Tag Archives: learn
#431371 Amazon Is Quietly Building the Robots of ...
Science fiction is the siren song of hard science. How many innocent young students have been lured into complex, abstract science, technology, engineering, or mathematics because of a reckless and irresponsible exposure to Arthur C. Clarke at a tender age? Yet Arthur C. Clarke has a very famous quote: “Any sufficiently advanced technology is indistinguishable from magic.”
It’s the prospect of making that… ahem… magic leap that entices so many people into STEM in the first place. A magic leap that would change the world. How about, for example, having humanoid robots? They could match us in dexterity and speed, perceive the world around them as we do, and be programmed to do, well, more or less anything we can do.
Such a technology would change the world forever.
But how will it arrive? While true sci-fi robots won’t get here right away—the pieces are coming together, and the company best developing them at the moment is Amazon. Where others have struggled to succeed, Amazon has been quietly progressing. Notably, Amazon has more than just a dream, it has the most practical of reasons driving it into robotics.
This practicality matters. Technological development rarely proceeds by magic; it’s a process filled with twists, turns, dead-ends, and financial constraints. New technologies often have to answer questions like “What is this good for, are you being realistic?” A good strategy, then, can be to build something more limited than your initial ambition, but useful for a niche market. That way, you can produce a prototype, have a reasonable business plan, and turn a profit within a decade. You might call these “stepping stone” applications that allow for new technologies to be developed in an economically viable way.
You need something you can sell to someone, soon: that’s how you get investment in your idea. It’s this model that iRobot, developers of the Roomba, used: migrating from military prototypes to robotic vacuum cleaners to become the “boring, successful robot company.” Compare this to Willow Garage, a genius factory if ever there was one: they clearly had ambitions towards a general-purpose, multi-functional robot. They built an impressive device—PR2—and programmed the operating system, ROS, that is still the industry and academic standard to this day.
But since they were unable to sell their robot for much less than $250,000, it was never likely to be a profitable business. This is why Willow Garage is no more, and many workers at the company went into telepresence robotics. Telepresence is essentially videoconferencing with a fancy robot attached to move the camera around. It uses some of the same software (for example, navigation and mapping) without requiring you to solve difficult problems of full autonomy for the robot, or manipulating its environment. It’s certainly one of the stepping-stone areas that various companies are investigating.
Another approach is to go to the people with very high research budgets: the military.
This was the Boston Dynamics approach, and their incredible achievements in bipedal locomotion saw them getting snapped up by Google. There was a great deal of excitement and speculation about Google’s “nightmare factory” whenever a new slick video of a futuristic militarized robot surfaced. But Google broadly backed away from Replicant, their robotics program, and Boston Dynamics was sold. This was partly due to PR concerns over the Terminator-esque designs, but partly because they didn’t see the robotics division turning a profit. They hadn’t found their stepping stones.
This is where Amazon comes in. Why Amazon? First off, they just announced that their profits are up by 30 percent, and yet the company is well-known for their constantly-moving Day One philosophy where a great deal of the profits are reinvested back into the business. But lots of companies have ambition.
One thing Amazon has that few other corporations have, as well as big financial resources, is viable stepping stones for developing the technologies needed for this sort of robotics to become a reality. They already employ 100,000 robots: these are of the “pragmatic, boring, useful” kind that we’ve profiled, which move around the shelves in warehouses. These robots are allowing Amazon to develop localization and mapping software for robots that can autonomously navigate in the simple warehouse environment.
But their ambitions don’t end there. The Amazon Robotics Challenge is a multi-million dollar competition, open to university teams, to produce a robot that can pick and package items in warehouses. The problem of grasping and manipulating a range of objects is not a solved one in robotics, so this work is still done by humans—yet it’s absolutely fundamental for any sci-fi dream robot.
Google, for example, attempted to solve this problem by hooking up 14 robot hands to machine learning algorithms and having them grasp thousands of objects. Although results were promising, the 10 to 20 percent failure rate for grasps is too high for warehouse use. This is a perfect stepping stone for Amazon; should they crack the problem, they will likely save millions in logistics.
