36

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

P R

O D

U C

T P

H O

T O

S B

Y L

E O

N A

R D

G R

E C

O

MA17_10_360cameras.indd 36 2/8/17 12:39 PM

T h

e 3

6 0

-D eg

ree Se

lfie

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW

VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

By Elizabeth Woyke

Inexpensive cameras that make spherical images are opening a new era in photography and changing the way people share stories.

Breakthrough Consumer cameras that produce 360° images, providing a realistic sense of events or places.

Why It Matters Photos and videos with this perspective could become the new standard for everything from news coverage to vacation shots.

Key Players - Ricoh - Samsung - 360fly - JK Imaging (maker of

Kodak Pixpro digital cameras)

- IC Real Tech (maker of the ALLie camera)

- Humaneyes Technologies

Availability Now

MA17_10_360cameras.indd 37 2/8/17 12:39 PM

38

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

Seasonal changes to vegetation fas- cinate Koen Hufkens. So last fall Hufkens, an ecological researcher at Harvard, devised a system to continuously broadcast images

from a Massachusetts forest to a website called VirtualForest.io. And because he used a camera that creates 360° pictures, visitors can do more than just watch the feed; they can use their mouse cursor (on a computer) or finger (on a smartphone or tablet) to pan around the image in a circle or scroll up to view the forest can- opy and down to see the ground. If they

look at the image through a virtual-reality headset they can rotate the photo by mov- ing their head, intensifying the illusion that they are in the woods.

Hufkens says the project will allow him to document how climate change is affecting leaf development in New Eng- land. The total cost? About $550, includ- ing $350 for the Ricoh Theta S camera that takes the photos.

We experience the world in 360 degrees, surrounded by sights and sounds. Until recently, there were two main options for shooting photos and video that

captured that context: use a rig to posi- tion multiple cameras at different angles with overlapping fields of view or pay at least $10,000 for a special camera. The production process was just as cumber- some and generally took multiple days to complete. Once you shot your footage, you had to transfer the images to a computer; wrestle with complex, pricey software to fuse them into a seamless picture; and then convert the file into a format that other people could view easily.

Today, anyone can buy a decent 360° camera for less than $500, record a video

C O

U R

T E

S Y

O F

A L

L IE

C A

M E

R A

MA17_10_360cameras.indd 38 2/8/17 12:39 PM

39

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW

VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

Chicago’s Millennium Park captured by the ALLie camera.

ALLie Camera It uses technology originally developed for the surveillance industry and can capture images in low light.

MA17_10_360cameras.indd 39 2/8/17 12:39 PM

40

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

within minutes, and upload it to Facebook or YouTube. Much of this amateur 360° content is blurry; some of it captures 360 degrees horizontally but not vertically; and most of it is mundane. (Watching footage of a stranger’s vacation is almost as boring in spherical view as it is in reg- ular mode.) But the best user-generated 360° photos and videos—such as the Vir- tual Forest—deepen the viewer’s apprecia- tion of a place or an event.

Journalists from the New York Times and Reuters are using $350 Samsung Gear 360 cameras to produce spherical photos and videos that document any- thing from hurricane damage in Haiti to a refugee camp in Gaza. One New York Times video that depicts people in Niger

fleeing the militant group Boko Haram puts you in the center of a crowd receiv- ing food from aid groups. You start by watching a man heaving sacks off a pickup truck and hearing them thud onto the ground. When you turn your head, you see the throngs that have gathered to claim the food and the makeshift carts they will use to transport it. The 360° for- mat is so compelling that it could become a new standard for raw footage of news events—something that Twitter is trying to encourage by enabling live spherical videos in its Periscope app.

Or consider the spherical videos of medical procedures that the Los Angeles startup Giblib makes to teach students about surgery. The company films the

Ballet dancers captured by the Samsung Gear 360.

C O

U R

T E

S Y

O F

S A

M S

U N

G

The 360° format is so compelling that it could become a new standard f

o r

r a

w f

o o

t a

g e

o

f n

e w

s e

v e

n t

s .

MA17_10_360cameras.indd 40 2/8/17 12:39 PM

41

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW

VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

samsung gear 360 Samsung has given these cameras to New York Times and Reuters journalists who are producing 360° news coverage.

MA17_10_360cameras.indd 41 2/8/17 12:39 PM

42

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

Utah’s Sidestep Canyon captured by the Ricoh Theta S.

Ricoh Theta S Ricoh put the image sensors on the camera’s sides instead of behind its lenses, making its thin shape possible.

J O

H N

F O

W L

E R

/F L

IC K

R

MA17_10_360cameras.indd 42 2/8/17 12:39 PM

43

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW

VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

operations by attaching a $500 360fly 4K camera, which is the size of a baseball, to surgical lights above the patient. The 360° view enables students to see not just the surgeon and surgical site, but also the way the operating room is organized and how the operating room staff interacts.

