Welcome to the December 21, 2015 edition of ACM TechNews, providing timely information for IT professionals three times a week.
Updated versions of the 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
The Billion-Dollar Robot Question--How Can We Make Sure They're Safe?
The Washington Post (12/18/15) Matt McFarland
The advent of unsupervised robots raises the question of how to ensure their safety, especially as they challenge traditional methods of safety assessment and regulation. For example, some are calling for the establishment of a new U.S. government agency to oversee robot regulation, given the lack of expertise in conventional agencies. "The government itself is not acting as a repository of [robotics] expertise here," notes University of Washington professor Ryan Calo. "I worry quite a bit that government will over-rely on experts from industry because they don't have their own internal knowledge." Some states are taking the initiative, with California's Department of Motor Vehicles releasing draft guidelines for autonomous cars, but without addressing fully driverless vehicles. Google has criticized the state's move, noting the point of its autonomous cars program is to improve road safety. Meanwhile, the U.S. Department of Transportation is mainly concentrating on communications signals between autonomous vehicles, instead of determining how safe driverless cars will be. The University of Michigan's Ryan Eustice suggests the robotics regulatory agency Calo envisions could employ a mix of hardware testing, laps on test tracks, tests in a virtual simulator, and mileage driven on public roads as a way to assess autonomous vehicles for licensing.
Is Evolution More Intelligent Than We Thought?
University of Southampton (United Kingdom) (12/18/15)
University of Southampton researchers have shown that evolution can learn from previous experience, a breakthrough they say could provide a better explanation of how evolution by natural selection produces such apparently intelligent designs. The researchers brought together the theory of evolution with learning theories to demonstrate it is possible for evolution to exhibit some of the same intelligent behaviors as learning systems, including neural networks. Formal analogies can be used to transfer specific models and results between the two theories to solve several important evolutionary puzzles, according to the researchers. "Showing that evolving systems can learn from past experience means that evolution has the potential to anticipate what is needed to adapt to future environments in the same way that learning systems do," says University of Southampton professor Richard Watson. His team thinks these biological evolutionary breakthroughs can be applied to evolutionary computer programs to develop better systems, such as neural networks. "Learning theory enables us to formalize how evolution changes its own processes over evolutionary time," Watson says.
An App to Digitally Detox From Smartphone Addiction: Lock n' LOL
Korea Advanced Institute of Science and Technology (KAIST) researchers have developed an application to help people rein in their impulse to use smartphones during meetings or social gatherings via peer pressure while still permitting their use for emergencies. The Lock Your Smartphone and Laugh Out Loud (Lock n' LoL) app enables users to create a new room or join an existing room, and then invite meeting participants or friends to the room and share its ID with them to put the Group Limit (lock) mode into action. All alarms and notifications are automatically muted in lock mode, and users must request permission to unlock their phones. However, in an emergency, users can access their phones for five minutes in a provisional unlimit mode. In addition, Lock n' LoL's Col-location Reminder senses and lists nearby users to encourage them to restrain their phone use. The app also displays key statistics to monitor users' behavior, such as the current week's total limit time, the weekly average usage time, top friends ranked by time spent together, and top activities in which users participated. "In an age of the Internet of Things, we expect that the adverse effects of mobile distractions and addictions will emerge as a social concern, and our Lock n' LoL is a key effort to address this issue," says KAIST professor Uichin Lee.
'Robot Locust' Can Traverse Rocky Terrain and Assist in Search and Rescue
American Friends of Tel Aviv University (12/17/15)
Tel Aviv University (TAU) researchers have developed TAUB, a locust-inspired robot that can jump twice as high as existing similar robots. The machine can jump 11 feet high, and cover a horizontal distance of 4.5 feet in a single leap. The researchers say TAUB should perform well in search-and-rescue missions and in reconnaissance operations in rough terrain. "Our locust-inspired miniature jumping robot is a beautiful example of bio-inspired technological innovation," says TAU professor Amir Ayali. The researchers printed the robot's body on a three-dimensional printer, and used stiff carbon rods and steel wire to make the legs and torsion springs. A small onboard battery powers the robot, which is remotely controlled through an onboard microcontroller. "Biological knowledge, gained by observing and studying locusts, was combined with state-of-the-art engineering and cutting-edge technologies, allowing biological principles to be implemented in a miniature robotic jumping mechanism," Ayali says. Instead of producing an exact mechanical replica of a locust, the researchers focused on some of the specific biomechanical features of the locust's highly successful jumping mechanism. Like a locust, the robot's jumping ability can be attributed to its ability to store energy in its torsion springs.