Another area where humanoid robotics—especially bipedal locomotion, or walking, has been seriously suggested—is in the last mile delivery problem. Amazon has shown willingness to be creative in this department with their notorious drone delivery service. In other words, it’s all very well to have your self-driving car or van deliver packages to people’s doors, but who puts the package on the doorstep? It’s difficult for wheeled robots to navigate the full range of built environments that exist. That’s why bipedal robots like CASSIE, developed by Oregon State, may one day be used to deliver parcels.
Again: no one more than Amazon stands to profit from cracking this technology. The line from robotics research to profit is very clear.
So, perhaps one day Amazon will have robots that can move around and manipulate their environments. But they’re also working on intelligence that will guide those robots and make them truly useful for a variety of tasks. Amazon has an AI, or at least the framework for an AI: it’s called Alexa, and it’s in tens of millions of homes. The Alexa Prize, another multi-million-dollar competition, is attempting to make Alexa more social.
To develop a conversational AI, at least using the current methods of machine learning, you need data on tens of millions of conversations. You need to understand how people will try to interact with the AI. Amazon has access to this in Alexa, and they’re using it. As owners of the leading voice-activated personal assistant, they have an ecosystem of developers creating apps for Alexa. It will be integrated with the smart home and the Internet of Things. It is a very marketable product, a stepping stone for robot intelligence.
What’s more, the company can benefit from its huge sales infrastructure. For Amazon, having an AI in your home is ideal, because it can persuade you to buy more products through its website. Unlike companies like Google, Amazon has an easy way to make a direct profit from IoT devices, which could fuel funding.
For a humanoid robot to be truly useful, though, it will need vision and intelligence. It will have to understand and interpret its environment, and react accordingly. The way humans learn about our environment is by getting out and seeing it. This is something that, for example, an Alexa coupled to smart glasses would be very capable of doing. There are rumors that Alexa’s AI will soon be used in security cameras, which is an ideal stepping stone task to train an AI to process images from its environment, truly perceiving the world and any threats it might contain.
It’s a slight exaggeration to say that Amazon is in the process of building a secret robot army. The gulf between our sci-fi vision of robots that can intelligently serve us, rather than mindlessly assemble cars, is still vast. But in quietly assembling many of the technologies needed for intelligent, multi-purpose robotics—and with the unique stepping stones they have along the way—Amazon might just be poised to leap that gulf. As if by magic.
Image Credit: Denis Starostin / Shutterstock.com Continue reading
Posted in Human Robots
#431350 The Internet of Things Needs to Be ...
In an interview at Singularity University’s Global Summit in San Francisco, Andreas Gal explained how his company is applying artificial intelligence to the Internet of Things (IoT). Gal is the former CTO of Mozilla and is currently CEO of Silk Labs.
“For us, the value of IoT is not really in making things connected,” Gal said. “It’s really about bringing intelligence to these devices, and that’s what we are focused on. We are bringing the latest advances in AI technology directly into these devices.”
Watch the interview to learn how infusing machine learning into IoT devices can take them beyond simple connection to add much greater value.
Image Credit: Anita Ponne / Shutterstock.com Continue reading
Posted in Human Robots
#431238 AI Is Easy to Fool—Why That Needs to ...
Con artistry is one of the world’s oldest and most innovative professions, and it may soon have a new target. Research suggests artificial intelligence may be uniquely susceptible to tricksters, and as its influence in the modern world grows, attacks against it are likely to become more common.
The root of the problem lies in the fact that artificial intelligence algorithms learn about the world in very different ways than people do, and so slight tweaks to the data fed into these algorithms can throw them off completely while remaining imperceptible to humans.
Much of the research into this area has been conducted on image recognition systems, in particular those relying on deep learning neural networks. These systems are trained by showing them thousands of examples of images of a particular object until they can extract common features that allow them to accurately spot the object in new images.
But the features they extract are not necessarily the same high-level features a human would be looking for, like the word STOP on a sign or a tail on a dog. These systems analyze images at the individual pixel level to detect patterns shared between examples. These patterns can be obscure combinations of pixel values, in small pockets or spread across the image, that would be impossible to discern for a human, but highly accurate at predicting a particular object.