Meanwhile, inexpensive 360° cameras such as Kodak’s $450 Pixpro SP360 4K are popping up on basketball backboards, football fields, and hockey nets during practice for professional and collegiate teams. Coaches say the resulting videos help players visualize the action and pre-

pare for games in ways that conventional sideline and end-zone videos can’t.

Component innovations These applications are feasible because of the smartphone boom and innova- tions in several technologies that combine images from multiple lenses and sensors. For instance, 360° cameras require more horsepower than regular cameras and generate more heat, but that is handled by the energy-efficient chips that power smartphones. Both the 360fly and the $499 ALLie camera use Qualcomm Snap-

dragon processors similar to those that run Samsung’s high-end handsets.

Camera companies also benefited in recent years from smartphone vendors’ continuous quest to integrate higher- quality imaging into their gadgets. The competition forced component makers like Sony to shrink image sensors and ensure that they offered both high reso- lution and good performance in low light. As the huge smartphone market helped bring down component prices, 360°-cam- era makers found it possible to price their devices accessibly, often at less than

MA17_10_360cameras.indd 43 2/8/17 12:39 PM

44

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES 360Fly 4K Dustproof and water-resistant, it’s often used to record extreme sports.

MA17_10_360cameras.indd 44 2/8/17 12:39 PM

45

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW

VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

$500. “There are sensors that now cost $1 instead of $1,000 because they’re used in smartphones, which have incredible economies of scale,” says Jeffrey Martin, the CEO of a 360°-camera startup called Sphericam. Advances in optics played a part as well. Unlike traditional cameras, which have fairly narrow fields of view, 360° cameras sport exaggerated fish-eye lenses that require special optics to align and focus images across multiple points.

Most 360° cameras lack displays and viewfinders. To compensate, cam- era makers developed apps that you can download to your phone to compose shots and review the resulting images.

The cameras connect to the apps wire- lessly, and many of them allow you to upload photos and video directly from your phone to Facebook and YouTube. In turn, those sites have made it possible over the past year for people not just to post recorded 360° content but to live- stream 360° videos as well.

Because creating 360° content requires stitching together multiple images, doing it on the fly for live stream- ing represents an impressive technical achievement. Computer-vision algorithms have simplified the process so that it can be done on the camera itself, which in turn allows people to live-stream video

Bicyclists in Taiwan captured by the 360fly 4K.

These applications are possible because of the smartphone boom and innovations in computer vision t

e c

h n

o l

o g

ie s

.

K IN

M U

N L

E E

/F L

IC K

R

MA17_10_360cameras.indd 45 2/8/17 12:39 PM

46

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

with minimal delays. (It helps that most consumer-grade cameras have only two lenses and thus one stitch line. Profes- sional versions can have six to 24 lenses.) The ALLie camera supports fast stitching and live-streaming, as do Ricoh’s upcom- ing Ricoh R development kit camera and Kodak’s Orbit360 4K, which will be avail- able later this year for $500.

Spherical cameras represented 1 per- cent of worldwide consumer camera shipments in 2016 and are set to reach 4 percent in 2017, according to the research fi rm Futuresource Consulting. The pop-

ularity of these devices will benefit the virtual-reality industry as well as cam- era makers. You don’t need special VR gear to view spherical videos, but YouTube says many people look at them on smart- phones slipped into VR headsets, such as Google’s Cardboard and Daydream devices. And more people experimenting with 360° cameras means more content for other people to watch in VR.

In fact, John Carmack, the chief tech- nology offi cer of Facebook’s Oculus VR subsidiary, has predicted that people will spend less than 50 percent of their VR

time playing games. Instead, they may don VR headsets to do things like virtu- ally attend a wedding.

Once people discover spherical videos, research suggests, they shift their viewing behavior quickly. The company Human- eyes, which is developing an $800 camera that can produce 3-D spherical images, says people need to watch only about 10 hours of 360° content before they instinc- tively start trying to interact with all vid- eos. When you see 360° imagery that truly transports you somewhere else, you want it more and more.

T O

M P

O O

L E

/F L

IC K

R

MA17_10_360cameras.indd 46 2/8/17 12:39 PM

47

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW

VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

An urban scene captured by the Kodak Pixpro SP360 4K.

Kodak pixpro sp360 4k It can be mounted on a drone to produce aerial 360° videos.

MA17_10_360cameras.indd 47 2/8/17 12:39 PM

Copyright of MIT Technology Review is the property of MIT Technology Review and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.

