This week’s list of data news highlights covers April 21-27, 2018, and includes articles about major initiatives in Europe to invest in AI and a method for mapping poverty with mobile phone data.
The European Commission has launched a new initiative to support the advancement of artificial intelligence to boost European competitiveness. The initiative’s goal is to increase public and private sector investment in AI research and development by €20 billion (US $24.2 billion) by 2020. To achieve this, the EU will invest an additional €1.5 billion (US $1.8 billion) in its Horizon 2020 research and innovation program by 2020 to support AI research, which it expects will trigger an additional €2.5 billion (US $3.0 billion) in private sector funding. The Commission will also encourage member states to prepare for social and economic changes caused by AI, such as by modernizing education systems, and develop ethical and legal frameworks for AI.
The United Kingdom has announced a similar AI investment initiative, planning to increase investment in AI to £1 billion (US $1.4 billion), which includes £300 million (US $417 million) in government spending in addition to its existing £400 million (US $556 million) investment, and £300 million of funding committed by the private sector. The government will use some of these funds to support 1,000 new AI doctorate students, train 8,000 new computer science teachers for secondary schools, and establish a Center for Data Ethics to research AI safety and ethics.
Researchers at the University of Washington have developed a method for mapping poverty in a region by using mobile phone data and machine learning. The researchers developed a machine learning algorithm that could estimate a person’s socioeconomic status by analyzing their phone usage patterns and applied this algorithm to data about 1.5 million Rwandan mobile phone users. This approach allowed the researchers to map the geographic distribution of wealth with nearly the same accuracy as the Rwandan government’s census data.
Researchers at Microsoft and Kyoto University have developed an AI system that generate poetry based on images. The researchers trained their system on images paired with data about rhyming patterns and relevant language, and while most of the poetry is nonsensical, some of it was capable of fooling human judges into thinking it was written by humans. Training AI to perform tasks like writing poetry is valuable because it helps researchers learn how to develop algorithms capable of creativity, which many believe to be an important component of intelligence.
Researchers at the University of Cambridge used data mining to measure the relationship between a city’s level of cultural activity, or “cultural capital,” and its economic prosperity. Anthropologists have long suspected there to be a positive correlation between cultural capital and prosperity, but while measuring economic development is straightforward, cultural development is significantly harder. The researchers analyzed 1.5 million geotagged photos of New York and London from Flickr tagged with words related to culture, such as architecture, films, and music. By combining this data with official statistics about poverty, social vulnerability, and house prices, the researchers were able to demonstrate this link between cultural capital and prosperity on a neighborhood level.
Researchers at Northwestern University, the Department of Energy, and the National Institute of Standards and Technology, are using machine learning to discover new alloys of metallic glass, a combination of metals with properties allowing it to be lighter and stronger than steel. Scientists have known about metallic glass for 50 years but have only tested 6,000 of the millions of potential combinations of metals that could create metallic glass alloys. The researchers used machine learning to develop a method for predicting combinations of alloys that could create metallic glass substantially faster than human analysis. Using this method, the researchers have already analyzed 20,000 combinations of metals and discovered 3 new alloys.
YouTube has increased its use of AI to identify and remove videos uploaded to the site containing objectionable content. YouTube’s AI system now flags 83 percent of videos that get deleted, allowing human reviewers to remove 75 percent of these before they receive any views. In the beginning of 2017, YouTube was able to remove only 8 percent of violent extremist videos before receiving 10 views, but it can now delete over 50 percent of these videos before they receive 10 views.
Norwegian research institute SINTEF Ocean has launched a project called SMARTFISH-H2020 to develop a suite of technologies that use machine learning to reduce the ecological impact of commercial fishing. The project will include developing image recognition algorithms for smartphones and video cameras on ships that can automatically quantify catches in real time and share this data with commercial fisheries, researchers, and policymakers, who can use it to track fish stocks and better comply with fishing regulations.
Astronomers at the University of California, Santa Cruz have developed a deep learning system that can analyze images of galaxies from the Hubble Space Telescope and classify them using similar techniques as facial recognition algorithms. Galaxies can exhibit distinct visual traits depending on their age, including density and color, as younger galaxies with more young stars emit more blue wavelengths of light, while older galaxies emit more red wavelengths, but parsing through telescopic imagery to classify these images can be time consuming. The astronomers trained the system on simulated images of galaxies of different ages and by testing it on real observational data, the system could reliably determine which stage of life a galaxy was in.
Nvidia has developed AI software capable of reconstructing corrupted images with a high degree of realism. In an image of a landscape with a road going through it, for example, a user can white-out the road and the software can generate replacement imagery, a technique known as inpainting, that blends in with its surroundings. Nvidia developed the inpainting software by training a neural network on large training datasets of landscapes, faces, and objects, with random streaks and holes overlayed on them so it could learn to predict what replacement imagery should look like.
Law enforcement in California was able to identify and arrest the Golden State Killer, a notorious serial killer and rapist active in California in the 1970s and 1980s, thanks to genetic analysis and a consumer genomics ancestry service. Authorities had the Golden State Killer’s DNA for decades but were unable to identify whose it was until they sequenced it and submitted this data to a free ancestry service called GEDmatch, which allows users to submit their DNA to map their family tree. GEDmatch revealed that the DNA was a partial match to a relative of the killer, which allowed authorities to investigate his family members and finally apprehend him.