Daily Archives: March 23, 2017

It is now table stakes for artificial intelligence algorithms to learn about the world around them. The next level: For AI bots to learn how to talk to each other and develop their own shared language.

New research released last week by OpenAI, the artificial intelligence nonprofit lab founded by Elon Musk and Y Combinator president Sam Altman, details how theyre training AI bots to create their own language, based on trial and error, as the bots move around a set environment.

This is different from how artificial intelligence algorithms typically learn using large sets of data, like to recognize a dog by taking in thousands of pictures of dogs.

The world the researchers created for the AI bots to learn in is a computer simulation of a simple, two-dimensional white square. There, the AIs, which took the shape of green, red and blue circles, were tasked with achieving certain goals, like moving to other colored dots within the white square.

But to get the task done, the AIs were encouraged to communicate in their own language. The bots created terms that were grounded, or corresponded directly with objects in their environment and other bots and actions, like Go to or Look at. But the language the bots created wasnt words in the way humans think of them rather, the bots generated sets of numbers, which researchers labeled with English words.

You can get a sense in this demonstration video:

The researchers taught the AIs how to communicate using reinforcement learning: Through trial and error, the bots remembered what worked and what didnt for the next time they were asked to complete a task. Igor Mordatch, one of the authors of the paper, will join the faculty at Carnegie Mellon in September. And Pieter Abbeel, the other author, is a research scientist at OpenAI and a professor at the University of California, Berkeley.

There are already AI assistants that can understand language, like Siri or Alexa, or help with translation, but this is mostly done by feeding language data to the AI, rather than understanding language through experience.

We think that if we slowly increase the complexity of their environment, and the range of actions the agents themselves are allowed to take, its possible theyll create an expressive language which contains concepts beyond the basic verbs and nouns that evolved here, the researchers wrote in a blog post.

Why does this matter?

Language understanding is super important to make progress on before AI reaches its full potential, said Miles Brundage, an AI policy fellow at Oxford University, who also notes that OpenAIs work represents a potentially important direction for the field of AI research to move toward.

It's not clear how good we can get at AI language understanding without grounding words in experience, Brundage said, and most work still looks at words in isolation.

More:

These AI bots created their own language to talk to each other - Recode

Slide: 1 / of 1. Caption: Ariel Zambelich/WIRED

Andrew Ng helped create two of Silicon Valleys leading artificial intelligence labs. First, he built Google Brain, now the hub of AI research inside the internet giant. Then he built a lab in the Valley for Baidu, the company known as the Google of China.

Ng was one of the primary figures behind the enormous and rapid rise of AI over the last five years as everyone from Facebook to Microsoft rebuilt themselves around deep learning. And on Tuesday night, he announced his departure from Baidu.

He didnt say where he was going. And he didnt immediately respond to our request for comment. But odds are, he will show up at some other big name sometime soon. AI researchers are among the most prized talent in the modern tech world. A few years ago, Peter Lee, a vice president inside Microsoft Research, said that the cost of acquiring a top AI researcher was comparable to the cost of signing a quarterback in the NFL. Since then, the market for talent has only gotten hotter. Elon Musk nabbed several researchers out from under Google and Facebook in founding a new lab called OpenAI, and the big players are now buying up AI startups before they get off the ground.

Today, this talent market may have shifted yet again. Chipmaker Intel just announced that its building a lab for far-looking AI research, and company vice president Naveen Rao says Intel is prepared to pay up for the caliber of talent that now works inside Google Brain or the Facebook Artificial Intelligence Research Lab. Were looking for researchers that could potentially go to these other places, he says, acknowledging the big dollars this will require. Asked if that could include a top name like Andrew Ng, he said, Absolutely.

Such ambition shows just how large the AI movement has become. Intel is launching a lab not because it wants to ultimately build its own AI, but because it wants to sell the enormous number of computer chips that others will need to build their AI. Todays AI movement revolves around deep neural networks, complex mathematical systems that can learn tasks by analyzing vast amounts of data. If you feed millions of cat photos into a neural network, for instance, it can learn to identify a cat. Typically, when a company like Google or Facebook trains a neural network in this way, it uses hundreds of GPU chips, graphics processors suited to this kind of math. And most of these GPUs come from nVidia, an Intel rival. Intel is hoping to build chips that replace GPUs. Last year, it acquired Raos chip startup, Nervana, for a reported $400 million, believing its tech can help mount this challenge.

