Inside YouTube's Struggles to Shut Down Video of the New Zealand Shooting -- and the Humans Who Outsmarted Its Systems

The Washington Post

As video recorded by the alleged perpetrator of Friday's shootings at two New Zealand mosques played out on YouTube and other social media platforms, a group of senior YouTube executives assembled to try to identify and remove the videos. However, as soon as the group took down one video, another would appear, as quickly as one per second in the hours after the shooting. The team temporarily disabled several search functions and cut off human review features to speed removal of videos flagged by automated systems. “Every time a tragedy like this happens we learn something new, and in this case it was the unprecedented volume” of videos, said YouTube’s Neil Mohan. “Frankly, I would have liked to get a handle on this earlier.”

From "Inside YouTube's Struggles to Shut Down Video of the New Zealand Shooting — and the Humans Who Outsmarted Its Systems"
The Washington Post (03/18/19) Elizabeth Dwoskin; Craig Timberg
View Full Article

Stanford University Launches Institute for Human-Centered AI

Stanford News

Stanford University today launched the Stanford Institute for Human-Centered Artificial Intelligence (HAI) as an interdisciplinary hub for policymakers, researchers, and students. HAI will be co-directed by AI pioneer and former Google vice president Fei-Fei Li, and former Stanford provost John Etchemendy. The Institute’s advisory council includes former Alphabet executive chairman Eric Schmidt, LinkedIn co-founder Reid Hoffman, former Yahoo chief executive Marissa Mayer and co-founder Jerry Yang, and Jeffrey Dean, co-recipient of the 2012 ACM-Infosys Foundation award (now the ACM Prize in Computing). Etchemendy said HAI’s “biggest role will be to reach out to the global AI community, including universities, companies, governments, and civil society to help forecast and address issues that arise as this technology is rolled out.”

From "Stanford University Launches Institute for Human-Centered AI"
Stanford News (03/18/19)
View Full Article

Racing Against China, U.S. Reveals Details of $500 Million Supercomputer

The New York Times

The U.S. Department of Energy (DoE) has disclosed the details of a $500-million supercomputer, which may play a critical role in U.S.-Chinese competition to develop exascale-class systems. The Aurora system's delivery is slated for 2021 to the DoE’s Argonne National Laboratory. Argonne officials expect Aurora will be the first U.S. machine to exceed a quintillion calculations per second, and supporters hope it will facilitate more accurate models of phenomena like drug responses and climate change. Among Aurora's purported components are unreleased Intel accelerator chips, a version of Intel's standard Xeon processor, new memory and communications technology, and a design that stacks chips on top of each other for space and power savings. Meanwhile, Cray's Shasta system design and other contributions will expedite the flow of data inside Aurora. Cray and Intel also are providing software to make the programming of supercomputers easier.

From "Racing Against China, U.S. Reveals Details of $500 Million Supercomputer"
The New York Times (03/18/19) Don Clark
View Full Article - May Require Paid Subscription

Google Unveils Stadia, A High-End Gaming Service Without a Console

The Wall Street Journal

Google's new Stadia service will allow players to stream videogames to their smartphones, tablets, or computers from the cloud, without the need for costly hardware. With cloud gaming, the labor of generating a game's visuals and audio is performed on remote servers, then channeled over the Internet to the player's screen. Streaming games with rich graphics or having many people play simultaneously risks performance slowdowns, which makes cloud providers well suited to game streaming. Google said it collaborated with Advanced Micro Devices to produce a customized graphics processing unit for Stadia. According to Google, games on Stadia will run at up to 4K resolutions with high dynamic range, at 60 frames per second.

From "Google Unveils Stadia, A High-End Gaming Service Without a Console"
The Wall Street Journal (03/19/19) Sarah E. Needleman
View Full Article - May Require Paid Subscription

U.S. Trails Europe in Technology, Data-Science Skills Ranking


Europe accounts for more than 80% of nations in the top 25% of countries across the technology, business, and data science domains, according to a recent Coursera Inc. index of technology skills. The data showed certain countries as consistent leaders across all three categories: Finland, Switzerland, Austria, Sweden, Germany, Belgium, Norway, and the Netherlands. The index places the United States 16th out of 60 nations for data science, 23rd in technology, and 18th in business. The results highlight the idea that the U.S. should invest more in developing skilled labor to support the kind of technological innovation that will keep the economy competitive on a global stage. Coursera created the measures to track global skill trends based on data compiled from 40 million assessments by 3 million platform users from 60 countries and 10 industries. Said Coursera's Emily Glassberg Sands, "The U.S. isn't really cutting-edge across any of the domains, so that suggests that there's real competition, at least from a skills perspective."

From "U.S. Trails Europe in Technology, Data-Science Skills Ranking"
Bloomberg (03/19/19) Jeff Kearns
View Full Article

The Computer Scientist Who Wants to Put a Name to Every Face in Civil War Photographs


Virginia Polytechnic Institute and State University's Kurt Luther has combined crowdsourcing and facial recognition to produce free, online software to help users identify unknown subjects in Civil War-era photos. The development of the free online Civil War Photo Sleuth software began with the accumulation of a large database of already identified photos, from national archives as well as private collections. The database helps users identify people in photos they upload themselves, having manually tagged special visual traits; a facial-recognition algorithm then analyzes and logs unique face ratios. Photo Sleuth compares the visual data of the unknown image to already identified photos in the database, and displays what it considers the best matches, based on facial similarity and information derived from the other metadata. Analysis determined 85% of the proposed identifications were either probably or definitely a match.