Machine Learning and the Market for Intelligence: Heavyweights of the AI World Gather at U of T
U of T News (12/16/15) Alan Christie
The recent Machine Learning and the Market for Intelligence conference at the University of Toronto (U of T) was a showcase for diverse opinions on artificial intelligence (AI) from various experts and entrepreneurs. "Nobody is certain whether we are on the brink of an AI revolution or simply enjoying a brief reprieve from the AI winter," said U of T professor Ajay Agrawal. "However, almost everyone is certain that significant economic turbulence is coming given the falling cost of processing power, storage, and sensors, coupled with the rising collection of data from mobile devices, wearables, and other so-called Internet of Things devices, all occurring as algorithm performance marches steadily forward." As an example of the use of machine learning (ML) in the new economy, entrepreneur Shivon Zilis cited a camera-outfitted tractor that made decisions in real time about whether crops did or did not need pesticides. Meanwhile, U of T professor Steve Mann argued AI should be rebranded as Humanistic Intelligence, pointing to the growing relevance of the tension between AI and technological augmentation. Agrawal acknowledged ML advancements have helped enhance lifestyle and medicine, but said they have not yet reached a point where they can facilitate an economic transformation.
Roadmap to Safer Cyberspace
National Science Foundation (12/15/15) Aaron Dubrow
Researchers from SRI International and the Information Sciences Institute of the University of Southern California have outlined a strategy for next-generation experimental cybersecurity research in a report commissioned by the U.S. National Science Foundation. "This report is a critical first step to re-think what is needed in cyber experimentation before we build the infrastructure," says U.S. Department of Homeland Security Cyber Security Division director Douglas Maughan. According to the study, the research community must develop a "science of cybersecurity experimentation" that taps techniques and approaches to generate reproducible studies the community can test, reuse, and build upon. The report stressed the necessity for infrastructure supporting and facilitating repeatable experiments by creating simple methods for researchers to test each others' results. Furthermore, the study said researchers should use common standards and the means for cross-discipline and cross-domain research. The final required elements are new approaches for sharing and creating data to expedite knowledge- and community-building across disciplines and organizations. The researchers cite multidisciplinary research, precise modeling/incorporation of human activity, adherence to common models of infrastructure and experiment components via open interfaces and standards, reusable designs, and a usable and intuitive infrastructure as essential ingredients for ensuring transformational outcomes.
Can This Man Make AI More Human?
Technology Review (12/17/15) Will Knight
New York University professor and Geometric Intelligence founder Gary Marcus believes the way children learn and reason may hold the key to making machines much more intelligent. Marcus has spent decades studying the way the human mind works and how children learn new skills. One of the big inspirations for Geometric Intelligence is the way toddlers pick up new concepts and generalize. The company aims to create algorithms for use in an artificial intelligence (AI) machine that can learn new and better ways. Computer scientists and mathematicians now at the forefront of AI have embraced an approach known as deep learning, which has produced some astonishing results in recent years, but Marcus believes researchers are missing a huge opportunity by ignoring many subtleties of the human mind. He believes the brain is capable of more than just recognizing patterns in large amounts of data; it acquires deeper abstractions from relatively little data, Marcus notes. His company is using a number of technologies such as probabilistic algorithms to recreate human learning.
How Machines Write Poetry
Motherboard (12/15/15) Elizabeth Preston
As a teenager in Vermont, Sarah Harmon used Java to develop software that wrote poetry. Today, Harmon is a computer science Ph.D. student at the University of California, Santa Cruz, who wants to develop software that is actually creative. Her latest effort, called FIGURE8, uses case-based reasoning to generate figurative language. Harmon wants to build a program that could evaluate its own ideas and choose the best ones, similar to a human author. Her goal for FIGURE8 is not a program that can compose great poetry, but rather one that can aid human creativity. "This is a discovery process for figuring out how humans are creative," Harmon says. Like a human author brainstorming and revising, FIGURE8 produces many possible similes, and then goes back to analyze what it has written, ranking all of its similes according to Harmon's criteria of clarity, novelty, aptness, and surprise. She says computers that can simulate how a person generates ideas may be able to help them with all kinds of creative problem solving. Harmon notes that might mean writing music, coming up with recipes, or inventing new techniques for processes such as building bridges.
Where Is Machine Learning Heading in 2016?