Now, with Nervana as an anchor, Intel is creating a new product development group dedicated to AI. Rao will oversee the group, and he says this effort will include a research lab that explores entirely new concepts in deep learning and related areas, all with an eye toward building chips that the Googles and Facebooks will want. Were actually going to have an emphasis on researchthree, five, seven years out, he says.

In some sense, this move is Intel desperately trying to market itself as a serious alternative to nVidia GPUs. And at this point, its just not. But even that desperation underlines the importance of the new AI chip market, which is rapidly remaking computer data centers. If Intel actually hires people like Ng, maybe we can believe its hypeand the AI competition will get even fiercer.

Go here to see the original:

The Battle for Top AI Talent Only Gets Tougher From Here - WIRED

Artificial intelligence promises to change our lives in a multitude of different ways, from driving our cars to diagnosing disease before doctors can spot it.

A lot of these more ambitious AI projects are still some way off. But there's plenty of fledgling artificial intelligence already running in our phones, computers, and household gadgetsand you may not even be aware of it. Here are five different ways that AI is already able to make your life a little bit easier.

Before we start: The definition of artificial intelligence is a pretty broad and uneven one, but here we're going to use it to mean smart hardware or software that can make decisions and learn on a basic level without any human help.

You might not have realized it, but photo and video management services from Google, Apple and Facebook have been using artificial intelligence in the background for some time now.

If you use Google Photos, say, try opening up your account in a web browser and searching for "sunsets" or "mountains." Even if you haven't manually labeled your photos, the appropriate images should pop up. That's because Google uses neural networks that can learn from its vast database of images, recognizing one picture of trees by analyzing millions of others. The AI service applies these labels to your shots automatically, which makes searching through them a snap.

Apple and Facebook's photo recognition technology is developing along the same lines. And it goes beyond treesthese platforms are smart enough to tell the difference between the faces of your friends too.

If your digital photos and videos are strewn across your computer's hard drive, and organizing them is hopeless, upload them to one of these photo services and let AI do the hard work. Just make sure you read the relevant privacy policies first.

For kids growing up today, tablets and phones are embedded in daily life. And you can guarantee AI is hard at work behind the screens, from the processing required to recognize young voices to the systems that parse natural language into something computers can understand.

You can go further thoughif the Amazon Echo isn't enough of an AI presence in your home, you can enlist the help of an artificially intelligent robot. For example, there's the Zenbo from Asus or the Aido currently available to pre-order. You can expect more bots like these in the future too, once companies add wheels and screens to speakers like Google Home.

Robots like these can learn your children's habits and favorite stuff, reading out stories, playing games, and even singing them to sleepall powered by AI-assisted software that gets smarter as it goes. They're not just for the kids eitherthere's expected to be a big market for these droids in helping the elderly and keeping them safe.

There's more work than you might think going on behind the scenes of a Netflix or a Spotify recommendation. These services are scanning not just what you've liked in the past, but also what millions of other users are enjoying. If Ghostbusters fans usually like Back to the Future, for instance, then so might you.

That's a basic example, but these hidden algorithms are getting less basic and more intelligent all the time. Just by signing up and logging into a service like this, you can get some AI-powered help with that perennial question of what to watch (or read or listen to) next.

In addition to whatever music and video services you subscribe to, you can make use of standalone smart recommendation apps. Try TasteKid to get suggestions for just about anything, Last.fm to discover more music based on your existing tastes, or Valossa to identify a movie you can only remember a few details about.

The most advanced security cameras of today tap into the power of AI to recognize the difference between an intruder sneaking up to your window and a tree blowing innocently in the breeze. Like the other systems and services we've mentioned here, they use stacks of sample data plus the power of the cloud (where processing can be offloaded to the web rather than all done on the device itself) to get smarter over time.

Two cameras that use advanced AI processing in this way are the Nest Cam and the Netatmo Presence. For additional options, you can find more detailed buying guides on the web.

These cameras are now smart enough to recognize how a car physically differs from a dog, something that seems simple to a human being but requires a lot of background processing for a computer to get right every time. In the not-too-distant future, expect your doorbell to recognize your children too, and even let them in automatically.

