Google Continues to Improve Voice Search
November 5, 2015
Google’s research arm continues to make progress on voice search. The Google Research Blog updates us in, “Google Voice Search: Faster and More Accurate.” The Google Speech Team begins by referring back to 2012, when they announced their Deep Neural Network approach. They have since built on that concept; the team now employs a couple of models built upon recurrent neural networks, which they note are fast and accurate: connectionist temporal classification and sequence discriminative (machine) training techniques. The write-up goes into detail about how speech recognizers work and what makes their latest iteration the best yet. I found the technical explanation fascinating, but it is too lengthy to describe here; please see the post for those details.
I am still struck when I see any article mention that an algorithm has taken the initiative. This time, researchers had to rein in their model’s insightful decision:
“We now had a faster and more accurate acoustic model and were excited to launch it on real voice traffic. However, we had to solve another problem – the model was delaying its phoneme predictions by about 300 milliseconds: it had just learned it could make better predictions by listening further ahead in the speech signal! This was smart, but it would mean extra latency for our users, which was not acceptable. We solved this problem by training the model to output phoneme predictions much closer to the ground-truth timing of the speech.”
At least the AI will take direction. The post concludes:
“We are happy to announce that our new acoustic models are now used for voice searches and commands in the Google app (on Android and iOS), and for dictation on Android devices. In addition to requiring much lower computational resources, the new models are more accurate, robust to noise, and faster to respond to voice search queries – so give it a try, and happy (voice) searching!”
We always knew natural-language communication with machines would present huge challenges, ones many said could never be overcome. It seems such naysayers were mistaken.
Cynthia Murrell, November 5, 2015
Sponsored by ArnoldIT.com, publisher of the CyberOSINT monograph
It Is Not a Bird in the Law Firm
November 3, 2015
In science-fiction, artificial intelligence is mostly toyed around with in robots and androids. Machines that bear artificial intelligence either try to destroy humanity for their imperfection or coexist with humanity in a manner that results in comedic situations. In reality, artificial intelligence exists in most everyday objects from a mobile phone to a children’s toy. Artificial intelligence is a much more common occurrence than we give our scientists credit for and it has more practical applications than we could imagine. According to PR Newswire one of the top artificial intelligence developers has made a new deal for their popular product, “RAVN Systems’ Artificial Intelligence Platform Is Deployed At Berwin Leighton Paisner.”
RAVN Systems is known for their top of line software in enterprise search, unstructured big data analytics, knowledge management, and, of course, artificial intelligence. The international law firm Berwin Leighton Paisner recently deployed RAVN Systems’s RAVN Applied Cognitive Engine (RAVN ACE). RAVN ACE will work in the law firm’s real estate practice, not as a realtor, but as the UK’s first contract robot. It will use cutting-edge AI to read and interpret information from documents, converting unstructured data into structured output. RAVN ACE will free up attorneys to complete more complex, less menial tasks.
“Matthew Whalley, Head of Legal Risk Consultancy at BLP commented, ‘The robot has fast become a key member of the team. It delivers perfect results every time we use it. Team morale and productivity has benefited hugely, and I expect us to create a cadre of contract robots throughout the firm. If the reaction to our first application is any indication, we will be leading the implementation of AI in the Law for some time to come.’ ”
RAVN ACE has more applications than writing real estate contracts. It can be deployed for financial services, media, telecommunications, and more. Taking over the menial tasks will save on time , allowing organizations to reinvest time into other projects.
Whitney Grace, November 3, 2015
Sponsored by ArnoldIT.com, publisher of the CyberOSINT monograph
Watch Anti-Money Laundering Compliances Sink
September 25, 2015
With a title like “AML-A Challenge Of Titanic Proportions” posted on Attivio metaphoric comparisons between the “ship of dreams” and icebergs is inevitable. Anti-money laundering compliances have seen an unprecedented growth between 2011-2014 of 53%, says KPMG’s Global Anti-Money Laundering (AML) Survey. The costs are predicted to increase by more than 25% in the next three years. The biggest areas that are requiring more money, include transaction monitoring systems, Know Your Customer systems, and recruitment/retention systems for AML staff.
The Titanic metaphor plays in as the White Star Line director Bruce Ismay, builder Thomas Andrew, and nearly all of the 3327 passengers believed the ship was unsinkable and the pinnacle of modern technology. The belief that humanity’s efforts would conquer Mother Nature was its downfall. The White Star Line did not prepare the Titanic for disaster, but AML companies are trying to prevent their ships are sinking. Except they cannot account for all the ways thieves can work around their system, just as the Titanic could not avoid the iceberg.
