Tech Talk iconTech Talk

null

The Fight Over the .africa Domain Name

An explosion of domain names has reshaped the Internet by offering hundreds of new ways to end a Web address. Lawyers can now advertise websites with “.lawyer” while toy companies can register with “.toys.” Jokers who want to build a site that finishes with “.fail” or “.wtf” can do that, too.  

But one highly sought domain remains stubbornly out of reach for roughly a billion people. Africans still can’t register sites to “.africa” because the right to operate that domain is the subject of a tussle between rival registries that is now dragging through its fourth year. While the domain’s ultimate fate could remain the subject of legal battles for years, a California court will decide on 4 April whether to finally permit .africa to go live.

The domain could prove quite lucrative for whichever registry wins it, though both competing registries pledge to spend profits on charitable activities. Registries act as domain name wholesalers. They sell the right to resell a domain name to many registrars such as GoDaddy, which make their money by signing people up.

Wayne Diamond, who runs a registrar based in South Africa, says many of his clients want to list websites with .africa, but are stuck waiting. “I think there's growing impatience with what's happening now that it's being held up in legal wranglings,” he says. “The delay has had a significant impact on the growth of the domain space in Africa.”

The two registries vying for control of .africa have also pitched the domain as an emerging economic engine and cultural exchange. DotConnectAfrica, a charitable trust that operates out of Kenya, promotes .africa as “your online African identity” while the South African nonprofit ZA Central Registry says the new domain will enable “e-commerce, technology and infrastructure to flourish.”

Those promises have so far gone unfulfilled. The tug-of-war began in 2011 when the Internet Corporation for Assigned Names and Numbers (ICANN), the nonprofit that manages domain names, noticed many of the shortest and most memorable addresses that ended in “.com,” “.org” and “.net” were taken.

The organization invited registries to apply to add more options, and has released 938 new domains in the time since. Most disputes were resolved amicably or by offering domains up for auction. For example, in January GMO Registry Inc. beat seven competitors with its $41.5 million bid for the “.shop” domain.

The .africa domain, however, didn’t go up for sale because of its geographic and cultural importance. In fact, ICANN requires applicants for a geographic domain to demonstrate support from 60 percent of national governments.

That requirement lies at the heart of the disagreement over which of the rival registries is better suited to manage .africa. Both candidates submitted their applications in 2012 and claimed that they had the support of the African Union Commission. DotConnectAfrica says it received the commission’s blessing in 2009, but the commission later formally withdrew that support and backed ZA Central Registry.

Sophia Bekele, head of DotConnectAfrica, says the process wasn’t “transparent and accountable” and that the commission failed to represent African governments. Neil Dundas, executive director of ZA Central Registry, points out that DotConnectAfrica has relatively few staff on the continent and would work with UK-based registry CentralNic to manage the domain. Filings with ICANN indicate DotConnectAfrica will charge only US $10 per year for website registrations, versus the $18 that ZA Central Registry plans to collect should it win the domain rights.

In 2014, ICANN agreed to issue .africa to ZA Central Registry. To fight back, DotConnectAfrica requested an internal review. After two more years, ICANN’s board passed a resolution in March reaffirming their decision and stating that DotConnectAfrica had not garnered enough government support.

That decision would have cleared the way for ZA Central Registry to begin registering Africa’s websites, but DotConnectAfrica filed a legal complaint against California-based ICANN, and asked a U.S. district court to block the organization from awarding .africa to ZA Central Registry while the case proceeds.

In a hearing scheduled for Monday, 4 April, the court will decide whether to grant DotConnectAfrica’s petition for a temporary stay that would prevent the transfer of the long-awaited .africa to ZA Central Registry while the suit against ICANN is adjudicated. If the court rules in its favor, ZA Central Registry estimates that it could have .africa sites up and running within four months. Dundas says he would love to see over a million sites signed up within three to five years.

But if not, the delay will drag on, and the promise of a new domain to jumpstart economic growth and build a shared online identity among Africans will remain nothing more than untapped potential.

null

My Subterranean Tour of London’s Crossrail

It’s a damp, freezing cold day in January, and I’m at the bottom of a massive hole in the ground. This is one of a pair of 41-meter-deep shafts in a part of east London called the Limmo peninsula, a spit of land on the banks of the River Thames. From a drone’s-eye view, it looks as though a giant hole punch has taken two neat circles out of the silty earth.