CIO Australia (12/15/15) Rebecca Merrett
Machine-learning (ML) innovations expected in 2016 include smarter mobile applications, intelligent digital assistants, and democratization of artificial intelligence (AI), according to analysts. Frost & Sullivan analyst Mark Koh anticipates a greater prevalence of machine learning in mobile apps that extends beyond automated cars, robotics, and drones to include enterprise and productivity apps. "For example, increasing use of [natural language processing], machine vision, and voice interface," Koh says. He also predicts AI/machine-learning democratization's continued progress in 2016, as more vendors produce easy-to-use products in response to demand to make standard machine-learning algorithms accessible. "The challenge of using ML, predictive analytics, and AI technologies in an automated manner is that in many cases it throws up a lot of false positives and the ability to understand the underpinning algorithm would be important--expertise that is lacking currently--to understand the results," Koh acknowledges. Meanwhile, Gartner analyst Ian Bertram expects machine learning-enabled digital personal assistants to debut next year, with the advent of more industry-specific digital assistants that learn over time. "The skills to do machine learning, the data science skills to do machine learning, we are still in a short supply globally," he points out. Deloitte's Stuart Johnston says executives' interest in applying machine learning to business strategies is growing, which will likely transform business models.
'Plucking' Light Particles From Laser Beams Could Advance Quantum Computing
LiveScience (12/14/15) Edd Gent
Researchers at Israel's Weizmann Institute of Science have devised a method of reliably "plucking" an individual particle of light out of a laser pulse. The ability to generate single photons is very important for the future of quantum computing and quantum cryptography, but current methods are very unreliable. The current method for generating single photons sometimes generates multiple photons, a phenomenon that would undermine quantum computing or quantum encryption. However, the Weizmann researchers have used a method know as single-photon Raman interaction (SPRINT), which involves using supercooled atoms and optical technology to consistently pluck just one photon out of a beam of light. The researchers had previously used the method to build an all-optical router for quantum communication systems. The method could be used to build reliable quantum computing or encryption systems, or to "purify" other methods of extra photons. Weizmann researcher Barak Dayan says their method has two advantages: "One: In principle, it always happens--it's deterministic. Two: You're not losing the photon, just diverting it, and you can use it for other processes."
These Are the Decisions the Pentagon Wants to Leave to Robots
Defense One (12/14/15) Patrick Tucker
The U.S. Pentagon's third offset strategy calls for developing new artificial intelligence (AI) and autonomy capabilities to gain a strategic advantage over the country's adversaries. Deputy defense secretary Robert Work cleared up some of the mysteries surrounding the research effort during a national security forum co-hosted by the Center for a New American Security and Defense One. Work cited a National Geospatial Intelligence program called Coherence Out of Chaos, which could cue human analysts to take a look at different situations as those situations evolve on the ground. "It will do so in situations that require faster-than-human reaction," Work notes. Robots also could conduct cyber defensive operations, electronic warfare, and over-the-horizon targeting. "We believe strongly that humans should be the only ones to decide to when use lethal force," Work says. "But when you're under attack, especially at machine speeds, we want to have a machine that can protect us." The Pentagon also wants robots to tell F-35 pilots what to point and shoot at, how to land and fly, and also wants them to fly drones and drive boats. A fast-learning system would back many of the AI aides and capabilities. Work says the Pentagon is playing catch-up in the area of automation and AI, noting "the commercial world has already made this leap."
How Long Until We Can Build R2-D2 and C-3PO?
The Conversation (12/17/15) Mary Ellen Foster
Creating socially intelligent, interactive robots such as those depicted in the "Star Wars" films is a challenging proposal, involving an integration of diverse technologies, writes University of Glasgow lecturer Mary Ellen Foster. The technical building blocks for such machines--such as walking technology, speech recognition/synthesis, and computer vision--are steadily advancing. However, the challenge lies in putting them together in a way the resulting robot can understand and provide non-verbal communication and appropriate contextual interpretation and response. Foster says the field of data science may hold the key to the realization of truly socially intelligent robots. Robot developers' approach to interactive machines has shifted from using preprogrammed rules to machine learning. The latter approach involves recording interactions between people or between people and robots, and then "teaching" the robot how to act based on what that data shows, enabling the device to be much more flexible and adaptive. Moreover, the addition of deep learning is helping robots learn behavior from raw data, giving them much more open-ended potential. Foster notes these methods already have led to significant advancements in speech recognition and computer vision.
Abstract News © Copyright 2015 INFORMATION, INC.
To submit feedback about ACM TechNews, contact: [email protected]
Current ACM Members: Unsubscribe/Change your email subscription by logging in at myACM.