Tag Archives: popular
Back in 2016, somewhere in a Google-owned warehouse, more than a dozen robotic arms sat for hours quietly grasping objects of various shapes and sizes. For hours on end, they taught themselves how to pick up and hold the items appropriately—mimicking the way a baby gradually learns to use its hands.
Now, scientists from MIT have made a new breakthrough in machine learning: their new system can not only teach itself to see and identify objects, but also understand how best to manipulate them.
This means that, armed with the new machine learning routine referred to as “dense object nets (DON),” the robot would be capable of picking up an object that it’s never seen before, or in an unfamiliar orientation, without resorting to trial and error—exactly as a human would.
The deceptively simple ability to dexterously manipulate objects with our hands is a huge part of why humans are the dominant species on the planet. We take it for granted. Hardware innovations like the Shadow Dexterous Hand have enabled robots to softly grip and manipulate delicate objects for many years, but the software required to control these precision-engineered machines in a range of circumstances has proved harder to develop.
This was not for want of trying. The Amazon Robotics Challenge offers millions of dollars in prizes (and potentially far more in contracts, as their $775m acquisition of Kiva Systems shows) for the best dexterous robot able to pick and package items in their warehouses. The lucrative dream of a fully-automated delivery system is missing this crucial ability.
Meanwhile, the Robocup@home challenge—an offshoot of the popular Robocup tournament for soccer-playing robots—aims to make everyone’s dream of having a robot butler a reality. The competition involves teams drilling their robots through simple household tasks that require social interaction or object manipulation, like helping to carry the shopping, sorting items onto a shelf, or guiding tourists around a museum.
Yet all of these endeavors have proved difficult; the tasks often have to be simplified to enable the robot to complete them at all. New or unexpected elements, such as those encountered in real life, more often than not throw the system entirely. Programming the robot’s every move in explicit detail is not a scalable solution: this can work in the highly-controlled world of the assembly line, but not in everyday life.
Computer vision is improving all the time. Neural networks, including those you train every time you prove that you’re not a robot with CAPTCHA, are getting better at sorting objects into categories, and identifying them based on sparse or incomplete data, such as when they are occluded, or in different lighting.
But many of these systems require enormous amounts of input data, which is impractical, slow to generate, and often needs to be laboriously categorized by humans. There are entirely new jobs that require people to label, categorize, and sift large bodies of data ready for supervised machine learning. This can make machine learning undemocratic. If you’re Google, you can make thousands of unwitting volunteers label your images for you with CAPTCHA. If you’re IBM, you can hire people to manually label that data. If you’re an individual or startup trying something new, however, you will struggle to access the vast troves of labeled data available to the bigger players.
This is why new systems that can potentially train themselves over time or that allow robots to deal with situations they’ve never seen before without mountains of labelled data are a holy grail in artificial intelligence. The work done by MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) is part of a new wave of “self-supervised” machine learning systems—little of the data used was labeled by humans.
The robot first inspects the new object from multiple angles, building up a 3D picture of the object with its own coordinate system. This then allows the robotic arm to identify a particular feature on the object—such as a handle, or the tongue of a shoe—from various different angles, based on its relative distance to other grid points.
This is the real innovation: the new means of representing objects to grasp as mapped-out 3D objects, with grid points and subsections of their own. Rather than using a computer vision algorithm to identify a door handle, and then activating a door handle grasping subroutine, the DON system treats all objects by making these spatial maps before classifying or manipulating them, enabling it to deal with a greater range of objects than in other approaches.
“Many approaches to manipulation can’t identify specific parts of an object across the many orientations that object may encounter,” said PhD student Lucas Manuelli, who wrote a new paper about the system with lead author and fellow student Pete Florence, alongside MIT professor Russ Tedrake. “For example, existing algorithms would be unable to grasp a mug by its handle, especially if the mug could be in multiple orientations, like upright, or on its side.”