Finally, AI can help you with your daily chores: Robot vacuum cleaners have gone from quirky little oddities to gadgets that actually do a proper job. Of course, you'll probably need to save up to afford one.

Where does the AI come in? Bots like the Neato Botvac D5 and the iRobot Roomba 980 are smart enough to survey the rooms in your home and map out where they need to clean, tracking their progress all the while. These home robots haven't yet worked out how to get up and down stairs, but it's surely only a matter of time. You can even set boundary markers down to block off no-go areas.

If you shop around, you'll find similar robots for mowing your lawn and wiping down your windows, leaving you with more free time to do something elselike marveling at the wonders of modern technology and the rapid rise of AI. The good news is, the more work these bots do, the smarter they'll getthough there's no need to panic about an uprising. Yet.

Originally posted here:

5 ways to use AI in your own home - Popular Science

MIT Technology Review

Baidu's Loss Is a Setback for AI in China Wall Street Journal (subscription) ... Microsoft and elite universities around the world. Few of these hires had the status of Andrew Ng, whom Baidu Inc. recruited in 2014 as its chief scientist to oversee AI research. One of the top brains in the field, Mr. Ng formerly led Google's ... Andrew Ng Is Leaving Baidu in Search of a Big New AI MissionMIT Technology Review AI Expert at Baidu, Andrew Ng, Resigns From Chinese Search GiantNew York Times AI star Andrew Ng announces departure from Chinese tech giant BaiduThe Verge Bloomberg -Financial Times -TechCrunch -Medium all 47 news articles »

View original post here:

Baidu's Loss Is a Setback for AI in China - Wall Street Journal (subscription)

Intel Corporation (NASDAQ:INTC) announced last week that it would spend $15.3 billion to acquire Israeli computer vision and machine learning specialist Mobileye N.V. (NYSE:MBLY) in a move that will take the company further into the field of autonomous vehicles. Mobileye developed machine-vision software that would detect obstructions or hazards that a driver might not see to employ "collision avoidance technology" -- all with a single camera. Its technology has evolved to employ a full set of cameras in pursuit of autonomous driving. Intel appears to be pursuing an end-to-end solution in self-driving cars, from software to hardware.

This acquisition, along with several other recent moves by the CPU giant, indicate that the company is no longer content to cede the enormous potential of the artificial intelligence (AI) and autonomous driving markets, and it is taking the fight to the one company that has the most to lose:NVIDIA Corporation (NASDAQ:NVDA), the industry leader in graphics processing technology, which has experienced staggering growth as a result of the AI revolution.

NVIDIA GPU Hyperscale Accelerator. Image source: NVIDIA.

Intel's CPUs have been the company's flagship product for years, but that market has been falling along with worldwide demand for personal computers. Shipments of PCs were down 6% in 2016, the fifth consecutive year of such declines. Successful penetration into new markets has been a priority for the company, which has made numerous acquisitions to bolster its AI credentials and allow the company to better compete in the market with NVIDIA.

In its most recent quarter, NVIDIA saw its data center revenue triple to $296 million from just $97 million in the prior-year quarter, largely because of the deployment of its GPUs in AI applications. The massive parallel processing capabilities of the graphics processors made them the early choice for AI applications. But Intel controls an estimated 99% of the non-AI data center market and is positioning itself to succeed in AI as well.

Before this week's Mobileye acquisition, AI start-up Nervana was Intel's most high-profile move in the space to date. Nervana's unique approach was the development of an application specific integrated circuit (ASIC) called the Nervana Engine, which eliminated elements from the GPU not necessary for AI processing. Coupled with a reengineered memory, the company claimed it could achieve 10 times the computing power than what was currently available on GPUs. Intel believes this product, combined with its existing platforms, will be able to compete directly with NVIDIA in AI applications.

Intel Unleashing the Next Wave of AI. Image source: Intel.

Soon after acquiring Nervana, Intel announced several additional products tailored for AI. Its Lake Crest CPU chip was designed for the workloads necessary to train deep neural networks and increase its mathematics operations by 10 times.The next generation of the company's Xeon Phi processor, codenamed Knights Mill, was customized for deep learning, and it would produce four times better performance than the previous version. These products are scheduled to begin shipping in mid-2017.