Back in late 2012, two enormous tunnel-boring machines were lowered into these shafts. Workers fired up the 1,000-metric-ton behemoths (named Elizabeth and Victoria), and their rotating cutting heads slowly gouged their way westwards. When they finally reached Central London in May 2015, it marked the completion of the tunneling work on Crossrail, a new underground railway system that spans London.

Crossrail will be fully operation by the end of 2019, with an expected 200 million passengers carried through its arteries every year. For now, it is Europe’s biggest construction project, with a budget of £14.8 billion (about US $21 billion).

This is my second visit to Crossrail. A few months earlier I strolled around the network’s immaculate new Canary Wharf station, where I learned about the innovations that are helping to bring this railway to life—including wireless sensors and lasers that monitor construction, smart components that warn of their impending failure, and a 3-D virtual model of the whole network that can be explored from an iPad. [For more on all that, see my article, “London’s Crossrail Is a $22 Billion Test of Virtual Modeling.”]

Here at the Limmo site, though, you can see the blunt end of this construction project. The area is packed with cranes, concrete mixers, and dumpsters. Warning alarms rip through the air as heavy loads are hoisted around. The workers wear scarves and balaclavas above their bright orange coats to ward off the cold.

My hosts and I descend the shaft using cramped stairs built from scaffolding and boards. Halfway down is a bizarre sight: a small shrine bolted to the wall, containing a figure of Saint Barbara, the patron saint of mineworkers and tunnelers. Dozens of these statues were blessed by priests and placed at the mouths of Crossrail’s tunnels before work began.

“It’s a tradition, hundreds of years old,” explains site manager Peter Kelly.

At the bottom, we walk along the eastbound tunnel as it curves upwards at a slight incline. Automated lighting brightens and dims as we pass, while bangs and crashes from the shaft echo around us. Rails laid out on the floor are ready to be welded together and mounted on ties by a mobile gantry, a vehicle that looks like a huge robotic spider on wheels. The gantry is currently less than 1 km to the east of here, and closing fast. Kelly says they’re now racing to finish all the civil engineering work at the site by the end of March. And in a few years’ time, trains will come barreling through here at about 65 kilometers per hour, tilting gently as they round the bend.

As we return to the surface, I notice that the short section of tunnel between the two shafts is strewn with rubble. Workmen are loading the hunks of concrete into wheelbarrows, their breath clouding the air from the exertion.

There’s no doubt that Crossrail is a high-tech railway—but building it still involves a whole lot of mud, concrete, shovels, and sweat. 

null

Machines Just Got Better at Lip Reading

Soccer aficionados will never forget the headbutt by French soccer great Zinadine Zidane during the 2006 World Cup final. Caught on video camera, Zidane’s attack on Italian player Marco Materazzi after a verbal exchange got him a red ticket. He left the field, making it easier for Italy to become world champions. The world found out later about Materazzi’s abusive words of Zidane’s female relatives.

“If we had good lip-reading technology Zidane’s reaction could have been explained or they would’ve both gotten sent out,” says Helen Bear, a computer scientist at the University of East Anglia in Norwich, UK. “Maybe the match outcome would be different.”

Bear and her colleague Richard Harvey have come up with a new lip-reading algorithm that improves a computer’s ability to differentiate between sounds—such as ‘p’, ‘b,’ and ‘m’—that all look similar on lips. The researchers presented their work at the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) in Shanghai.

Read More
null

Apple Reveals iPhone SE, and New iPad Pro With Chameleon-Like "True Tone" Display

At Apple’s latest press event in Cupertino, Calif., on Monday, the company revealed a new Retina display that automatically adjusts its color to match the light of its surroundings.

The first-of-its-kind technology, called the True Tone display, makes its debut on a smaller iPad Pro with a 9.7-inch screen. True Tone measures the brightness and temperature of the lighting in the immediate area (such as the oranges and yellows in a warmly lit room or the bluish hues of a cool summer night) through ambient light sensors. Then, it adjusts the hues on the screen to match.

The company says this chameleon-like quality should make reading or working on the iPad easier on users’ eyes. A separate function built into the iOS 9.3 update, called Night Shift, uses the internal clock and GPS system in a device to automatically move the colors on the screen away from the blue end of the spectrum as the sun sets. Artificial light from electronic devices is known to make it difficult to fall off to sleep. Blue wavelengths, say scientists, appear to affect sleep the most.