62

BREAKTHROUGH TECHNOLOGIESTECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2

Self-Driving Trucks

MA17_10_trucks.indd 62 2/8/17 12:43 PM

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW

VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

Tractor-trailers without a human at the wheel will soon barrel onto highways near you. What will this mean for the nation’s 1.7 million truck drivers?

Self-Driving Trucks

Breakthrough Long-haul trucks that drive themselves for extended stretches on highways.

Why It Matters The technology might free truck drivers to complete routes more efficiently, but it could also erode their pay and eventually replace many of them altogether.

B y D

a v

id H

. F r

e e

d m

a n

Key Players - Otto - Volvo - Daimler - Peterbilt

Availability 5 to 10 years

R oman Mugriyev was driving his long-haul 18-wheeler down a two-lane Texas highway when he saw an oncoming car drift into his lane just a few hundred feet

ahead. There was a ditch to his right and more oncoming cars to his left, so there was little for him to do but hit his horn and brake. “I could hear the man who taught me to drive telling me what he always said was rule number one: ‘Don’t hurt anybody,’” Mugriyev recalls.

But it wasn’t going to work out that way. The errant car collided with the front of Mugriyev’s truck. It shattered his front axle, and he struggled to keep his truck and the wrecked car now fused to it from hitting anyone else as it barreled down the road. After Mugriyev finally came to a stop, he learned that the woman driving the car had been killed in the collision.

Could a computer have done better at the wheel? Or would it have done worse?

We will probably find out in the next few years, because multiple companies are now testing self-driving trucks. Although

many technical problems are still unre- solved, proponents claim that self-driving trucks will be safer and less costly. “This system often drives better than I do,” says Greg Murphy, who’s been a professional truck driver for 40 years. He now serves as a safety backup driver during tests of self-driving trucks by Otto, a San Fran- cisco company that outfits trucks with the equipment needed to drive themselves.

At first glance, the opportunities and challenges posed by self-driving trucks might seem to merely echo those asso- ciated with self-driving cars. But trucks aren’t just long cars. For one thing, the economic rationale for self-driving trucks might be even stronger than the one for driverless cars. Autonomous trucks can coördinate their movements to platoon closely together over long stretches of highway, cutting down on wind drag and saving on fuel. And letting the truck drive itself part of the time figures to help truck- ers complete their routes sooner.

But the technological obstacles fac- ing autonomous trucks are higher than

the ones for self-driving cars. Otto and other companies will need to demon- strate that sensors and code can match the situational awareness of a professional trucker—skills honed by years of expe- rience and training in piloting an easily destabilized juggernaut, with the momen- tum of 25 Honda Accords, in the face of confusing road hazards, poor surface con- ditions, and unpredictable car drivers.

And perhaps most important, if self- driving trucks do take hold, they figure to be more controversial than self- driving cars. At a time when our politics and economy are already being upended by the threats that automation poses to jobs (see “The Relentless Pace of Automation,” page 92), self-driving trucks will affect an enormous number of blue-collar work- ers. There are 1.7 million trucking jobs in the U.S., according to the Bureau of Labor Statistics. Technology is unlikely to replace truckers entirely anytime soon. But it will almost certainly alter the nature of the job, and not necessarily in ways that all would welcome.A

N D

R E

W P

A Y

N T

E R

; C O

U R

T E

S Y

O F

O T

T O

MA17_10_trucks.indd 63 2/8/17 4:59 PM

64

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2

“We’re not waiting” Otto’s headquarters, in the once-seedy South of Market section of San Francisco, isn’t much like many of the other tech startups that have transformed the area. Proudly oblivious to that neighborhood upgrade, it’s a barely renovated former furniture warehouse converted to a garage and machine shop, with semi trucks in various states of dismantlement hulk- ing over benches of tools and comput- ers. “No fancy, shiny offices here,” brags Eric Berdinis, Otto’s young and clean-cut- looking product manager.

Berdinis shows off the latest gen- eration of the company’s fast-evolving technology, which is currently installed on Volvo semis. Unlike the bolted-on, kludgy-looking hardware that’s been on testing runs for the past year, the newer versions of the company’s sensor and pro- cessing arrays are more sleekly integrated throughout the Volvo cab. The equipment includes four forward-facing video cam- eras, radar, and a box of accelerometers that Berdinis boasts is “as close as the government allows you to get to missile- guidance quality.”

Particularly key to Otto’s technology is a lidar system, which uses a pulsed laser to amass detailed data about the truck’s sur- roundings. The current third-party lidar box costs Otto in the vicinity of $100,000 each. But the company has a team design- ing a proprietary version that could cost less than $10,000.

