Welcome to the December 23, 2016 edition of ACM TechNews, providing timely information for IT professionals three times a week.
Editor's Note: In Wednesday’s edition of TechNews, we included an article reporting on a project that purported to use artificial intelligence to predict a woman’s personality from her appearance. ACM is committed to respect for all its members, and we should not have included in TechNews an article on a project whose purpose is to objectify women. We regret the error, and apologize to our readers.
Please Note: In observance of the upcoming U.S. holiday, TechNews will not be published on Monday, Dec. 26. Publication will resume Wednesday, Dec. 28.
ACM TechNews mobile apps are available for Android phones and tablets (click here) and for iPhones (click here) and iPads (click here).
HEADLINES AT A GLANCE
Researcher Proposes Parallel Intelligence, a Move Toward the Intelligent Future
Chinese Academy of Sciences professor Fei-Yue Wang says he saw the birth of a new age of intelligent technology nine months ago when a computer program beat a champion Go player. "It marked the beginning of a new era in [artificial intelligence (AI)]--parallel intelligence," he contends. Parallel intelligence is described as the interaction between physical and virtual reality, and Wang says it inverts the current view of traditional AI in that small, complex laws guide huge data instead of large and universal laws directing small amounts of data. Wang says the human Go champion "was not defeated by a computer program, but by all the humans standing behind the program, combined with the significant cyber-physical information inside it. This also verifies the belief of many AI experts that intelligence must emerge from the process of computing and interacting." Wang proposes a strategy, artificial/computational/parallel, to produce big data from small data, and then abstract big data into specific laws, where AI systems learn from millions of computational experiments to make the best decisions while engaging in parallel with real-world physical systems. "AI is not 'artificial' anymore," Wang emphasizes. "Ultimately, it becomes the 'real' intelligence that can be embodied into machines, artifacts, and our societies."
Experts Split on How Soon Quantum Computing Is Coming, but Say We Should Start Preparing Now
Network World (12/22/16) Maria Korolov
Although opinion is divided on the precise timeline for quantum computing's arrival, experts agree enterprises should start preparing the groundwork for new forms of encryption. Professor Kevin Curran at the U.K.'s University of Ulster expects quantum computing to make many modern and popular public key algorithms obsolete. Possible post-quantum alternatives to public key encryption include lattice-based, hash-based, and multivariate cryptographic algorithms as well as those that update the current Diffie-Hellman algorithm with supersingular elliptic curves. Some future-proof encryption algorithms have been developed and are undergoing tests, but enterprises should begin checking now to see whether their in-house and vendor-provided systems have the flexibility to permit early replacement of obsolete algorithms with new ones. Because longer keys will not work for public key encryption and companies will need to replace their algorithms, the encryption technology should be modular, says Echoworx's Sam Elsharif. Still, he notes while the theory of quantum computing cracking modern encryption systems is proven, the engineering to make it a reality could be some time off. However, Masergy Communications' Mike Stute says the threat is more imminent. "It's probably safe to say that nation states are not on the first generation of the technology but are probably on the second," he contends.
For Powerful Arguments
Technical University of Darmstadt (Germany) (12/19/16)
The Ubiquitous Knowledge Processing Lab (UKP) at Germany's Technical University of Darmstadt (TU Darmstadt) is developing software tools to check the quality of technical texts intended to support arguments posted on the Internet. TU Darmstadt professor and UKP director Iryna Gurevych hypothesizes a scenario where a learning machine identifies arguments in a collection of documents and analyzes them with associated corroborations on whether parents should encourage their children to restrict mobile phone use. The analysis system recognizes the topic and applies keywords to locate text fragments relevant to the query, then begins a predicate argument analysis in the individual fragments, seeking the deed and its frame of reference in the sentences. Following fragment analysis, references with regard to content are generated between all text passages found, and the system uses its own knowledge database and user feedback to identify premises, assertions, and supporting or contradictory corroborations for the specific argument. The system will recognize shortcomings in the argument--such as vague resources or one-sidedness--and produce a graphic that rates the argument's plausibility and credibility. The researchers have compiled an argument database to train the system on. "It allows us to start a new discussion about the possibilities of machine learning," says UKP scientist Ivan Habernal.
Artificial Intelligence: The 3 Big Trends to Watch in 2017
TechRepublic (12/20/16) Hope Reese
Experts expect artificial intelligence (AI) trends to progress in three major areas in 2017, with University of Maryland professor Marie desJardins anticipating "increasing use of machine learning and knowledge-based modeling methods" in the coming year. Duke University professor Vince Conitzer envisions growing interest--and concern--from the general public on AI's societal ramifications as a result of "specific new technological developments and their failures." Meanwhile, Sundown AI CEO Fabio Cardenas thinks instances of "AI going rogue" could become a fact of life in 2017, ranging from system-hacking AI created to commit fraud to hackers' introduction of bias and exemptions into existing AI to subvert its predictive capabilities. Roman Yampolskiy, director of the University of Louisville's Cybersecurity Lab, warns "AI failures will grow in frequency and severity proportionate to AI's capability." The third trend experts foresee is increased concentration on AI's moral and ethical implications. "The traditional AI community within computer science will increasingly address societal and moral issues in their work," Conitzer predicts. Professor Toby Walsh at Australia's University of New South Wales expects to see an accidental fatality caused by a driverless car highlighting the value of such issues as steps are taken to develop and regulate AI.