At the same event, Apple announced a new rose gold iPhone SE that it says melds the compact size of the iPhone 5 with the high performance of the iPhone 6. However, there was little about the underlying technology that breaks from the mold of its predecessors. Before the announcement, some analysts were already calling it “underwhelming.”

The new phone’s screen measures four inches diagonally, compared with the 4.7-inch screen for the iPhone 6 and 5.5-inch screen for the iPhone 6 Plus, which both debuted in 2014. That makes the iPhone SE very similar in appearance to the iPhone 5s.

In the U.S., there’s at least some evidence to suggest that a segment of customers may simply prefer a smaller device. Apple says that 60 percent of iPhone users who owned an iPhone before the launch of the iPhone 6 and 6s have neglected to upgrade.

Despite its iPhone 5–like size, the iPhone SE packs the same 64-bit A9 chip and M9 motion coprocessor as the iPhone 6s, enabling the “Hey, Siri” function to always remain on. It also comes equipped with Apple Pay and incorporates Touch ID, which allows a user to unlock their phone with their fingerprint.

As for price, the iPhone SE will cost $399 for 16GB, compared with $549 for the iPhone 6. That lower price, as well as the inclusion of Apple Pay, is likely part of Apple’s strategy to encourage customers in fast-growing cellphone markets such as China and India. CEO Tim Cook has said that China will eventually be home to the company’s largest customer base.

Overall, sales growth for the iPhone has already slowed worldwide. In January, the company said for the first time it expects iPhone sales to decline in the current quarter compared to the same period a year ago.

It’s a big week for Apple in several arenas. Tomorrow, the company faces the FBI in a U.S. federal court hearing over its refusal to build a new operating system that could break into the phone of a gunman from last year’s San Bernardino, Calif., mass shooting.

Fingers hold a small circuit board which supports four cylinders with lenses. A yellow/green piece of plastic partially covers the circuit board.

CeBIT 2016: Terabee’s Range Sensor Helps Make Drones Fast, Cheap, and Under Control

Editors Note: This week IEEE Spectrum is covering CeBIT, the enormous information and communications technology show that takes place annually in Hanover, Germany. For up-to-the-second updates, you can follow our CeBIT Ninja, Stephen Cass, on Twitter (@stephencass), or catch daily highlights throughout the week here.

The World Wide Web is the most famous technology to emerge from the needs of the international particle physics research center CERN, but it’s not the only one. In the latest example, a lightweight, inexpensive (and maker-friendly) range sensor has come about because scientists want to use drones to survey tunnels and vaults without smashing into expensive and difficult to replace equipment.

CERN’s massive subterranean facility lies underneath farm fields between Geneva and the Jura mountains. The centerpiece is the Large Hadron Collider, housed in a tunnel that forms a ring with a 27-kilometer circumference. As well as lots of interesting physics, these accelerators can also produce lethal amounts of radiation (hence the need to keep everything underground). A few years ago, CERN looked into the possibility of having drones create three-dimensional surveys of the radiation levels in the accelerator tunnels and the vaults that house CERN’s giant particle detectors. 

They approached a drone services company, but it was soon discovered that there was no way to create a drone that was small enough to operate in the cluttered spaces and yet had sensors that would let it locate itself with enough precision to avoid collisions. The result was the founding of Terabee in 2012, explains Massimiliano Ruffo, the company’s CEO, who I met at CeBIT’s airy press center yesterday. (I know you guys don’t care about #journalismproblems, but over the last 15 years of covering events I’ve had to work out of a lot of windowless pits, fighting with other reporters over desk space and wall sockets, so by all that’s holy, I’m going to give the Hannover Fair press center—which even has its own bar—a shout out.)

In 2015 Terabee was recognized formally as a CERN spin-off and selected to join the research center’s business incubation partner Innogex. Terabee began selling its first sensor, the TeraRanger One, the same year.

The matchbox-sized TeraRanger One sells for 125 euro (US $140) and weighs just 8 grams. Measuring the time-of-flight for infrared pulses generated by an LED allows the TeraRanger One to determine the distance to a single point-like region ahead 1000 times per second. The maximum distance that can be measured indoors (or, of course, underground) is about 14 meters, with a range accuracy of four centimeters. Ruffo says that with some sensor-by-sensor calibration and a slower rate of measurement, the accuracy can be increased to about two centimeters. Maximum distances closer to five to six meters are possible in sunlight.