Class-specific descriptors, which can be applied to the object features, can allow the robot arm to identify a mug, find the handle, and pick the mug up appropriately. Object-specific descriptors allow the robot arm to select a particular mug from a group of similar items. I’m already dreaming of a robot butler reliably picking my favourite mug when it serves me coffee in the morning.
Google’s robot arm-y was an attempt to develop a general grasping algorithm: one that could identify, categorize, and appropriately grip as many items as possible. This requires a great deal of training time and data, which is why Google parallelized their project by having 14 robot arms feed data into a single neural network brain: even then, the algorithm may fail with highly specific tasks. Specialist grasping algorithms might require less training if they’re limited to specific objects, but then your software is useless for general tasks.
As the roboticists noted, their system, with its ability to identify parts of an object rather than just a single object, is better suited to specific tasks, such as “grasp the racquet by the handle,” than Amazon Robotics Challenge robots, which identify whole objects by segmenting an image.
This work is small-scale at present. It has been tested with a few classes of objects, including shoes, hats, and mugs. Yet the use of these dense object nets as a way for robots to represent and manipulate new objects may well be another step towards the ultimate goal of generalized automation: a robot capable of performing every task a person can. If that point is reached, the question that will remain is how to cope with being obsolete.
Image Credit: Tom Buehler/CSAIL Continue reading →
Recently, I picked up Kai-Fu Lee’s newest book, AI Superpowers.
Kai-Fu Lee is one of the most plugged-in AI investors on the planet, managing over $2 billion between six funds and over 300 portfolio companies in the US and China.
Drawing from his pioneering work in AI, executive leadership at Microsoft, Apple, and Google (where he served as founding president of Google China), and his founding of VC fund Sinovation Ventures, Lee shares invaluable insights about:
The four factors driving today’s AI ecosystems;
China’s extraordinary inroads in AI implementation;
Where autonomous systems are headed;
How we’ll need to adapt.
With a foothold in both Beijing and Silicon Valley, Lee looks at the power balance between Chinese and US tech behemoths—each turbocharging new applications of deep learning and sweeping up global markets in the process.
In this post, I’ll be discussing Lee’s “Four Waves of AI,” an excellent framework for discussing where AI is today and where it’s going. I’ll also be featuring some of the hottest Chinese tech companies leading the charge, worth watching right now.
I’m super excited that this Tuesday, I’ve scored the opportunity to sit down with Kai-Fu Lee to discuss his book in detail via a webinar.
With Sino-US competition heating up, who will own the future of technology?
Let’s dive in.
The First Wave: Internet AI
In this first stage of AI deployment, we’re dealing primarily with recommendation engines—algorithmic systems that learn from masses of user data to curate online content personalized to each one of us.
Think Amazon’s spot-on product recommendations, or that “Up Next” YouTube video you just have to watch before getting back to work, or Facebook ads that seem to know what you’ll buy before you do.
Powered by the data flowing through our networks, internet AI leverages the fact that users automatically label data as we browse. Clicking versus not clicking; lingering on a web page longer than we did on another; hovering over a Facebook video to see what happens at the end.
These cascades of labeled data build a detailed picture of our personalities, habits, demands, and desires: the perfect recipe for more tailored content to keep us on a given platform.
Currently, Lee estimates that Chinese and American companies stand head-to-head when it comes to deployment of internet AI. But given China’s data advantage, he predicts that Chinese tech giants will have a slight lead (60-40) over their US counterparts in the next five years.
While you’ve most definitely heard of Alibaba and Baidu, you’ve probably never stumbled upon Toutiao.
Starting out as a copycat of America’s wildly popular Buzzfeed, Toutiao reached a valuation of $20 billion by 2017, dwarfing Buzzfeed’s valuation by more than a factor of 10. But with almost 120 million daily active users, Toutiao doesn’t just stop at creating viral content.
Equipped with natural-language processing and computer vision, Toutiao’s AI engines survey a vast network of different sites and contributors, rewriting headlines to optimize for user engagement, and processing each user’s online behavior—clicks, comments, engagement time—to curate individualized news feeds for millions of consumers.
