Home BlogWeekly News 10 Bits: the Data News Hotlist

10 Bits: the Data News Hotlist

by Joshua New
Changi airport

This week’s list of data news highlights covers April 28 – May 4, 2018, and includes articles about the full launch of the United States’ precision medicine initiative and a virtual scavenger hunt that teaches AI common sense.

1. Opening the Vatican’s Secret Archives

Scientists at Roma Tre University in Rome have launched a project called In Codice Ratio in partnership with the Vatican Secret Archives (VSA), which maintains one of the world’s largest collections of historical handwritten texts, to use AI to digitize and share VSA’s collection. The VSA houses 53 miles of shelving holding works dating back 12 centuries, but hardly any of it has been made available online, making it practically useless for scholars. The scientists will use an AI system they developed to augment existing character recognition software to allow it to interpret and transcribe handwritten texts.

2. Kicking Off Precision Medicine in the United States

The U.S. National Institutes of Health (NIH) is launching open enrollment for the All of Us precision medicine initiative, which aims to compile a database of comprehensive health data from 1 million Americans to advance medical research. NIH has been piloting the initiative since 2017 and has already enrolled 16,000 people to test how to best collect health and behavioral data, including genetic data and data from wearable fitness trackers.

3. AI Can Be Your Guide for the Royal Wedding

Sky News has partnered with Amazon Web Services to use AI to identify attendees in video coverage of the wedding of Prince Harry and Meghan Markle on May 19 in real time. Viewers using the Sky News app or watching online will be able to select famous guests during the livestream and the AI system will identify them and provide background information.

4. Keeping Bullies Off Instagram

Instagram has announced that it will implement an AI system called DeepText to analyze and filter user comments that contain toxic content such as bullying, sexual harassment, and racism. DeepText relies on deep learning to interpret language contextually, which enables it to understand words or phrases that are used literally, as well as identify content, such as slurs, even if users misspell them in an attempt to bypass the filter. DeepText can automatically hide these comments and flag users that are repeat offenders.

5. Data Sharing Could Make Autonomous Cars Safer

Autonomous vehicle company Oxbotica is piloting a system for sharing data from self-driving cars with other cars and third parties, including insurance company XL Catlin, to help improve navigation and reduce potentially risky behavior. As Oxbotica cars transmit data about speed, direction, and their surroundings in real time, they alert each other about road conditions, such as whether construction work is taking place on a particular stretch of road. By sharing this data with third parties, such as insurance companies, when a car encounters a difficult situation, such as large amounts of unexpected obstacles caused by kids leaving a school, a third party could review this data and recommend ways a car could reduce risk, such as by rerouting or reducing its speed.

6. Making Airports Smarter

Singapore’s Changi airport is testing a variety of different technologies, including facial recognition and connected sensors, that could help reduce travel delays and better predict travel times. The airport is testing how facial recognition could help locate travelers late for their flights to reduce delays, as well as automate immigration and bag check. Additionally, Cangi is trialing the use of sensors that monitor when planes leave the gate and take off, which has helped reduce taxiing time by 90 seconds during peak hours. Another test involves using AI to analyze wind and weather data to accurately predict flight arrival times two hours out, whereas Changi could only predict arrival times 30 minutes out using traditional methods.

7. Building an AI Scavenger Hunt

Researchers at Georgia Tech and Facebook have developed a virtual scavenger hunt challenge designed to evaluate whether an AI system can successfully use rudimentary common sense. The challenge involves asking an AI-controlled agent questions about the location of objects in a simulated home, so that over time the AI can predict where it makes sense to look for certain objects, such as by learning that cars are usually found in a garage and that it can usually access a garage by going out the house’s front or back door.

8. AI Could Help Prevent Childhood Blindness Better than Doctors

Researchers from Oregon Health and Science University have developed an AI system that can detect whether a patient has retinopathy of prematurity (ROP), which can cause blindness in premature babies, with 91 percent accuracy, whereas human experts can only do so with 82 percent accuracy. Diagnosing ROP involves evaluating whether blood vessels in the back of an eye are overly dilated or malformed and requires specialized training. This disease is difficult to diagnose and can cause blindness if doctors do not detect it early. The researcher’s system could help alleviate the shortage of ophthalmologists trained to diagnose ROP.

9. Performing Background Checks in Real-Time

Israeli startup Intelligo has developed background check software, called Clarity, that uses AI to automatically conduct background checks with 90 percent accuracy. Background checks are crucial for a wide variety of business applications, but are typically manual processes that can be time consuming, expensive, and inaccurate. Clarity can analyze data from social media, legal records, and other sources typically used in background checks to automate this process and identify potential red flags about a person in real-time.

10. Turning Instagram into Training Data

Facebook has developed a method for training computer vision systems on public Instagram posts that relies on user-created hashtags to provide clues about the contents of each image. Developers typically have to manually label training data for an AI system to be able to learn to recognize different objects in images, but curating large training datasets can be resource intensive. By using hashtags instead, Facebook bypassed the need to manually label the contents of images. A computer vision system Facebook trained on this data performed two percent better on a benchmarking test than any other existing system.

Image: Nan-Cheng Tsai

You may also like

Show Buttons
Hide Buttons