The TeraRanger One’s onboard electronics takes care of all the post processing required from the sensor’s raw time-of-flight data and spits out the distance as a number that represents the number of millimeters measured. A 5-volt UART serial interface is used by default, and a 5 V I2C bus can be used with a firmware change, making the sensor trivial to hook up to an Arduino, and only slightly more complicated to connect to a Raspberry Pi. (That’s due to the latter’s aversion to voltages higher than 3.3 volts).

Inexpensive ultrasonic range finding sensors of TeraRanger One’s size and weight have been available for some time, but they lack its speed and angular precision. LIDAR systems, which often employ a rotating mirror to scan a sensing laser beam around, are also fast and provide high resolution at good distances but are bulkier and more expensive than the TeraRanger One. (Although that may change if DARPA’s phased-array LIDAR-on-a-chip ever comes to the mass market.) Stereo vision systems are another alternative for rangefinding, and can provide depth information over a wide field of pixels, but Ruffo believes that TeraBee again has the edge because the time-of-flight data produces more reliable distance measurements over a longer range.

For systems that require more complex measurements, such as those the doing the kind of SLAM (simultaneous location and mapping) required for the original CERN surveying, TeraBee currently offers a hub which allows measurements from up to eight separate sensors to be integrated. A pre-built eight sensor “tower” is in the works, says Ruffo, and the company also offers an evaluation version of a small LIDAR-type scanning platform

I’m hoping to try the TeraRanger One out for myself for Spectrum’s Hands On section in the coming months and put it through its paces. Perhaps, in honor of its design history, I can make a wearable people detector for the deep dark tunnels of the New York City subway—one that will solve the recurring problem of being oblivious to an empty subway seat opening up directly behind you and then losing that seat to another rider.

A glowing floating balloon patterned to look like an eyeball hovers in the evening sky outdoors.

CeBIT 2016: The Aerotain Skye Could Be Your Friendly Floating Camera Drone

Editors Note: This week IEEE Spectrum is covering CeBIT, the enormous information and communications technology show that takes place annually in Hanover, Germany. For up-to-the-second updates, you can follow our CeBIT Ninja, Stephen Cass, on Twitter (@stephencass), or catch daily highlights throughout the week here. 

Once upon a time there was a very odd British television show called The Prisoner, which featured a secret agent repeatedly attempting to escape from a mysterious village. One of the biggest threats the agent faced was a giant balloon called Rover, which would pursue and subdue rule-breaking villagers. Now Rover has been brought to reality, albeit in a much more adorable version, thanks to the engineers at Aerotain and their Skye inflatable drone.

The Skye is a 3-meter-diameter controllable balloon that’s filled with helium for buoyancy. Dotted around the surface are propellers whose direction can be adjusted, spinning the balloon or moving it around as required. There’s also the option to add an internal projector to display moving images on the balloon’s skin. Skye has been used at events as a crowd-pleaser, but it can also be used a platform for aerial photography by adding cameras.

Read More
null

Meet the Guy Whose Software Keeps the World’s Clocks in Sync

Clocks sprang forward last weekend in about 75 countries. Over time, technology has spared many citizens who observe daylight saving time the headache of physically changing their clocks. Electronic clocks automatically reset the time—a subtle convenience made possible by the rise of the global Internet, a network of real-life atomic clocks, and a physicist who has spent decades finding smarter ways to distribute time.  

In many cases, the internal clock that ticks away in a laptop or desktop computer is synchronized to an official time service maintained by the U.S. National Institute of Standards and Technology (NIST). This free service shares Coordinated Universal Time with personal devices, Web browsers, financial trading software and e-mail programs throughout the world. The service receives 150,000 requests per second (roughly 16 billion a day) from systems that repeatedly ask, “What time is it?

“If you have a PC, it's probably synchronized to the time service,” says Judah Levine, the man who originally built servers and programmed software to send time over the Internet for NIST back in 1993.

Here’s how it works: electronic clocks are programmed to check in (once an hour, on average) and record the time from a network of 20 “timeservers” scattered throughout the U.S. Three of those servers—two in Boulder, Colo., and a third in Fort Collins, Colo.—are physically linked to atomic clocks, the newest of which is so accurate, it gains or loses only a second every 300 million years. At those sites, an electric pulse signifying the start of each second is generated by the atomic clock and delivered to a server.