And as users grow more engaged with Toutiao’s content, the company’s algorithms get better and better at recommending content, optimizing headlines, and delivering a truly personalized feed.
It’s this kind of positive feedback loop that fuels today’s AI giants surfing the wave of internet AI.
The Second Wave: Business AI
While internet AI takes advantage of the fact that netizens are constantly labeling data via clicks and other engagement metrics, business AI jumps on the data that traditional companies have already labeled in the past.
Think banks issuing loans and recording repayment rates; hospitals archiving diagnoses, imaging data, and subsequent health outcomes; or courts noting conviction history, recidivism, and flight.
While we humans make predictions based on obvious root causes (strong features), AI algorithms can process thousands of weakly correlated variables (weak features) that may have much more to do with a given outcome than the usual suspects.
By scouting out hidden correlations that escape our linear cause-and-effect logic, business AI leverages labeled data to train algorithms that outperform even the most veteran of experts.
Apply these data-trained AI engines to banking, insurance, and legal sentencing, and you get minimized default rates, optimized premiums, and plummeting recidivism rates.
While Lee confidently places America in the lead (90-10) for business AI, China’s substantial lag in structured industry data could actually work in its favor going forward.
In industries where Chinese startups can leapfrog over legacy systems, China has a major advantage.
Take Chinese app Smart Finance, for instance.
While Americans embraced credit and debit cards in the 1970s, China was still in the throes of its Cultural Revolution, largely missing the bus on this technology.
Fast forward to 2017, and China’s mobile payment spending outnumbered that of Americans’ by a ratio of 50 to 1. Without the competition of deeply entrenched credit cards, mobile payments were an obvious upgrade to China’s cash-heavy economy, embraced by 70 percent of China’s 753 million smartphone users by the end of 2017.
But by leapfrogging over credit cards and into mobile payments, China largely left behind the notion of credit.
And here’s where Smart Finance comes in.
An AI-powered app for microfinance, Smart Finance depends almost exclusively on its algorithms to make millions of microloans. For each potential borrower, the app simply requests access to a portion of the user’s phone data.
On the basis of variables as subtle as your typing speed and battery percentage, Smart Finance can predict with astounding accuracy your likelihood of repaying a $300 loan.
Such deployments of business AI and internet AI are already revolutionizing our industries and individual lifestyles. But still on the horizon lie two even more monumental waves— perception AI and autonomous AI.
The Third Wave: Perception AI
In this wave, AI gets an upgrade with eyes, ears, and myriad other senses, merging the digital world with our physical environments.
As sensors and smart devices proliferate through our homes and cities, we are on the verge of entering a trillion-sensor economy.
Companies like China’s Xiaomi are putting out millions of IoT-connected devices, and teams of researchers have already begun prototyping smart dust—solar cell- and sensor-geared particulates that can store and communicate troves of data anywhere, anytime.
As Kai-Fu explains, perception AI “will bring the convenience and abundance of the online world into our offline reality.” Sensor-enabled hardware devices will turn everything from hospitals to cars to schools into online-merge-offline (OMO) environments.
Imagine walking into a grocery store, scanning your face to pull up your most common purchases, and then picking up a virtual assistant (VA) shopping cart. Having pre-loaded your data, the cart adjusts your usual grocery list with voice input, reminds you to get your spouse’s favorite wine for an upcoming anniversary, and guides you through a personalized store route.
While we haven’t yet leveraged the full potential of perception AI, China and the US are already making incredible strides. Given China’s hardware advantage, Lee predicts China currently has a 60-40 edge over its American tech counterparts.
Now the go-to city for startups building robots, drones, wearable technology, and IoT infrastructure, Shenzhen has turned into a powerhouse for intelligent hardware, as I discussed last week. Turbocharging output of sensors and electronic parts via thousands of factories, Shenzhen’s skilled engineers can prototype and iterate new products at unprecedented scale and speed.
With the added fuel of Chinese government support and a relaxed Chinese attitude toward data privacy, China’s lead may even reach 80-20 in the next five years.