“Systems need to be smarter – even capable of learning patterns of transaction and ownership. Staff needs more productive ways of investigating and positively concluding their caseload. Alerting methods need to generate fewer ‘false positives’ – reducing the need for costly human investigation. New sources of information that can provide evidence need to come online faster and quickly correlate with existing data sources.”
The Titanic crew accidentally left the binoculars for the crow’s nest in England, which did not help the lookouts. The current AML solutions are like the forgotten binoculars and pervasive action needs to be taken to avoid the AML iceberg.
Whitney Grace, September 25, 2015
Sponsored by ArnoldIT.com, publisher of the CyberOSINT monograph
The AI Evolution
September 10, 2015
An article at WT Vox announces, “Google Is Working on a New Type of Algorithm Called ‘Thought Vectors’.” It sounds like a good use for a baseball cap with electrodes, a battery pack, WiFi, and a person who thinks great thoughts. In actuality, it’s a project based on the work of esteemed computer scientist Geoffrey E. Hinton, who has been exploring the idea of neural networks for decades. Hinton is now working with Google to create the sophisticated algorithm of our dreams (or nightmares, depending on one’s perspective).
Existing language processing software has come a very long way; Google Translate, for example, searches dictionaries and previously translated docs to translate phrases. The app usually does a passably good job of giving one the gist of a source document, but results are far from reliably accurate (and are often grammatically comical.) Thought vectors, on the other hand, will allow software to extract meanings, not just correlations, from text.
Continuing to use translation software as the example, reporter Aiden Russell writes:
“The technique works by ascribing each word a set of numbers (or vector) that define its position in a theoretical ‘meaning space’ or cloud. A sentence can be looked at as a path between these words, which can in turn be distilled down to its own set of numbers, or thought vector….
“The key is working out which numbers to assign each word in a language – this is where deep learning comes in. Initially the positions of words within each cloud are ordered at random and the translation algorithm begins training on a dataset of translated sentences. At first the translations it produces are nonsense, but a feedback loop provides an error signal that allows the position of each word to be refined until eventually the positions of words in the cloud captures the way humans use them – effectively a map of their meanings.”
But, won’t all efficient machine learning lead to a killer-robot-ruled dystopia? Hinton bats away that claim as a distraction; he’s actually more concerned about the ways big data is already being (mis)used by intelligence agencies. The man has a point.
Cynthia Murrell, September 10, 2015
Sponsored by ArnoldIT.com, publisher of the CyberOSINT monograph
Watson Still Has Much to Learn About Healthcare
July 9, 2015
If you’ve wondered what is taking Watson so long to get its proverbial medical degree, check out IEEE Spectrum’s article, “IBM’s Dr. Watson Will See You… Someday.” When IBM’s AI Watson won Jeopardy in 2011, folks tasked with dragging healthcare into the digital landscape naturally eyed the software as a potential solution, and IBM has been happy to oblige. However, “training” Watson in healthcare documentation is proving an extended process. Reporter Brandon Keim writes:
“Where’s the delay? It’s in our own minds, mostly. IBM’s extraordinary AI has matured in powerful ways, and the appearance that things are going slowly reflects mostly on our own unrealistic expectations of instant disruption in a world of Uber and Airbnb.”
Well that, and the complexities of our healthcare system. Though the version of Watson that beat Jeopardy’s human champion was advanced and powerful, tailoring it to manage medicine calls for a wealth of very specific tweaking. In fact, there are now several versions of “Doctor” Watson being developed in partnership with individual healthcare and research facilities, insurance companies, and healthcare-related software makers. The article continues:
“Watson’s training is an arduous process, bringing together computer scientists and clinicians to assemble a reference database, enter case studies, and ask thousands of questions. When the program makes mistakes, it self-adjusts. This is what’s known as machine learning, although Watson doesn’t learn alone. Researchers also evaluate the answers and manually tweak Watson’s underlying algorithms to generate better output.
“Here there’s a gulf between medicine as something that can be extrapolated in a straightforward manner from textbooks, journal articles, and clinical guidelines, and the much more complicated challenge of also codifying how a good doctor thinks. To some extent those thought processes—weighing evidence, sifting through thousands of potentially important pieces of data and alighting on a few, handling uncertainty, employing the elusive but essential quality of insight—are amenable to machine learning, but much handcrafting is also involved.”
Yes, incorporating human judgement is time-consuming. See the article for more on the challenges Watson faces in the field of healthcare, and for some of the organizations contributing to the task. We still don’t know how much longer it will take for the famous AI (and perhaps others like it) to dominate the healthcare field. When that day arrives, will it have been worth the effort?