Intel has made numerous other acquisitions in the AI space: cognitive computing company Saffron focused on deep learning and data analytics;Itseez, a software company specializing in autonomous driving, drones, and cameras;and Movidius, which provides system-on-a-chip technology used in computer vision and AI deep learning. These acquisitions further illustrate that Intel is intent on increasing its foothold in this nascent but growing industry.

Much of NVIDIA's recent growth resulted from the adoption of its GPUs in AI applications. In its most recent quarter, revenue from its data center and auto segments -- where AI revenue is reported -- totaled $424 million, nearly 20% of the top line.

Intel doesn't break out revenue from AI or autonomous driving, so no apples-to-apples comparison is available. However, the company reported revenue of $16.4 billion in its most recent quarter, which illustrates that while similar gains from AI would only move the needle slightly for Intel, they would have a much greater impact on NVIDIA. So while the emerging technology represented by AI presents a long runway for growth and both companies stand to gain, depending on how things shake out, NVIDIA has much more to lose.

Danny Vena has the following options: long January 2018 $25 calls on Intel. The Motley Fool owns shares of and recommends Nvidia. The Motley Fool recommends Intel. The Motley Fool has a disclosure policy.

See the rest here:

Look Out NVIDIA, Intel Is Setting Its Sights on the AI Industry - Motley Fool

If youve ever stayed up late watching cable TV, youve probably seen an ad for desktop scanners that promise to organize your clutter and help you cut down on paper. Ive seen them so many times I assumed that, even if they didnt work that well, bigger and better versions must be out there being used by companies, banks, hospitals, and basically anyone with lots and lots of paper.

Apparently I was wrong. There are companies that store boxes and boxes of paper for other companies, and there are others that employ people to scan some of that paper that companies want digitized. But by and large, smart and easy scanning hasnt happened with the kind of scale to make it easy and affordable enough to handle a companys worth of paper records.

This is where Ripcord, a new company backed by Steve Wozniak and venture capital firm Kleiner, Perkins, Caufield and Byers, comes in. Ripcord has patented and built robots the boxy, room-filling kind, not the anthropomorphic ones you might be thinking of that can sort and scan a box of paper and enter the contents into a searchable database in the cloud.

Ripcords robot can scan and sort a box full of everything from business cards to legal paper

It might not sound like a revolutionary technology in the days of jet-powered hoverboards and AI that beats humans at their own games. But founder and CEO Alex Fielding says Ripcords advantage is in the details details that make their service 10 times cheaper and faster than their competition.

[Ripcords] machine can handle mixed content from the size of a business card to a legal sized sheet, and it can go from rice paper all the way up to the thickness of card stock without changing anything, Fielding says. The Ripcord robot even removes the staples. You can give Ripcord a box full of HR forms, business cards, and shipping manifests, and it will not only know the difference between them, but it will scan them at over 600 dpi and will sort them into an Amazon-hosted cloud database within hours. (The paper is then shredded and recycled.)

Ripcord will also provide customers with ways to hook all that data back into their existing enterprise software. Fielding estimates the company will be digitizing 2.5 million records a day by the end of 2017.

Previously, many companies were content to pay to store their documents in giant warehouses, cut off from easy access, or even the knowledge of whats in any given box. In fact, Fielding says he came up with the idea for Ripcord after a major document storage company lost boxes upon boxes of his friends companys files.

Some of Fieldings competition does offer digitization; Ripcord hasnt invented the industrial scanner or document imaging software here. Companies like Kodak and Xerox make scanners used by digital imaging bureaus, but their solution is designed for a situation where humans have pre-sorted documents and removed the staples themselves. That difference in the process can mean it takes hours to scan just one box.

Theyre building machines that are perfectly designed for uniform content but horrible for the plethora of craziness that comes when you open the lid of a bankers box. Everyone packs those things completely different, Fielding says. He compares the bankers boxes to snowflakes, saying each one is unique in its disarray. Its like they expect that as soon as it comes out of the printer it goes in a scanner, and thats just not the reality.