From "The Computer Scientist Who Wants to Put a Name to Every Face in Civil War Photographs"
Smithsonian (03/19) Sarah Wells
View Full Article

Prisoners in England to Be Taught to Code

BBC News

As part of a project to help people in disadvantaged communities expand their digital skills, the U.K. government will fund a program in which "carefully vetted" prisoners will be taught software coding. To be rolled out initially at two prisons and an employment hub, the government will provide £100,000 ($132,600) to underwrite the program, in which participants will be trained in basic coding, before graduating to more advanced levels. Prisoners will work on real-world projects for outside clients, then for clients on temporary day release, with the goal of helping them secure full-time employment as developers when their sentences are complete. The government hopes the program, modeled after the Last Mile project at California's San Quentin prison, will lead to a network of coding workshops in U.K. prisons. Said Neil Barnby with the Code 4000 organization, "The workshops are reducing reoffending at a measurable rate, because we keep in touch with our graduates."

From "Prisoners in England to Be Taught to Code"
BBC News (03/15/19)
View Full Article

Quantum Computing Should Supercharge Machine Learning Technique

Technology Review

IBM and Massachusetts Institute of Technology (MIT) researchers have demonstrated how an IBM quantum computer can expedite the machine learning task of feature matching, which should allow future quantum computers to accelerate machine learning. Feature mapping transforms data into a mathematical representation that lends itself to machine learning analysis, depending on the efficiency and quality of this process. The researchers employed a 2-quantum-bit system to execute the calculation. The researchers said they are “still far off from achieving quantum advantage for machine learning,” adding, “what we've shown is a promising path forward."

From "Quantum Computing Should Supercharge Machine Learning Technique"
Technology Review (03/13/19) Will Knight
View Full Article

U.S. Regulators Want Public's View on Cars with No Steering Wheel, Brake Pedals


A U.S. regulator is asking for public comments on whether autonomous cars lacking steering wheels and brake pedals should be allowed on the nation’s streets. The U.S. National Highway Traffic Safety Administration (NHTSA) has delayed action for 15 months on General Motors' request to deploy a small number of vehicles on U.S. roads without human controls. GM wants the NHTSA to compare a vehicle in which all driving decisions are made by a computer to a vehicle piloted by a human driver. NHTSA is also seeking public comment on a separate petition by driverless delivery startup Nuro to deploy a limited number of low-speed highly automated delivery vehicles without human occupants. Both petitions seek exemptions from U.S. vehicle safety regulations written decades ago that assume human drivers would always be in control of a vehicle. The NHTSA will accept public comments on the proposals for at least 60 days.

From "U.S. Regulators Want Public's View on Cars with No Steering Wheel, Brake Pedals"
Reuters (03/15/19) David Shepardson
View Full Article

Seeing Through a Robot's Eyes Helps Those With Profound Motor Impairments

Georgia Tech News Center

Georgia Institute of Technology researchers have developed an interface that uses augmented reality to help the mobility-challenged operate a humanoid robot to perform routine personal care tasks. The Web-based interface displays a "robot's eye view" of surroundings to help users interact with their environment. The researchers employed a PR2 mobile manipulator, a wheeled robot with 20 degrees of freedom, two arms, and a "head," so it can manipulate objects. In one experiment, participants with motor impairments controlled the robot remotely online, operating a mouse cursor to perform personal care tasks; another test had the PR2 and interface used by a movement-impaired subject at home, which allowed the participant to devise novel uses combining the operation of both robot arms simultaneously. The researchers suggested these "robotic body surrogates" could enhance users' quality of life, and offer a platform for designing faster, more capable assistive robots.

From "Seeing Through a Robot's Eyes Helps Those With Profound Motor Impairments"
Georgia Tech News Center (03/15/19)
View Full Article

The Best Image-Recognition AIs are Fooled by Slightly Rotated Images

New Scientist

Researchers at Auburn University have identified dozens of examples of how artificial intelligence (AI) is much worse at identifying objects by sight than many people realize. For example, distinguishing between a yellow taxi and a pair of binoculars seems simple, but if an image of a taxi is flipped upside down, many AI systems see a pair of binoculars. The Auburn researchers took images of common objects from ImageNet and randomly rotated and changed the position of the objects in the pictures. The team found this minor change was enough to confuse several state-of-the-art image-recognition systems 97% of the time, averaged across all of the systems. The biggest hurdle to progress for this technology is that when an AI system looks at an image, it cannot extract rules about the object shown that would help it identify a similar object in the future. Said Auburn researcher Anh Nguyen, "To reach a human level of reasoning, we need a way to extract rules from images."

From "The Best Image-Recognition AIs are Fooled by Slightly Rotated Images"
New Scientist (03/13/19) Douglas Heaven
View Full Article