Artificial Intelligence  

Google's natural language voice search capability just took a tremendous leap in its understanding of your voice. More than just single words and short phrases, the company claims it is now able to understand the meaning behind your questions.

Google's work on artificial intelligence and natural language understanding has yielded some big advances, the company has reported on their blog. With voice enabled interfaces increasingly becoming important, especially on mobile devices, the competition to create the most accurate and smart systems is on.

"Now we’re “growing up” just a little more. The Google app is starting to truly understand the meaning of what you’re asking."

Google took its first steps understanding and answering questions with voice search in 2008, and with he Knowledge Graph in 2012. While the first systems could only make out single words and phrases items like “mama” or “car,” the Knowledge Graph started by providing information on individual entities like “Barack Obama” or “Shah Rukh Khan.”

"We graduated to answering simple questions about those entities, so you could ask “How old is Stan Lee?” or “What did Leonardo da Vinci invent?” We soon got a little smarter, so if you asked “What are the ingredients for a screwdriver?”, we understood you meant the cocktail and not the tool," writes Google  Product Manager Satyajeet Salgar,

"Now we’re “growing up” just a little more. The Google app is starting to truly understand the meaning of what you’re asking. We can now break down a query to understand the semantics of each piece..."

Related articles

• Synaptic Transistors Point To Computer Hardware That Learns 
• BRAIN Project Infographic Released
• William Hertling Looks At The Not-Too-Distant Future in the Last Firewall

According to the post, Google voice search is now starting to understand the intent behind the question. This allows the system to use the Knowledge Graph much more reliably to find the right facts and compose a useful answer—and they can use this base to answer harder questions.

Here are a few new types of complex questions Google can now handle. Google now understands superlatives—”tallest,” “largest,” etc.—and ordered items. So you can ask the Google app:

• “Who are the tallest Mavericks players?”
• “What are the largest cities in Texas?”
• “What are the largest cities in Iowa by area?”

Second, Google now has a much better understanding of questions with dates in them. So you can ask:

• “What was the population of Singapore in 1965?”
• “What songs did Taylor Swift record in 2014?”
• “What was the Royals roster in 2013?”

Finally, complex combinations are now better understood too. So Google can now respond to questions like:

• “What are some of Seth Gabel's father-in-law's movies?”
• “What was the U.S. population when Bernie Sanders was born?”
• “Who was the U.S. President when the Angels won the World Series?”

"We’re still growing and learning," admits Salgar which means the system still make mistakes. But the progress is very apparent, and shows just how far artificial intelligence is progressing.

SOURCE  Google

By 33rd Square

Embed

 Machine Learning  

Deep learning start-up MetaMind has published details of a system that is more accurate than other language processing methods.  The company is developing technology designed to be capable of a range of different artificial-intelligence tasks.

Google and Facebook, are investing huge sums into the  research and development of improved artificial intelligence algorithms for processing language.

Around the world, various research groups are making steady progress toward improving a computer's language skills especially using recent advances in machine learning.

"The insight—and it’s almost trivial—is that every task in NLP is actually a question-and-answer task."

The most recent development in Natural Language Processing (NLP) comes from a start-up called MetaMind, which has developed a language recognition system that is more accurate than the leading systems available on the market.

MetaMind has published new research detailing how their neural networking system uses a kind of artificial short-term memory to answer a wide range of questions about a piece of natural language.

According to MetaMind, the system can answer everything from very specific queries about what the text describes to more general questions like “What’s the sentiment of the text?” or “What’s the French translation?” The research, due to appear next week at Arxiv.org, a popular online repository for academic papers, echoes similar research from Facebook and Google, but it takes this work at step further.

MetaMind was founded by Richard Socher, a prominent machine-learning expert who obtained his PhD from Stanford where he worked with Chris Manning and Andrew Ng. Socher tested his algorithms using a data set compiled by Facebook for measuring machine performance at routine comprehension tasks. MetaMind's software ended up outperforming Facebook's own algorithms.

The new technology is designed to be capable of different artificial-intelligence tasks including image classification and sentiment analysis. The work is indicative of ongoing success in giving machines more efficient learning and comprehension.

“The insight—and it’s almost trivial—is that every task in NLP is actually a question-and-answer task,” Socher told Wired.

A key to this progress is an approach known as deep learning, a relatively new field of artificial intelligence research that aims to perfect tasks such as face and language recognition.

"[MetaMind's] deep learning technology is going to have enormous impact in multiple industries …” says Marc Benioff, the CEO of Salesforce.

Related articles The Clever World of Artificial Intelligence  Ben Goertzel On Aspects of Artificial General Intelligence  Gary Marcus Urges People To Think About The Threat of Artificial Intelligence

The MetaMind system processes data using what Socher calls “episodic memory.” Like how Demis Hassabis describes DeepMind's algorithms, this is something akin to the way the brain treats short-term memory in the hippocampus. The system must “remember” one fact before determining what another is, based on the natural language data supplied.