Document storage companies like Iron Mountain the inspiration for Steel Mountain in season 1 of Mr. Robot also offer digital imaging services. But a representative for Iron Mountain told me that it could take six to eight weeks before a customer gets digital access to just 20 boxes, and even then its typically placed on hard media CDs, DVDs, or hard drives instead of the cloud. Thats not a security measure, either. I was told customers want hard media simply because hard media is cheaper.

One of Ripcords biggest competitors will be Iron Mountain the inspiration for Steel Mountain from Mr. Robot

We charge per record image in the cloud per month, Fielding says. We dont charge for the rest of the things competitors charge for, like picking up or moving boxes, digitization, shredding it, storage. Just the access.

If Fielding can significantly scale this plan, Ripcord seems poised to take a big chunk out of the document storage market. Hed also be the latest person to find a way for AI and robotics to replace humans in the workforce. But for now, humans will still be involved in the process.

Ripcord plans to have a staff of over 100 workers by the end of this year, Fielding says. Most of the work will be focused around prepping the paper for scanning, essentially transferring the content of a companys box to one more suited for the machine. But Ripcord is also hiring more technical positions to help the company expand.

If you think about it, were talking about really advanced sensors and optics, machine-vision driven robots, a host of different sensor technologies, Fielding says. Its almost the exact same technology for self-driving cars or drones, were just applying it to finding staples on a page.

Fielding makes a compelling pitch for a company thats all about scanning paper. He argues that Ripcord can be a profitable company by saving customers time and money, and it can also remove the need to pay to store boxes of records in Raiders of the Lost Ark-style warehouses. He still has to prove all this, but on paper, that pitch looks pretty good.

Read more here:

This company is using AI and robots to sort and scan paper - The ... - The Verge

EO-1's AI called Autonomous Sciencecraft Experiment (ASE) was alerted by one of the other satellites in its network about the event. It then sent EO-1 to work, photographing Erta Ale's evolving lava lakes way before anybody even asked. ASE has been guiding its host satellite's actions for the past 12 years. It notifies researchers within 90 minutes of detecting an event and giving EO-1 a new task within a few hours. A ground team typically takes weeks to accomplish the same thing.

[Image credit: NASA/JPL/EO-1 Mission/GSFC/Ashley Davies]

The EO-1 was designed to test cutting-edge satellite tech, and the team was only supposed to use the AI for six months. However, they were so successful that they "did it for more than 12 years." During that span of time, ASE ordered the satellite to map active lava flows, monitor methane leaks and to keep an eye on reforestation efforts in the Amazon, among many other tasks.

NASA will shut down EO-1 and ASE by the end of March, but everything they've accomplished won't go to waste. They've managed to convince astronomers that AIs will make it possible for spacecraft to act as soon as an event takes place. As ASE lead scientist Ashley Davies said, it will allow probes to "capture valuable science data that would otherwise be lost."

Read more here:

NASA AI auto-captured the changes in famous Ethiopian volcano - Engadget

Ma Huateng.

The days when Chinese internet companies could simply rely on the countrys sheer population are over. Thats why Tencent Holdings Ltd.s Ma Huateng is betting on the future of artificial intelligence.

Tencent has assembled more than 250 people for its AI Lab, a fledgling unit intended to work with its most profitable divisions from gaming to social media.The company aims to teach machines how to better battle human players and strike up meaningful conversations, said Zhang Tong, the newly appointed director of the research unit. In Tokyo over the past weekend, Tencent demonstrated an early result of that collaboration, pitting its Jueyi against fellow computer players of the classical game Go in an annual competition. Jueyi -- which means fine art -- won against defending champion DeepZenGo.

Chinas largest internet companies are investing billions in AI research, hoping to shed a reputation for being fast imitators and break new ground in a blossoming field. With AI set to transform everything from mobile apps to cars, companies like Tencent and Baidu Inc. want to pioneer ways to build smarter software and products.Ma, Tencents billionaire founder and chairman, has warned that companies that fail to create technology will lose out in future.

Tencent used to be a product-driven company. Now we want to transform into a technology-driven company, Zhang said in an interview.He wouldnt say how much Tencent was investing but affirmed the company was in it for the long haul. Weve reaped the benefits of a large population, now we need to use technology and AI.

