This week’s list of data news highlights covers November 25 – December 1, 2017, and includes articles about an AI fashion assistant for retail stores and the first approved smart drug in the United States.
1. AI Teaches Itself to Be Bilingual Without a Dictionary
Two different groups of computer science researchers have both developed AI systems capable of translating between two languages without prior understanding of one of the languages. The systems use unsupervised machine learning to identify similarities in how words are grouped together in both languages to gradually piece together a bilingual dictionary to translate between each language. The systems cannot translate whole sentences as accurately as humans or more developed systems such as Google Translate, but it is adept at translating individual words.
2. Using AI to Keep Extremist Content Off Facebook
Facebook has announced that its automated content moderating system is now capable of removing 99 percent of the propaganda and other content spread by ISIS and al Qaeda before users report it. In addition, the system allows Facebook to remove 83 percent of known terrorrism content in less than an hour after it is posted. Facebook is now working to expand the system to identify and remove a broader array of extremist content, such as pages of white supremacist groups.
3. Compiling the World’s Open Data Standards
Bloomberg Philanthropies has partnered with Johns Hopkins University and Canadian open data group Geothink to create the Open Data Standards Directory, consisting of 60 different open data standards governments can use to publish open data. The directory provides information about how different governments use each standard, their compatibility, and their origins. According to its creators, the Open Data Standards Directory is the first effort to systematically index civic open data standards, which can enable governments to better differentiate between good and bad standards and promote interoperability.
4. Rapidly Identifying People with Their DNA
Researchers at Columbia University and the New York Genome Center have developed a DNA analysis method that can match a person’s genetic profile to verify their identify with near-perfect accuracy within minutes. The method relies on a portable DNA sequencing instrument called the MinION that, though quick, typically produces high error rates. The researchers developed an algorithm that can rapidly cross-check up to 300 genetic variants against a database to filter out erroneous sequencing data, making it possible to use the device for identity verification.
5. Stopping People from Reading Over Your Shoulder
Researchers at Google have created a machine learning system that uses a smartphone’s front-facing camera to alert users when it detects a face that is not theirs looking at the screen. The system uses eye-tracking to make this determination and can alert users if someone behind them is looking at their screen in real time.
A Stanford University-led group of researchers called the AI100 have published their annual AI Index report, which tracks the growth of AI using three categories of metrics: the volume of activity, which tracks factors such as the value of investments in AI; technical performance, such as benchmarks for how well algorithms can understand images; and derivative measures, which examine the relationship between trends in fields such as academia and industry. The report found that since 2000, the number of AI startups has increased 14-fold and venture capital investment in AI has increased 6-fold.
7. Spotting Suicidal Intent on Social Media
Facebook has developed a machine learning system that can spot users’ activity indicating they might attempt suicide and suggest resources to the user or their friends that can help, as well as alert a team at Facebook that can contact local authorities to intervene. Facebook began testing the system in the United States in March 2017, and after successful tests, it is now deploying the system in other countries.
8. Predicting When to Give Patients Palliative Care
Researchers from Stanford University have developed an AI system that can predict a patient’s likelihood of mortality within 3 to 12 months, which can help doctors make better decisions about when to provide palliative care. The system analyzes patient records to make its predictions and also provides doctors with a report explaining how it made its prediction. Hospitals could use this system to provide better treatment for patients despite the shortage of palliative-care professionals.
9. Understanding the Biological Roots of Mental Illness
Researchers at Sichuan University are analyzing magnetic resonance imaging (MRI) scans of human brains using machine learning to identify physiological differences in the brain structures of patients with depression or social anxiety, compared to healthy patients. The researchers found differences in cortical thickness and abnormalities in brain grey matter in patients with depression and social anxiety, and though their analysis used too few subjects to be conclusive, it can serve as the foundation for new avenues of research and potential treatments.
10. AI is Inventing New Video Games
A researcher from the University of Falmouth named Michael Cook has developed an AI system called Angelina capable of generating video games from scratch. Though game designers have used algorithms to create content for games for decades, such as by randomizing environments or generating straightforward tasks for players, sometimes called “throwaway content” due to its shallowness, Angelina is capable of generating more substantial kinds of content. For example, Angelina can generate rules for a game, populate games with openly licensed graphics from online repositories, and incorporate characters and ideas pulled from social media and online newspapers.
Image: Joseph Wright of Derby.