Cynthia Murrell, July 9, 2015
Sponsored by ArnoldIT.com, publisher of the CyberOSINT monograph
Deep Learning System Surprises Researchers
June 24, 2015
Researchers were surprised when their scene-classification AI performed some independent study, we learn from Kurzweil’s article, “MIT Deep-Learning System Autonomously Learns to Identify Objects.”
At last December’s International Conference on Learning Representations, a research team from MIT demonstrated that their scene-recognition software was 25-33 percent more accurate than its leading predecessor. They also presented a paper describing the object-identification tactic their software chose to adopt; perhaps this is what gave it the edge. The paper’s lead author, and MIT computer science/ engineering associate professor, Antonio Torralba ponders the development:
“Deep learning works very well, but it’s very hard to understand why it works — what is the internal representation that the network is building. It could be that the representations for scenes are parts of scenes that don’t make any sense, like corners or pieces of objects. But it could be that it’s objects: To know that something is a bedroom, you need to see the bed; to know that something is a conference room, you need to see a table and chairs. That’s what we found, that the network is really finding these objects.”
Researchers being researchers, the team is investigating their own software’s initiative. The article tells us:
“In ongoing work, the researchers are starting from scratch and retraining their network on the same data sets, to see if it consistently converges on the same objects, or whether it can randomly evolve in different directions that still produce good predictions. They’re also exploring whether object detection and scene detection can feed back into each other, to improve the performance of both. ‘But we want to do that in a way that doesn’t force the network to do something that it doesn’t want to do,’ Torralba says.”
Very respectful. See the article for a few more details on this ambitious AI, or check out the researchers’ open-access paper here.
Cynthia Murrell, June 24, 2015
Sponsored by ArnoldIT.com, publisher of the CyberOSINT monograph
Don’t Fear the AI
May 14, 2015
Will intelligent machines bring about the downfall of the human race? Unlikely, says The Technium, in “Why I Don’t Worry About a Super AI.” The blogger details four specific reasons he or she is unafraid: First, AI does not seem to adhere to Moore’s law, so no Terminators anytime soon. Also, we do have the power to reprogram any uppity AI that does crop up and (reason three) it is unlikely that an AI would develop the initiative to reprogram itself, anyway. Finally, we should see managing this technology as an opportunity to clarify our own principles, instead of a path to dystopia. The blog opines:
“AI gives us the opportunity to elevate and sharpen our own ethics and morality and ambition. We smugly believe humans – all humans – have superior behavior to machines, but human ethics are sloppy, slippery, inconsistent, and often suspect. […] The clear ethical programing AIs need to follow will force us to bear down and be much clearer about why we believe what we think we believe. Under what conditions do we want to be relativistic? What specific contexts do we want the law to be contextual? Human morality is a mess of conundrums that could benefit from scrutiny, less superstition, and more evidence-based thinking. We’ll quickly find that trying to train AIs to be more humanistic will challenge us to be more humanistic. In the way that children can better their parents, the challenge of rearing AIs is an opportunity – not a horror. We should welcome it.”
Machine learning as a catalyst for philosophical progress—interesting perspective. See the post for more details behind this writer’s reasoning. Is he or she being realistic, or naïve?
Cynthia Murrell, May 14, 2015
Sponsored by ArnoldIT.com, publisher of the CyberOSINT monograph
The Elusive Video Recognition
April 22, 2015
Pictures and video still remain a challenge for companies like Google, Facebook, Apple, and more. These companies want to be able to have an algorithm pick up on the video or picture’s content without relying on tags or a description. The reasons are that tags are sometimes vague or downright incorrect about the content. VentureBeat reports that Google has invested a lot of funds and energy in a deep learning AI. The article is called “Watch Google’s Latest Deep Learning System Recognize Sports In YouTube Clips.”
The AI is park of a neural network that is constantly fed data and programmed to make predictions off the received content. Google’s researchers fed their AI consists of a convolutional neural network and it was tasked with watching sports videos to learn how to recognize objects and motions.
The researchers learned something and wrote a paper about it:
“ ‘We conclude by observing that although very different in concept, the max-pooling and the recurrent neural network methods perform similarly when using both images and optical flow,’ Google software engineers George Toderici and Sudheendra Vijayanarasimhan wrote in a blog post today on their work, which will be presented at the Computer Vision and Pattern Recognition conference in Boston in June.”
In short, Google is on its way to making video and images recognizable with neural networks. Can it tell the differences between colors, animals, people, gender, and activities yet?
Whitney Grace, April 22, 2015
Stephen E Arnold, Publisher of CyberOSINT at www.xenky.com
-
- Subscribe to Beyond Search
Feature archive
News archive
-