Once the pulse has arrived at a server, the signal is translated to the precise hour, minute and second of the day in Coordinated Universal Time using a string of characters sent separately from the clock to the server. This code enables the server to identify a given second as 16:02:56 UTC, for example. 

Next, this information is shared with the other NIST servers through a phone line and distributed to devices and systems over the Internet, primarily in 48-byte packets via the Network Time Protocol. Adjustments for time zones, daylight saving time, or leap seconds are made by Internet service providers or handled by instructions built into a network or device.

While many nations operate an official time service, NIST has the highest-capacity timekeeping network. It’s also the most popular. Jeff Sherman, a NIST physicist, recently tracked requests to two NIST timeservers for one month. He found requests originating from 316 million unique IP addresses, which he estimates represents about 8.5 percent of all the devices currently connected to the Internet. Those addresses were registered in 244 nations.

This system has served the Internet well for more than 20 years. But Levine expects demand for the time service to grow exponentially with the rise of internet-connected wearables and household electronics. These days, the 76-year-old physicist is thinking hard about how to prepare the network for the onslaught of requests it expects to receive in the era of the Internet of Things.

“The growth rate has sort of been steady at a few percent a month,” Levine says. “We should be able to handle the next two or three years of growth. After that, we're going to have to think again.” He isn’t even sure how many requests per day the current system could handle.

One way to absorb such growth might be to add more layers to the time service. For personal devices that do not require the precise time down to the thousandth of a second, companies such as Apple, Google or Microsoft could make a single request to NIST and then provide the time indirectly to thousands of customers through layers, or stratums, of service.

“One would hope that if a toaster is connected to the Internet,” Sherman says, “it doesn't need the same degree of accuracy as a power plant and someone would arrange for that toaster to be in that stratum system.” In fact, many companies already do this—but more may need to adopt the practice as the Internet of Things takes off.

Another option would be to rely more heavily on alternatives to the NIST time service that already provide the time to many devices and networks. For example, most cell phones rely on GPS satellites maintained by the U.S. Naval Observatory to track time. Many Web browsers synchronize to clocks maintained by other sources through the Network Time Protocol, which pre-dates Levine’s service.

For now, NIST is considering restructuring its timeservers so that every server is directly linked to an atomic clock. Levine says the plan is to build a fourth atomic clock at NIST’s headquarters in Gaithersburg, Md., and maintain about 12 timeservers total between the four sites. This would simultaneously improve the system’s accuracy and boost its capacity for delivering the answer to the question, “What time is it?”

Boys and girls smile as they hold tablets with bright yellow cases

CeBIT2016: The Kio Kit Is a Classroom in a Box

Editors Note: This week IEEE Spectrum is covering CeBIT, the enormous information and communications technology show that takes place annually in Hanover, Germany. For up-to-the-second updates, you can follow our CeBIT Ninja, Stephen Cass, on Twitter (@stephencass), or catch daily highlights throughout the week here.

Founded in 2013, BRCK is a Kenyan company that makes a rugged router designed for places with poor communications or power infrastructure. It can act as a traditional WiFi access point, but if a wired connection goes out or is simply not available, the router can switch over to cellular networks. Power outages are compensated for by an 8-hour battery. But now the company has gone beyond its basic product with the launch in September 2015 of BRCK Education and its US $5,000 Kio Kit.

The Kio Kit is an almost literal turnkey connected classroom: A water-resistant trunk-sized travel case contains 40 Kio 7-inch tablets and a BRCK router. The travel case wirelessly each charges each tablet, and the whole unit can be charged from either a wall outlet, solar power, or even a car battery. The tablets come pre-loaded with educational software chosen by the purchaser, which can be a mix of free and paid material from providers such as the Kahn Academy or eKitabu, a Kenyan e-book company. Updates can provided through the cloud when connectivity is available.

BRCK’s business development manager, Alex Masika, was at CeBIT to present early results from Kio Kit deployments at the invitation of the German Federal Ministry of Economic Cooperation and Development. Since January, Kio Kits have been sold into schools in Kenya, Tanzania, and the Solomon Islands, with additional orders coming in from Sudan, and queries from many other countries around the globe including the United States.

“The impetus for BRCK Education was the lack of education around the world, with hundreds of millions of kids going without,” says Masika. Educational content was available, but existing set of tools, such as typical consumer-grade tablets, “wasn’t able to address the challenges faced in Africa with power and connectivity,” he adds. Even something as basic as charging multiple mobile devices proved difficult in many schools, so BRCK tried to develop an all-in-one-solution with an emphasis on durability. The tablets are designed to survive a drop of least 70 centimeters, and “we haven’t had report of a single broken screen yet,” says Masika. Other touches—such as color coding the headphones yellow to make them easy to identify when giving instructions—were designed to make the system as hassle-free for teachers as possible.

