, has developed industry principles for developing ethical artificial intelligence systems. The principles span three areas, including: industry’s responsibility to promote the responsible development and use of AI; opportunities for governments to support the development of beneficial and safe AI systems; supporting public-private partnerships to accelerate and democratize AI development. Specific principles include ensuring companies test AI systems for bias and use representative training data, as well as developing a diverse workforce capable of working with AI by expanding science, technology, engineering, and math education.
Beginning in January 2018, banks in the European Union will be required to allow other firms to access customer data if the customer gives their consent, which should help spur competition and allow financial firms to develop new products and services. Banks have an incentive to not share customer data with competitors so they often refuse to do so, however new rules called the Public Services Directive (PSD2) establish that customer data belongs to customers and prohibits banks from preventing customers from sharing their data with third parties. European antitrust officials have already begun investigating banks to determine if they are withholding access to customer data even though customers have consented for this data to be shared. PSD2 will also require banks to develop technological infrastructure to ensure they can quickly share data with other firms.
The U.S. Food and Drug Administration (FDA) has approved a device called the Confirm Rx Insertable Cardiac Monitor (ICM), developed by medical device company Abbott, that embeds under a patient’s skin and wirelessly shared data with their smartphone about cardiac arrhythmias. Though ICM’s are common, this is the first approved by the FDA that works with a smartphone. Sensors in the ICM continuously monitor heart rhythms and shares this data with a smartphone app at regular intervals, which patients’ doctors can also access, and patients can use the app to log information about any symptoms they experience.
Researchers at the Mitsubishi Electric Research Laboratory have developed a machine learning system capable of overcoming the “cocktail party effect” by distinguishing between multiple people speaking simultaneously. The system identifies distinct features in a speaker’s voice to create a voiceprint and then uses these voiceprints to reconstruct speech from multiple voices. The system can separate and reconstruct speech from two never-before-heard speakers with 90 percent accuracy, and from three speakers with 80 percent accuracy.
Connecticut-startup Butterfly Network has developed a small ultrasound device called Butterfly IQ that links with a smartphone and in a test, Butterfly Network’s chief medical officer identified a previously unknown cancerous tumor in his throat. Ultrasound devices typically use a vibrating crystal to shoot sound waves into a body so it can map an area based on the echoes, but these machines are large and costly to operate. Butterfly IQ, the first solid-state ultrasound device to reach the market in the United States, instead uses 9,000 drums etched onto a semiconductor which enables it to be portable and uses a smartphone to analyze these soundwaves.
The developer of DoNotPay, a chatbot that uses AI to automate the process of contesting parking tickets, has repurposed the chatbot to help people file for divorce in the United States without the need to pay a lawyer. Approximately 95 percent of divorces are uncontested, and in many of these cases, divorcing couples pay lawyers thousands of dollars to prepare and file paperwork that could be generated automatically. DoNotPay uses language processing algorithms from IBM’s Watson and has already helped 400,000 people contest parking tickets, saving them $11 million in fines.
Image: Voice of America News.