“You can’t do transitive reasoning without episodic memory,” Socher says.

And, he explains, you can use much the same setup to do analyze sentiment or translate words into a new language. “One model—one dynamic memory network—can solve these very different problems,” he says.

MetaMind's founder says that his work has made significant progress toward more generalizable artificial intelligence. “This idea of adding memory components is something that’s in the air right now,” he says. “A lot of people are building different kinds of models, but our goal is to try to find one model that can perform lots of different tasks.”

In the talk below, Socher describes how deep learning algorithms can learn language.




SOURCE  MIT Technology Review

By 33rd Square

Embed

 Artificial Intelligence  

Developers of artificial intelligence software in Japan hope to pass the Tokyo University entrance exam by 2021. The team just doubled their performance over last year.

Artificial intelligence in Japan is getting closer to entering college. An AI software system scored higher on the English section of Japan’s standardized college entrance test than the average Japanese high school senior recently, the development team said.

The software, known as To-Robo, nearly doubled its score on a multiple choice test from its performance a year ago, indicating progress toward a goal set by its developers to eventually pass the entrance exam for Tokyo University, Japan’s most prestigious college.

Related articles KIROBO Robot Prepares For Work Aboard The International Space Station  Top 10 Jobs for the Future  Dennis Hong On Robot Evolution and The DARPA Robotics Challenge

“The average score for the English section of the standardized entrance exam was 93.1 (out of 200), but the AI scored 95,” a spokesman for NTT Science and Core Technology Laboratory Group said. Last year the software scored 52.

The NTT lab is developing the software alongside the National Institute of Informatics, and is in charge of developing the software’s English capabilities. The project began in 2011 with a 10-year time frame for reaching its goal.

Questions from the test were turned into data that could be recognized by the software. To-Robo then processed the information, distinguishing the logic of exchanges and correctly identifying the right answer out of multiple choices.

For example, it was able to correctly choose the answer that best fits the following conversation:

A: I hear your father is in the hospital.
B: Yes, and he has to have an operation next week.
A: 【　　】. Let me know if I can do anything.
B: Thanks a lot. 

–Exactly, Yes.
–No problem.
–That’s a relief.
–That’s too bad.

"The average score for the English section of the standardized entrance exam was 93.1 (out of 200), but the AI scored 95."

While NTT lab said To-Robo was getting better at completing conversations, structuring sentences appropriately and grasping the context of a dialogue, it added that the software still needs to improve at understanding more complex exchanges and comprehending the emotions of speakers.

So far the system is said to be better at liberal arts topics, and has limits in some questions, such as when a diagram needs to be generated.  Other science and math questions involving image recognition and comparison were not answerable by To-Robo.

The robot project leader, Professor Noriko Arai National Institute of Informatics, said, "Exploring the limits of artificial intelligence can be said to be the purpose of this project. Finding out how to coordinate people and machines to get along, is one of Japan's key economic developments. "

The technology may be developed for further use in the future, the lab said, with natural language processing and translation seen among its possible applications.


SOURCE  Wall Street Journal

By 33rd Square

Embed

 Artificial Intelligence  

Summarizing what is going on in a video is another task that may soon be done automatically thanks to work done through the Video In Sentences Out study. Using artificial intelligence deep learning methodology, a team has already been able to achieve accurate results in almost half of the videos the system has examined.

In a DARPA-funded research effort, a team using computer vision, robotics, and natural-language processing and deep learning have created a system that provides sentence descriptions of what is occurring in a video.

Apart from the obvious use of summarizing YouTube videos, the applications of this narrow artificial intelligence system are many, including robotics and the development of smart cameras.

The system called Video In Sentence Out produces what the researches call "sentential descriptions" of video.  This includes the  who did what to whom, and where and how they did it.

The research for Video in Sentences Out was conducted at the University of Toronto, Purdue University and the University of South Carolina.

Video In Sentences Out was developed by the Purdue-University of South Carolina-University of Toronto team under the DARPA Mind's Eye program. The Mind's Eye program seeks to develop the capability for visual intelligence by automating the ability to learn generally applicable and generative representations of action between objects in a scene directly from visual inputs, and then reason over those learned representations.

The study has been published in Proceedings of the Twenty-Eighth Conference on Uncertainty in Artificial Intelligence, and has been fully open-sourced on GitHub.

Actions are returned by the system as a verb, participant objects as noun phrases, properties of those objects as adjectival modifiers in those noun phrases, spatial relations between those participants as prepositional phrases, and characteristics of the event as prepositional-phrase adjuncts and adverbial modifiers.