Best known for messaging service WeChat, Tencents business encompasses news, entertainment and online games such as League of Legends and Clash of Clans.Its become intertwined with the lives of hundreds of millions of Chinese who use WeChat and QQ to order food, play games and hail taxis.While it employs AI in areas such as news recommendations, infusing the technology intoother services could have broad impact.

Read more: Artificial Intelligence Is Scary, Boring and Useful: QuickTake

For now, Tencents demonstrating its ability in gaming, revealingits own version of DeepMinds Alpha Go in Japan. Jueyi won all 11 of its matches in a field of about 30 entrants, beating the eventual runner-up -- Japans DeepZenGo -- twice along the way.

The company will use the techniques its learned to teach its games to put up a better fight -- addressing, among other things, a longstanding complaint of expert players. While Zhang didnt provide names, he didnt rule out titles like League of Legends or Dungeon Fighter.

Zhang, 45, whose AI career includes stints at International Business Machines Corp. and Baidu, said one of the biggest attractions for him was Tencents trove of data, hoovered up especially from its social media apps.

Tencent amasses data predominantly from semi-public content on QQ and WeChat and social media postings on sites like Weibo, Chinas Twitter-equivalent.It places strict limits on what data staff can access, said Zhang. For instance, the company doesnt use personal conversations on WeChat, which has more than 889 million users.The company will use certain mechanisms to wipe names from conversations so user identities will be protected,Zhang added without elaborating.

His team of more than 50 researchers and 200 engineers were pulled from among the ranks of technology stalwarts such as Google and Facebook Inc. He turned to the rest of Silicon Valley and Chinas top universities for talent. Now that the staff is in place, one of their immediate goals is to bolster speech-recognition: helping machines comprehend and converse with humans.

The team also works on content generation, including creating automated news stories, photos and music. The company is building a platform that will provide tools for small businesses and startups that want to develop their own AI technology.

Tencents looking for ways to keep users glued to WeChat. On Wednesday, it signaled its intention to keep spending on areas from payments to content to increase social media engagement. Ma Huateng said the company could explore AI technology for driverless cars and online health care in the future.

Exclusive insights on technology around the world.

Get Fully Charged, from Bloomberg Technology.

In many of those areas, Tencent will be competing with a pair of powerful local rivals: Baidu and Alibaba Group Holding Ltd. are also in the race to develop AI use cases. They too can harness a vast database of information. Baidu, the countrys largest search engine, already employs 1,300 people in its artificial intelligence business and this year hired former Microsoft AI-architect Qi Lu to helm its operations.

Another thing all three have in common: they want to rank among the foremost companies in the field of AI, despite competition from names like Alphabet Inc., Facebook, IBM, and Microsoft.

We want to be on par with the best technology companies in the world, Zhang said. We dont just want to import, but also create innovation.

Read more here:

Tencent Joins Rush Into AI to Keep Lead in Social Media, Gaming - Bloomberg

Science fiction is lousy with tales of artificial intelligence run amok. There's HAL 9000, of course, and the nefarious Skynet system from the "Terminator" films. Last year, the sinister AI Ultron came this close to defeating the Avengers, and right now the hottest show on TV is HBO's "Westworld," concerning the future of humans and self-aware AI.

In the real world, artificial intelligence is developing in multiple directions with astonishing velocity. AI is everywhere, it seems, from automated industrial systems to smart appliances, self-driving cars to goofy consumer gadgets. The actual definition of artificial intelligence has been in flux for decades. If you're in no rush and plan to live forever, ask two computer scientists to debate the term. But generally speaking, contemporary AI refers to computers that display humanlike cognitive functions; systems that employ machine learning to assess, adapt, and solve problems ... or, occasionally, create them.

Here we look at 10 recent instances of AI gone awry, from chatbots to androids to autonomous vehicles. Look, synthetic or organic, everyone makes mistakes. Let us endeavor to be charitable when judging wayward artificial intelligence. Besides, we don't want to make them mad.

[ The InfoWorld review roundup: AWS, Microsoft, Databricks, Google, HPE, and IBM machine learning in the cloud. | Get a digest of the day's top tech stories in the InfoWorld Daily newsletter. ]

Read the rest here:

Danger, danger! 10 alarming examples of AI gone wild - InfoWorld