“An attacker can trick the object recognition algorithm into seeing something that isn’t there, without these alterations being obvious to a human.”
What this means is that by identifying these patterns and overlaying them over a different image, an attacker can trick the object recognition algorithm into seeing something that isn’t there, without these alterations being obvious to a human. This kind of manipulation is known as an “adversarial attack.”
Early attempts to trick image recognition systems this way required access to the algorithm’s inner workings to decipher these patterns. But in 2016 researchers demonstrated a “black box” attack that enabled them to trick such a system without knowing its inner workings.
By feeding the system doctored images and seeing how it classified them, they were able to work out what it was focusing on and therefore generate images they knew would fool it. Importantly, the doctored images were not obviously different to human eyes.
These approaches were tested by feeding doctored image data directly into the algorithm, but more recently, similar approaches have been applied in the real world. Last year it was shown that printouts of doctored images that were then photographed on a smartphone successfully tricked an image classification system.
Another group showed that wearing specially designed, psychedelically-colored spectacles could trick a facial recognition system into thinking people were celebrities. In August scientists showed that adding stickers to stop signs in particular configurations could cause a neural net designed to spot them to misclassify the signs.
These last two examples highlight some of the potential nefarious applications for this technology. Getting a self-driving car to miss a stop sign could cause an accident, either for insurance fraud or to do someone harm. If facial recognition becomes increasingly popular for biometric security applications, being able to pose as someone else could be very useful to a con artist.
Unsurprisingly, there are already efforts to counteract the threat of adversarial attacks. In particular, it has been shown that deep neural networks can be trained to detect adversarial images. One study from the Bosch Center for AI demonstrated such a detector, an adversarial attack that fools the detector, and a training regime for the detector that nullifies the attack, hinting at the kind of arms race we are likely to see in the future.
While image recognition systems provide an easy-to-visualize demonstration, they’re not the only machine learning systems at risk. The techniques used to perturb pixel data can be applied to other kinds of data too.
“Bypassing cybersecurity defenses is one of the more worrying and probable near-term applications for this approach.”
Chinese researchers showed that adding specific words to a sentence or misspelling a word can completely throw off machine learning systems designed to analyze what a passage of text is about. Another group demonstrated that garbled sounds played over speakers could make a smartphone running the Google Now voice command system visit a particular web address, which could be used to download malware.
This last example points toward one of the more worrying and probable near-term applications for this approach: bypassing cybersecurity defenses. The industry is increasingly using machine learning and data analytics to identify malware and detect intrusions, but these systems are also highly susceptible to trickery.
At this summer’s DEF CON hacking convention, a security firm demonstrated they could bypass anti-malware AI using a similar approach to the earlier black box attack on the image classifier, but super-powered with an AI of their own.
Their system fed malicious code to the antivirus software and then noted the score it was given. It then used genetic algorithms to iteratively tweak the code until it was able to bypass the defenses while maintaining its function.
All the approaches noted so far are focused on tricking pre-trained machine learning systems, but another approach of major concern to the cybersecurity industry is that of “data poisoning.” This is the idea that introducing false data into a machine learning system’s training set will cause it to start misclassifying things.
This could be particularly challenging for things like anti-malware systems that are constantly being updated to take into account new viruses. A related approach bombards systems with data designed to generate false positives so the defenders recalibrate their systems in a way that then allows the attackers to sneak in.
How likely it is that these approaches will be used in the wild will depend on the potential reward and the sophistication of the attackers. Most of the techniques described above require high levels of domain expertise, but it’s becoming ever easier to access training materials and tools for machine learning.
Simpler versions of machine learning have been at the heart of email spam filters for years, and spammers have developed a host of innovative workarounds to circumvent them. As machine learning and AI increasingly embed themselves in our lives, the rewards for learning how to trick them will likely outweigh the costs.
Image Credit: Nejron Photo / Shutterstock.com Continue reading
Posted in Human Robots