Using an approach called event recognition, the research team was able to extract the information needed to create sentence descriptions from nearly 750 short videos.  This included recognition of object tracks (where things are going and from where), to whom or what objects were going, and changing body posture of the people in the videos.

For the language processing elements of the system, uses a simple vocabulary of 118 words: 1 coordination, 48 verbs, 24 nouns, 20 adjectives, 8 prepositions, 4 lexical prepositional phrases, 4 determiners, 3 particles, 3 pronouns, 2 adverbs, and 1 auxiliary.

Related articles DARPA Looks To New Form Of Computation That Mimics The Human Brain  Google and NASA's Quantum Artificial Intelligence Lab  Our Brains Shown To Act Computer-Like In New Situations

While it may seem at first to be a very straightforward activity to summarize a short video clip for a human it is far from easy for an artificial system to do.

Creating a descriptive sentence from video requires  recognizing the primary action being performed, because such actions are rendered as verbs and verbs serve as the central structure for sentences. However, event recognition alone is insufficient to generate the remaining components. Video In Sentence Out must recognize object classes in order to render nouns. But even object recognition alone is insufficient to generate meaning full sentences. The system must determine the roles that such objects play in the event.

The overall architecture of Video In Sentence Out first uses detectors for each object on each frame of the video. The objects are cross-checked for false positives at this stage by a number of sub-systems.  Then, dynamic programming algorithm is employed to select the main objects detected within the time flow of the action. These objects are then tracked by the system.

People in the videos are also checked and tracked, as well as postures and actions determined based on deep learning. Hidden Markov Models (HMMs) are then used to determine the verbs that will be used in the developing sentences.  Adjectives and prepositional phrases are processed into the final sentence generation.

To test the system, the human judges rated each video-sentence pair to assess whether the sentence was true of the video and whether it described the event depicted in the video. 26.7% of the video-sentence pairs were found to be true and 7.9% of the video-sentence pairs were deemed to be salient, covering the main elements seen in the video.

The video above represents highlights from the study. The full videos are available here.

By 33rd Square

Embed

 Artificial Intelligence  

With the initial report that a computer modeled after a 13-year-old boy had passed the Turing Test, followed much skepticism and doubt, Nuance has announced a new way to measure Artificial Intelligence with the Winograd Schema Challenge, an annually hosted hosted competition with Commonsense.org to push the boundaries of Artificial Intelligence.

The firm behind Siri, Nuance Communications has announced an annual competition to develop programs that can solve the Winograd Schema Challenge, a test developed by Hector Levesque, Professor of Computer Science at the University of Toronto, and winner of the 2013 IJCAI Award for Research Excellence.

The announcement was made at the 28th AAAI Conference in Quebec, Canada.

Nuance is sponsoring the yearly competition in cooperation with CommonsenseReasoning.org, a research group dedicated to furthering and promoting research in the field of formal commonsense reasoning. CommonsenseReasoning.org will organize, administer, and evaluate the Winograd Schema Challenge. The winning program that passes the test will receive a grand prize of $25,000. The test is designed to judge whether a program has truly modeled human level intelligence.

"The Winograd Schema Challenge provides us with a tool for concretely measuring research progress in commonsense reasoning, an essential element of our intelligent systems."

Artificial Intelligence (AI) has long been measured by the "Turing Test, " proposed in 1950 by one of the great pioneers of computer science, Alan Turing, who sought a way to determine whether a computer program exhibited human level intelligence. The test is considered passed if the program can convince a human that he or she is conversing with a human and not a machine. No system has ever passed the Turing Test, and most existing programs that have tried rely on considerable trickery to fool humans. Even the recently unveiled program modeling a 13-year-old boy, Eugene Goostman, has left many skeptical. These efforts have also suggested that the Turing Test may not be an ideal way to judge a machine's intelligence.

The Winograd Schema (WS) Challenge is an alternative to the Turing Test that provides a more accurate measure of genuine machine intelligence. Rather than base the test on the sort of short free-form conversation suggested by the Turing Test, the Winograd Schema Challenge poses a set of multiple-choice questions that have a form where the answers are expected to be fairly obvious to a layperson, but ambiguous for a machine without human-like reasoning or intelligence.

The schema is named after Terry Winograd, an American professor of computer science at Stanford University, and co-director of the Stanford Human-Computer Interaction Group and author of, Bringing Design to Softwareand Understanding Computers and Cognition: A New Foundation for Design. He is known within the philosophy of mind and artificial intelligence fields for his work on natural language using the SHRDLU program.

An example of a Winograd Schema question is the following: "The trophy would not fit in the brown suitcase because it was too big. What was too big? Answer 0: the trophy or Answer 1: the suitcase?" A human who answers these questions correctly typically uses his abilities in spatial reasoning, his knowledge about the typical sizes of objects, and other types of commonsense reasoning, to determine the correct answer.