Jumping on this wave are companies like Xiaomi, which aims to turn bathrooms, kitchens, and living rooms into smart OMO environments. Having invested in 220 companies and incubated 29 startups that produce its products, Xiaomi surpassed 85 million intelligent home devices by the end of 2017, making it the world’s largest network of these connected products.
One KFC restaurant in China has even teamed up with Alipay (Alibaba’s mobile payments platform) to pioneer a ‘pay-with-your-face’ feature. Forget cash, cards, and cell phones, and let OMO do the work.
The Fourth Wave: Autonomous AI
But the most monumental—and unpredictable—wave is the fourth and final: autonomous AI.
Integrating all previous waves, autonomous AI gives machines the ability to sense and respond to the world around them, enabling AI to move and act productively.
While today’s machines can outperform us on repetitive tasks in structured and even unstructured environments (think Boston Dynamics’ humanoid Atlas or oncoming autonomous vehicles), machines with the power to see, hear, touch and optimize data will be a whole new ballgame.
Think: swarms of drones that can selectively spray and harvest entire farms with computer vision and remarkable dexterity, heat-resistant drones that can put out forest fires 100X more efficiently, or Level 5 autonomous vehicles that navigate smart roads and traffic systems all on their own.
While autonomous AI will first involve robots that create direct economic value—automating tasks on a one-to-one replacement basis—these intelligent machines will ultimately revamp entire industries from the ground up.
Kai-Fu Lee currently puts America in a commanding lead of 90-10 in autonomous AI, especially when it comes to self-driving vehicles. But Chinese government efforts are quickly ramping up the competition.
Already in China’s Zhejiang province, highway regulators and government officials have plans to build China’s first intelligent superhighway, outfitted with sensors, road-embedded solar panels and wireless communication between cars, roads and drivers.
Aimed at increasing transit efficiency by up to 30 percent while minimizing fatalities, the project may one day allow autonomous electric vehicles to continuously charge as they drive.
A similar government-fueled project involves Beijing’s new neighbor Xiong’an. Projected to take in over $580 billion in infrastructure spending over the next 20 years, Xiong’an New Area could one day become the world’s first city built around autonomous vehicles.
Baidu is already working with Xiong’an’s local government to build out this AI city with an environmental focus. Possibilities include sensor-geared cement, computer vision-enabled traffic lights, intersections with facial recognition, and parking lots-turned parks.
Lastly, Lee predicts China will almost certainly lead the charge in autonomous drones. Already, Shenzhen is home to premier drone maker DJI—a company I’ll be visiting with 24 top executives later this month as part of my annual China Platinum Trip.
Named “the best company I have ever encountered” by Chris Anderson, DJI owns an estimated 50 percent of the North American drone market, supercharged by Shenzhen’s extraordinary maker movement.
While the long-term Sino-US competitive balance in fourth wave AI remains to be seen, one thing is certain: in a matter of decades, we will witness the rise of AI-embedded cityscapes and autonomous machines that can interact with the real world and help solve today’s most pressing grand challenges.
Webinar with Dr. Kai-Fu Lee: Dr. Kai-Fu Lee — one of the world’s most respected experts on AI — and I will discuss his latest book AI Superpowers: China, Silicon Valley, and the New World Order. Artificial Intelligence is reshaping the world as we know it. With U.S.-Sino competition heating up, who will own the future of technology? Register here for the free webinar on September 4th, 2018 from 11:00am–12:30pm PST.
Image Credit: Elena11 / Shutterstock.com Continue reading →
Elon Musk Presents His Tunnel Vision to the People of LA
Jack Stewart and Aarian Marshall | Wired
“Now, Musk wants to build this new, 2.1-mile tunnel, near LA’s Sepulveda pass. It’s all part of his broader vision of a sprawling network that could take riders from Sherman Oaks in the north to Long Beach Airport in the south, Santa Monica in the west to Dodger Stadium in the east—without all that troublesome traffic.”