Bat Chat: Machine Learning Algorithms Provide Translations for Bat Squeaks
The Guardian (12/22/16) Nicola Davis
Researchers from Tel Aviv University in Israel are studying Egyptian fruit bats and have found a way to determine which bats are communicating with each other, what they are communicating about, and predict the outcome of a disagreement between bats. The new approach involves harnessing machine-learning algorithms designed for human voice recognition. The researchers spent 75 days continuously recording audio and video footage of 22 bats that were divided into two groups and housed in separate cages. The researchers studied the video footage and were able to identify which bats were arguing with each other and the outcome of each disagreement. They divided each argument into one of our categories--sleep, food, perching position, and unwanted mating attempts, and then trained the algorithm with about 15,000 bat calls from seven adult females, each categorized using information taken from the video footage. The researchers found that, based only on the frequencies within the bats' calls, the algorithm correctly identified the bat making the call about 71 percent of the time, and what the animals were fighting about around 61 percent of the time. The system was also able to identify, with less accuracy, which bat the call was directed towards and predict the result of the disagreement.
Open Source Challenger Takes on Google Translate
InfoWorld (12/20/16) Serdar Yegulalp
A new open source machine translation framework could serve as an alternative to closed-source projects such as Google Translate. Open Source Neural Machine Translation (OpenNMT) is built on the work of researchers from Harvard University and machine-language software creator Systran. OpenNMT runs on the Torch scientific computing framework and uses the Lua language to interface with Torch. The new open source neural network system for performing language translations works like other products in its class. The training process for OpenNMT models can be accelerated on any graphics-processing-unit-equipped system, but it can still take a long time--sometimes many weeks. The training process can be snapshotted and resumed on demand if needed. For common language pairs such as English/French, the translations are very accurate, but OpenNMT does not supply pretrained language model data, which will limit its usefulness out of the box. A live demo provided by Systran claims to use OpenNMT in conjunction with Systran's own work.
We Don't Understand How AI Make Most Decisions, So Now Algorithms Are Explaining Themselves
Quartz (12/20/16) Dave Gershgorn
Artificial intelligence (AI) algorithms are normally only programmed to provide an answer based on the data they have learned, and not to provide rationales for their conclusions. "Engineers have developed deep-learning systems that 'work'--in that they can automatically detect the faces of cats or dogs, for example--without necessarily knowing why they work or being able to show the logic behind a system's decision," says Microsoft principal researcher Kate Crawford. A growing field of research seeks to fill this gap so there is less confusion and uncertainty of what goes wrong in the event of failure. Research from the University of California, Berkeley (UC Berkeley) and Germany's Max Planck Institute for Informatics might lead to an AI algorithm that analyzes data in two ways. Its first task is to answer the original question, and its second involves identifying the data used to answer the question so it can translate it into English. "The difficulty is explaining the individual decisions in a way that is human interpretable," says Devi Parikh at Virginia Polytechnic Institute and State University. A system such as the UC Berkeley and Max Planck Institute's analyzes long sets of numbers, finds commonalities between them to determine what the machine was looking at, and describes it in a human-readable manner.
Artificial Intelligence Is Going to Make It Easier Than Ever to Fake Images and Video
The Verge (12/20/16) James Vincent
The Twitter bot Smile Vector is a hint of what is to come as artificial intelligence (AI) opens a new world of image, audio, and video fakery. Smile Vector scrapes the Web for pictures of celebrities, then automatically morphs their facial expression into a smile using a deep-learning-powered neural network. Tom White, a lecturer in creative coding at Australia's Victoria University School of Design, created the bot because he wanted to show people what is happening in the AI space. Other AI multimedia manipulation tools in development right now would enable users to create three-dimensional face models from a single two-dimensional image, change the light source and shadows in any picture, generate sound effects based on mute video, or even live-stream the presidential debates but make Donald Trump bald. AI-powered image generation will be useful in creative industries, but another obvious beneficiary will be hoaxes. A tool that permits users to change the facial expression of Trump or Obama on video in real time could be combined with Adobe's new prototype software that can edit human speech. People could use video of politicians and celebrities and make them say whatever they want, and AI-powered manipulations are already difficult to spot and will only get better.