According to Levesque, "The WS challenge does not allow a subject to hide behind a smokescreen of verbal tricks, playfulness, or canned responses. Assuming a subject is willing to take a WS test at all, much will be learned quite unambiguously about the subject in a few minutes."  Eugene Goostman was already close to "passing" the Turing Test when Levesque wrote about the Winograd Schema in 2011, clearly targeting AIs that were created to use deception to defeat the test.  In the case of Goostman, the fact that the creator made him a young boy with English as a second language was key factor in it fooling 30 percent of the judges it conversed with.

Related articles AI System Breaks CATCHA Turing Test  Marvin Minsky on Singularity 1 on 1: The Turing Test is a Joke!  Angry Birds To Be Used As New Turing Test

"There has been renewed interest in AI and Natural Language Processing (NLP) as a means of humanizing the complex technological landscape that we encounter in our day-to-day lives," said Charles Ortiz, Senior Principal Manager of AI and Senior Research Scientist, Natural Language and Artificial Intelligence Laboratory, Nuance Communications. "The Winograd Schema Challenge provides us with a tool for concretely measuring research progress in commonsense reasoning, an essential element of our intelligent systems. Competitions such as the Winograd Schema Challenge can help guide more systematic research efforts that will, in the process, allow us to realize new systems that push the boundaries of current AI capabilities and lead to smarter personal assistants and intelligent systems."

The test will be administered on a yearly basis by CommonsenseReasoning.org starting in 2015. The first submission deadline will be October 1, 2015. The 2015 Commonsense Reasoning Symposium, to be held at the AAAI Spring Symposium at Stanford from March 23-25, 2015, will include a special session for presentations and discussions on progress and issues related to this Winograd Schema Challenge. Contest details can be found at http://commonsensereasoning.org/winograd.html.

The winner that meets the baseline for human performance will receive a grand prize of $25,000. In the case of multiple winners, a panel of judges will base their choice on either further testing or examination of traces of program execution. If no program meets those thresholds, a first prize of $3,000 and a second prize of $2,000 will be awarded to the two highest scoring entries. In the case of teams, the prize will be given to the team lead whose responsibility will be to divide the prize among its teammates as appropriate.

Clearly defining exactly what intelligence is continues to be a both a philosophical and technical issue for defining human-level artificial intelligence tests.  By creating a variety of such tests, we will be better able to judge the progress of our AI systems, and, by extension, learn more about ourselves.


SOURCE  Business Wire

By 33rd Square

Embed

 Artificial Intelligence  

A Canadian predictive intelligence technology company has partnered with the Austrian Research Institute for Artificial Intelligence (OFAI) to collaborate on natural language processing and prediction research and development.

WordLogic, a predictive intelligence technology company that creates patented solutions for mobile devices, tablets and wearables, has partnered with the Austrian Research Institute for Artificial Intelligence (OFAI) to collaborate on natural language processing and prediction research and development which will be coordinated by WordLogic's CTO, Mark Dostie.

Since its inception in 1984, OFAI has conducted research in modelling and processing human language. This includes research and development of methods and tools for text processing, question answering, computational linguistics and conversational systems.

Related articles Richard Gordon On Building Intelligent Machines  Ben Goertzel On Aspects of Artificial General Intelligence  The New Era of Cognitive Computing

A key area the combined team will be focusing on is developing algorithmic methods that are capable of capturing and reproducing all major idiosyncrasies displayed by a language variety; be they syntactic, lexical or phonetic in nature. By using active learning techniques, this strategy allows for reduced manual effort by automatically choosing for annotation, by a human, those sentences that will result in the largest increase in quality and accuracy of the predicting system.

"WordLogic's new NLP enhanced prediction engine, we are moving to a new level of prediction. Prediction based not only on known linguistic word patterns but also enhanced with meta-patterns of the types of words and how they are organized."

According to the company's press release, the intent is to combine OFAI's expertise in the subject areas of Natural Language Processing (NLP) and artificial intelligence including machine learning, to contribute resources and research to further enhance the development of WordLogic's data structures and algorithms while WordLogic will provide the operational technologies that can predict these complex linguistic structures in real time.

"With the development of Gen4, WordLogic's new NLP enhanced prediction engine, we are moving to a new level of prediction. Prediction based not only on known linguistic word patterns but also enhanced with meta-patterns of the types of words and how they are organized. The staff and existing research at OFAI are uniquely qualified to understand what we have achieved and to help us reach entirely new levels in computational linguistics, natural language processing and language prediction," states Mark Dostie, CTO of WordLogic.




SOURCE  Wordlogic

By 33rd Square

Embed