Feel What This Robot Feels Through Tactile Expressions
Evan Ackerman | IEEE Spectrum
“Guy Hoffman’s Human-Robot Collaboration & Companionship (HRC2) Lab at Cornell University is working on a new robot that’s designed to investigate this concept of textural communication, which really hasn’t been explored in robotics all that much. The robot uses a pneumatically powered elastomer skin that can be dynamically textured with either goosebumps or spikes, which should help it communicate more effectively, especially if what it’s trying to communicate is, ‘Don’t touch me!’”
In Virtual Reality, How Much Body Do You Need?
Steph Yin | The New York Times
“In a paper published Tuesday in Scientific Reports, they showed that animating virtual hands and feet alone is enough to make people feel their sense of body drift toward an invisible avatar. Their work fits into a corpus of research on illusory body ownership, which has challenged understandings of perception and contributed to therapies like treating pain for amputees who experience phantom limb.”
How Graphene and Gold Could Help Us Test Drugs and Monitor Cancer
Angela Chen | The Verge
“In today’s study, scientists learned to precisely control the amount of electricity graphene generates by changing how much light they shine on the material. When they grew heart cells on the graphene, they could manipulate the cells too, says study co-author Alex Savtchenko, a physicist at the University of California, San Diego. They could make it beat 1.5 times faster, three times faster, 10 times faster, or whatever they needed.”
Robotic Noses Could Be the Future of Disaster Rescue—If They Can Outsniff Search Dogs
Eleanor Cummins | Popular Science
“While canine units are a tried and fairly true method for identifying people trapped in the wreckage of a disaster, analytical chemists have for years been working in the lab to create a robotic alternative. A synthetic sniffer, they argue, could potentially prove to be just as or even more reliable than a dog, more resilient in the face of external pressures like heat and humidity, and infinitely more portable.”
Image Credit: Sergey Nivens / Shutterstock.com Continue reading →
Would you get on a plane that didn’t have a human pilot in the cockpit? Half of air travelers surveyed in 2017 said they would not, even if the ticket was cheaper. Modern pilots do such a good job that almost any air accident is big news, such as the Southwest engine disintegration on April 17.
But stories of pilot drunkenness, rants, fights and distraction, however rare, are reminders that pilots are only human. Not every plane can be flown by a disaster-averting pilot, like Southwest Capt. Tammie Jo Shults or Capt. Chesley “Sully” Sullenberger. But software could change that, equipping every plane with an extremely experienced guidance system that is always learning more.
In fact, on many flights, autopilot systems already control the plane for basically all of the flight. And software handles the most harrowing landings—when there is no visibility and the pilot can’t see anything to even know where he or she is. But human pilots are still on hand as backups.
A new generation of software pilots, developed for self-flying vehicles, or drones, will soon have logged more flying hours than all humans have—ever. By combining their enormous amounts of flight data and experience, drone-control software applications are poised to quickly become the world’s most experienced pilots.
Drones That Fly Themselves
Drones come in many forms, from tiny quad-rotor copter toys to missile-firing winged planes, or even 7-ton aircraft that can stay aloft for 34 hours at a stretch.
When drones were first introduced, they were flown remotely by human operators. However, this merely substitutes a pilot on the ground for one aloft. And it requires significant communications bandwidth between the drone and control center, to carry real-time video from the drone and to transmit the operator’s commands.
Many newer drones no longer need pilots; some drones for hobbyists and photographers can now fly themselves along human-defined routes, leaving the human free to sightsee—or control the camera to get the best view.
University researchers, businesses, and military agencies are now testing larger and more capable drones that will operate autonomously. Swarms of drones can fly without needing tens or hundreds of humans to control them. And they can perform coordinated maneuvers that human controllers could never handle.
Could humans control these 1,218 drones all together?
Whether flying in swarms or alone, the software that controls these drones is rapidly gaining flight experience.
Importance of Pilot Experience
Experience is the main qualification for pilots. Even a person who wants to fly a small plane for personal and noncommercial use needs 40 hours of flying instruction before getting a private pilot’s license. Commercial airline pilots must have at least 1,000 hours before even serving as a co-pilot.