Masika, who is currently looking for investors and industry partners who can help scale up production and distribution of the Kio Kit, notes that one thing he’d like to see is Kio Kits popping up in places like refugee camps along with other emergency infrastructure like tents. In the meantime, the Nairobi-based Kio Kit and BRCK engineers and designers are continuing to improve the system based on user feedback. 

An orange-colored air-plane-shaped drone with two propellers and two large flaps. It is standing vertically, supported by fins from its wings and tail.

CeBIT 2016: Wingtra Wants To Be Your Hybrid Drone

Editors Note: This week IEEE Spectrum is covering CeBIT, the monster information and communications technology show that takes place annually in Hanover, Germany. For up-to-the-second updates, you can follow our CeBIT Ninja, Stephen Cass, on Twitter (@stephencass), or catch daily highlights throughout the week here.

Quadcopters and other helicopter-style drones can take off and land vertically with pinpoint precision, but they aren’t as fuel efficient or as fast as fixed-wing drones. On the other hand, fixed-wing aircraft normally require either catapults or relatively long runways to get up to speed before taking off. A spin-off company from the Autonomous Systems Lab at ETH Zurich is trying to provide the best of both worlds with its eponymous Wingtra drone.

The Wingtra takes off vertically (it’s held upright on the ground by fins projecting from the wings and tail), then levels out into horizontal flight. For landing, the general process is reversed, but with the assistance of a camera located in the tail. This camera allows the drone to spot a printed target placed on the ground. Once in sight, the Wingtra will autonomously descend to touch down on the target, within about 10 centimeters of bullseye, says Wingtra’s Leoplold Flechsenberger. 

The battery-powered Wingtra can fly for about an hour, during which time it can travel 60 kilometers. There’s no need for continuous control by the operator, as the Wingtra will follow its flight path autonomously. A removable module can carry different payloads: Those looking to inspect railway lines or survey crops for precision agriculture might choose to equip the drone with a high-resolution LIDAR or camera package, for example. Alternatively, an add-on freight module lets the Wingtra carry up to 0.5 kilograms, which Flechsenberger says might prove invaluable in dispatching medical supplies to rural areas.

The drone was designed with simplicity in mind. There are just five primary components: a set of wings combined with the fuselage to form a single body, plus two propellers and two flaps. The drone doesn’t even have a forward-looking camera (although Flechsenberger says one may in added in later versions). The price has yet to be announced, but as the Wingtra is aimed at professional and institutional users, it’s likely to be considerably more than what one would expect for anything aimed at consumers or prosumers. The system is expected to be commercially available in 2017, says Flechsenberger (who adds that Wingtra is hiring to accommodate its rapid expansion). 

null

AlphaGo Wins Final Game In Match Against Champion Go Player

AlphaGo, a largely self-taught Go-playing AI, last night won the fifth and final game in a match held in Seoul, South Korea, against that country’s Lee Sedol. Sedol is one of the greatest modern players of the ancient Chinese game. The final score was 4 games to 1.

Thus falls the last and computationally hardest game that programmers have taken as a test of machine intelligence. Chess, AI’s original touchstone, fell to the machines 19 years ago, but Go had been expected to last for many years to come.

The sweeping victory means far more than the US $1 million prize, which Google’s London-based acquisition, DeepMind, says it will give to charity. That’s because AlphaGo, for all its processing power, mainly owes its victory to a radical new way of using that power: via deep neural networks. These networks can train themselves with only a little intervention from human beings, and DeepMind’s researchers had already demonstrated that they can master a wide range of computer video games. The researchers hope that this generalizability can be carried over to the mastering of practical tasks in many other domains, including medicine and robotics.

Game programming began with chess, using methods first sketched out by Claude Shannon and Alan Turing in the 1940s. A machine calculates every possible continuation for each side, working its way as many moves ahead as it can and so generating a tree of analysis with millions of game positions. It then grades the positions by applying rules of thumb that even beginning chess players know, such as the differing values of the various pieces and the importance of controlling the center of the board. Finally, the algorithm traces its way from those end positions back to the current position to find the move that leads to the best outcome, assuming perfect play on both sides.