Enabling Access to Reproducible Research
University of Southampton (12/19/16)
Researchers from the University of Southampton in the U.K. are developing Stat-JR, a new software technology that lets students and other researchers access reproducible statistical research. Stat-JR is an online universal gateway to many specialized statistical packages that provides tools to help users learn about statistical methods and how they can be implemented. "When researchers are using Stat-JR, we record their actions in the form of provenance traces," says Southampton research fellow Danius Michaelides. "These traces can be turned into workflows so that the analyses can not only be rerun but also be edited and reused. When used in conjunction with Stat-JR's e-book system it allows other researchers and students to adopt and adapt complex analyses for their own research." The Stat-JR team has built tools to support reproducibility over the course of the three-year project. "Provenance is a key technology in supporting reproducible research and with this project we've explored how best to use provenance to improve the Stat-JR tools," notes Southampton professor Luc Moreau.
A Level Playing Field: Lab Adapts Toys for Disabled Children
Associated Press (12/19/16) Jason Dearen
A University of North Florida (UNF) program is now in its third year of adapting toys from store shelves for disabled kids. The Adaptive Toy Project brings engineering and physical therapy students together in a lab with the goal of customizing toys so children with limitations can use and play with them. UNF professor Mary Lundy says students meet with families, and go to therapy appointments and schools. "Engineering students teach the physical therapy students how to modify basic electronics...and in the process engineers learn how to do people-centered designs, and how to look at their clients differently," she says. Students recently worked on a toy car for Scarlett Wilgis, a four-year-old girl with cerebral palsy, replacing the steering with a large push button, and adding light sensors underneath to enable it to follow a line of tape along the floor whenever the girl hits the button. Scarlett's parents can now design routes for the car with tape or use a remote-control mode for family walks. The customized car costs more than $1,000, but the family gets it for free. The project has received a five-year grant from the U.S. National Institutes of Health.
Georgia Tech Lands $17 Million Cybersecurity Research Grant
Atlanta Business Chronicle (12/19/16) Urvaksh Karkaria
The Georgia Institute of Technology (Georgia Tech) has received a $17.3-million research contract from the U.S. Department of Defense to establish new science focused on "attribution," a technique that enables researchers to quickly, objectively, and positively identify virtual actors behind cyberattacks. Although the tools and methods developed over the four-and-a-half year project will not point directly to the individuals responsible, the initiative will provide proof of involvement by specific groups. The groups will be identified by their methods of attack, consistent errors, and other unique characteristics, according to the Georgia Tech researchers. "We owe it to the people of this country to objectively reason about the actors attacking systems, stealing intellectual property, and tampering with our data," says Georgia Tech professor Manos Antonakakis. Attributing attacks to specific groups or individuals is partially possible today, but the new research will accelerate the process and provide scientific reasoning and hard evidence about the guilty parties. "In this project, we will use machine learning and algorithms to scale up the attribution process to help companies and the government protect against those bad actors," Antonakakis says. Ultimately, the researchers hope to combine intrusion detection with attribution, enabling a quicker response and helping victims cut off attackers faster.
Mimicking Biological Movements With Soft Robots
Harvard University (12/19/16) Leah Burrows
To create a soft robot that moves organically, Harvard University researchers have developed a method to automatically design soft actuators based on the desired movement. "We wanted a tool where you could plug in a motion and it would tell you how to design the actuator to achieve that motion," says Harvard professor Katia Bertoldi. One actuator type is not enough to produce complex motions, as the technology requires a sequence of actuator segments, each performing a different motion. They also must be actuated using a single input, notes Harvard graduate student Fionnuala Connolly. The new method uses mathematical modeling of fluid-powered, fiber-reinforced actuators to optimize the design of an actuator to perform a certain motion. The researchers used this model to design a soft robot that bends like an index finger and twists like a thumb when powered by a single pressure source. Harvard professor Conor Walsh says the research "can be used to design a robot arm that moves along a certain path or a wearable robot that assists with motion of a limb." The new method will be included in the Soft Robotic Toolkit, an online, open source resource to assist researchers, educators, and innovators to build and control soft robots.
Ford Studies Using Drones to Guide Self-Driving Cars
Ford Motor researchers are studying a system to use drones to help guide self-driving vehicles, including on off-road adventures. The hypothetical system would involve autonomous vehicles relying on a drone to help guide it by mapping the surrounding area beyond what the car's sensors can detect. Passengers in the vehicle could control the drone using the car's infotainment and navigation system. Earlier this year, Ford held a competition for programmers to see if they could teach a drone to fly from and return to a moving vehicle. The goal was to determine if a drone could use its cameras to direct a vehicle into and out of a disaster area where communications and roads have been destroyed or disrupted, says Ford spokesperson Alan Hall. He notes the plan is to create drone-to-vehicle communications using Ford Sync, the company's car-based wireless connection, as a means to inspect areas in an emergency. Only one of the 10 participants successfully launched a drone from a moving vehicle and got it to return after completing an assigned task. Ford has partnered with researchers in the Silicon Valley Research Center to find a way for drones to help autonomous vehicles solve future navigation problems.
Abstract News © Copyright 2016 INFORMATION, INC.
To submit feedback about ACM TechNews, contact: [email protected]