On-the-ground training and in-flight experience prepare pilots for unusual and emergency scenarios, ideally to help save lives in situations like the “Miracle on the Hudson.” But many pilots are less experienced than “Sully” Sullenberger, who saved his planeload of people with quick and creative thinking. With software, though, every plane can have on board a pilot with as much experience—if not more. A popular software pilot system, in use in many aircraft at once, could gain more flight time each day than a single human might accumulate in a year.
As someone who studies technology policy as well as the use of artificial intelligence for drones, cars, robots, and other uses, I don’t lightly suggest handing over the controls for those additional tasks. But giving software pilots more control would maximize computers’ advantages over humans in training, testing, and reliability.
Training and Testing Software Pilots
Unlike people, computers will follow sets of instructions in software the same way every time. That lets developers create instructions, test reactions, and refine aircraft responses. Testing could make it far less likely, for example, that a computer would mistake the planet Venus for an oncoming jet and throw the plane into a steep dive to avoid it.
The most significant advantage is scale: Rather than teaching thousands of individual pilots new skills, updating thousands of aircraft would require only downloading updated software.
These systems would also need to be thoroughly tested—in both real-life situations and in simulations—to handle a wide range of aviation situations and to withstand cyberattacks. But once they’re working well, software pilots are not susceptible to distraction, disorientation, fatigue, or other human impairments that can create problems or cause errors even in common situations.
Rapid Response and Adaptation
Already, aircraft regulators are concerned that human pilots are forgetting how to fly on their own and may have trouble taking over from an autopilot in an emergency.
In the “Miracle on the Hudson” event, for example, a key factor in what happened was how long it took for the human pilots to figure out what had happened—that the plane had flown through a flock of birds, which had damaged both engines—and how to respond. Rather than the approximately one minute it took the humans, a computer could have assessed the situation in seconds, potentially saving enough time that the plane could have landed on a runway instead of a river.
Aircraft damage can pose another particularly difficult challenge for human pilots: It can change what effects the controls have on its flight. In cases where damage renders a plane uncontrollable, the result is often tragedy. A sufficiently advanced automated system could make minute changes to the aircraft’s steering and use its sensors to quickly evaluate the effects of those movements—essentially learning how to fly all over again with a damaged plane.
Boosting Public Confidence
The biggest barrier to fully automated flight is psychological, not technical. Many people may not want to trust their lives to computer systems. But they might come around when reassured that the software pilot has tens, hundreds, or thousands more hours of flight experience than any human pilot.
Other autonomous technologies, too, are progressing despite public concerns. Regulators and lawmakers are allowing self-driving cars on the roads in many states. But more than half of Americans don’t want to ride in one, largely because they don’t trust the technology. And only 17 percent of travelers around the world are willing to board a plane without a pilot. However, as more people experience self-driving cars on the road and have drones deliver them packages, it is likely that software pilots will gain in acceptance.
The airline industry will certainly be pushing people to trust the new systems: Automating pilots could save tens of billions of dollars a year. And the current pilot shortage means software pilots may be the key to having any airline service to smaller destinations.
Both Boeing and Airbus have made significant investments in automated flight technology, which would remove or reduce the need for human pilots. Boeing has actually bought a drone manufacturer and is looking to add software pilot capabilities to the next generation of its passenger aircraft. (Other tests have tried to retrofit existing aircraft with robotic pilots.)
One way to help regular passengers become comfortable with software pilots—while also helping to both train and test the systems—could be to introduce them as co-pilots working alongside human pilots. Planes would be operated by software from gate to gate, with the pilots instructed to touch the controls only if the system fails. Eventually pilots could be removed from the aircraft altogether, just like they eventually were from the driverless trains that we routinely ride in airports around the world.
This article was originally published on The Conversation. Read the original article.
Image Credit: Skycolors / Shutterstock.com Continue reading →
Robby Pepper can answer questions in Italian, English and German. Billed as Italy's first robot concierge, the humanoid will be deployed all season at a hotel on the popular Lake Garda to help relieve the desk staff of simple, repetitive questions. Continue reading →