With modern hardware, this “brute-force” method can produce a strong chess-playing program. Add a grab-bag of tricks to “prune” the analysis tree, throwing out bad lines so the program can explore promising lines more deeply, and you get world-champion-level play. That came in 1997, when IBM’s Deep Blue supercomputer defeated then-World Chess Champion Garry Kasparov. Today you can download a US $100 program that plays even better—on a laptop.

Though some researchers have argued for some time that brute-force searching can in principle conquer Go, the game has long resisted such efforts. Compared to chess, the Chinese game offers far more moves in a given position and far more moves in a typical game, creating an intractably huge tree of analysis. It also lacks reliable rules of thumb for the grading of positions. 

In recent years, many programmers have tried to get around this problem with Monte Carlo simulation, a statistical means of finding the best first move from a vast database of the games that might begin from a given position. That method is also used a bit in AlphaGo, together with the tree-generating methods of yore. But the key improvement is AlphaGo’s use of deep neural networks to recognize patterns.

At a quiet moment, 42 minutes into the streaming of the match’s second game, on 10 March, one of the online commenters, Google’s Thore Graepel, described his first over-the-board encounter with an early form of AlphaGo a year ago—on his first day of work at Deep Mind’s London office. “I thought, neural network, how difficult can it be? It cannot even do reading of positions, it just does pattern recognitions,” Graepel said. “I sat down in front of the board, a small crowd gathered round, and in a small time, my position started to deteriorate… I ended up losing, I tried again and lost again. At least at that point the office knew me, I had a good introduction!”

AlphaGo uses two neural networks, a policy network that was trained on millions of master games with the goal of imitating their play, and a value network, that tries to assign a winning probability to each given position. That way, the machine can focus its efforts on the most promising continuations.  Then comes the tree-searching part, which tries to look many moves ahead.

“One way to think of it is that the policy network provides a guide, suggesting to AlphaGo moves to consider; but AlphaGo can then go on beyond that and come up with a new conclusion that overwhelms the suggestion by the policy network,” explained David Silver, the leader of the AlphaGo team, in online commentary last night, just before the final game. “At every part of the search tree, it’s using the policy network to suggest moves and the value network to evaluate moves. The policy network alone was enough to beat Graepel, an accomplished amateur player, on his first day in the office.”  

A strange consequence of AlphaGo’s division of labor is the way it plays once it thinks it has a clearly winning game. A human player would normally try to win by the largest possible margin, by capturing not just one extra point on the board, but 10 or 20 points, if possible. That way, the human would be likely to win even if he later makes a small mistake. But AlphaGo prefers to win by one point, at what it considers a high probability, over winning, say, by 20 points, at a rather lower probability.

You might think that this tendency to go for the safe-but-slack move is what enabled Lee Sedol to win the fourth game, on Sunday. And indeed, commentators at the time noted that the machine seemed to have the upper hand when Sedol pounced with an unexpected move, after which the machine played some weak moves. Sedol had used up a lot of time on his clock and so had to scramble to make his following moves, but in the end he was able to sustain his advantage and finally win.

However it wasn’t slackness but sheer surprise that caused the problem, members of the DeepMind team said last night, in commentary before thev final game. “That crucial move that Lee found, move 78, was considered very unlikely to be played by a human—[the program] estimated a one in 10,000 chance,” said David Silver, the team leader. “It’s fair to say that AlphaGo was surprised by this move; it had to start replanning at that point.”

A human player faced with a strange-looking move would study it deeply, if there was enough time to do so—and AlphaGo had plenty of time. “But AlphaGo has a simple time-control strategy,” Silver noted. “Maybe that’s something we can work on in future.”

So, it seems, efforts to improve AlphaGo will continue.

“An exciting direction for future research is to consider whether a machine can learn completely by itself, without any human examples, to achieve this level of performance,” Silver said. 

The final goal, of course, is to create an all-around learning machine, one that can learn to do a lot of things that people now get paid to do. Like, say, reporting and writing blog posts like this one. 

Advertisement

Tech Talk

IEEE Spectrum’s general technology blog, featuring news, analysis, and opinions about engineering, consumer electronics, and technology and society, from the editorial staff and freelance contributors.

Newsletter Sign Up

Sign up for the Tech Alert newsletter and receive ground-breaking technology and science news from IEEE Spectrum every Thursday.

Advertisement
Load More