Inside the cab is a custom-built, liquid-cooled, breadbox-size micro- supercomputer that, Berdinis claims, provides the most computing muscle ever crammed into so small a package. It is needed to crunch the vast stream of sensor data and shepherd it through the guidance algorithms that adjust braking and steering commands to compensate for the truck’s load weight. Rounding out the hardware lineup is a drive-by-wire box to turn the computer’s output into physical truck-control signals. It does

this through electromechanical actua- tors mounted to the truck’s mechanical steering, throttling, and braking systems. Two big red buttons in the cab—Otto calls them the Big Red Buttons—can cut off all self-driving activity. But even without them, the system is designed to yield to any urgent tugs on the steering wheel or heavy pumps of the pedals from anyone in the driver’s seat.

Otto was founded early in 2016 by Anthony Levandowski, who had been with Google’s self-driving-car effort, and Lior Ron, who headed up Google Maps, along with two others. It was a natural move to build on Google’s vast experience with its autonomous cars, which have driven more than two million miles on U.S. roads in several states, with an eye toward the four million trucks in the U.S. alone. Volvo Trucks, Daimler Trucks, and Peterbilt have been working on their own autonomous-truck technology.

Then, as further validation, Uber snatched Otto up for a reported $680 million last August. That deal has given Otto’s team access to roughly 500 engi- neers at Uber working on self-driving technology, according to Berdinis. Levan- dowski now heads that effort for Uber, which has said it envisions providing an overarching and largely automated trans- portation network for both goods and people.

Otto has only seven trucks on the road with its technology, but it hopes owners of many more trucks will eventually take on the equipment for free to test it out. Ber- dinis says the company is working to drive down the cost of the technology to the point where it offers a one- or two-year payback. That’s likely to mean something in the vicinity of $30,000 for a retrofit. “We expect the government to mandate this technology eventually, and for truck manufacturers to integrate it into their vehicles,” says Berdinis. “But new-truck development is on an eight-year cycle, and we’re not waiting.”

A human can push the red buttons to the right of the steering wheel to instantly take over from the self-driving system.

A shipment of Budweiser was loaded onto an autonomous

Otto truck last year.

MA17_10_trucks.indd 64 2/8/17 12:43 PM

65

The driver can sit in the back of the cab while the truck drives itself—albeit in the right lane only.

MA17_10_trucks.indd 65 2/8/17 12:43 PM

66

Greg Murphy, a longtime long- haul trucker, keeps an eye on things during tests of Otto trucks.

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

MA17_10_trucks.indd 66 2/8/17 12:43 PM

67

Roman Mugriyev wonders how well self-driving trucks would handle dangerous situations.

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW

VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

MA17_10_trucks.indd 67 2/8/17 12:43 PM

68

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

with no driver in it,” says Berdinis. But Otto does expect to free up the driver during highway cruising to remain in the back of the cab relaxing, working, or even napping. And therein lies the strongest part of the economic case for self-driving trucks. Drivers are legally restricted to 11 hours of driving a day and 60 hours a week. Given that a new big rig goes for about $150,000, and taking into account the vast delays that pulling over to rest injects into the movement of goods, trucks that can cruise nearly 24/7 could dramati- cally lower freight costs.

There are other anticipated savings from having trucks drive themselves across America’s 230,000 miles of highway. Fuel is about a third of the cost of operating a long-haul truck, and while drivers are capable of wringing maximum miles per gallon from their trucks, many are too heavy-footed on the pedals. (Berdinis says the best drivers are 30 percent more fuel-efficient than the worst ones.) Otto’s equipment is programmed to keep trucks pegged to optimal speeds and acceleration.

Then there’s the potential to cut down on accidents. Truck and bus crashes kill about 4,000 people a year in the U.S. and injure another 100,000. Driver fatigue is a factor in roughly one of seven fatal truck accidents. More than 90 percent of all acci- dents are caused at least in part by some form of driver error. We don’t yet know what fraction of those errors would be eliminated by autonomous technology— or what new errors might be introduced by it—but tests of self-driving cars suggest the technology will cut down on mistakes.

As long as self-driving trucks require a driver to remain on board, driving jobs seem safe. In some ways those jobs, which pay an average of about $40,000 a year, could even improve. For one thing, driv- ing a truck 11 hours a day is stressful. “You get physically and mentally tired,” says Mugriyev, the driver in the Texas acci- dent, which occurred in 2013. (He was not found to be at fault.) Besides being able

Ev en

if dr

ive rs

sta y o

n i n t

he ca

b, it’s

no t c

lea r

t h

e e

c o

n o

m ic

s w

il l

w o

r k

o

u t

i n

t h

e ir

f a

v o

r .

Pay cuts Last October an Otto-outfitted self-driv- ing truck carried 2,000 cases of Budweiser beer 200 kilometers down Interstate 25 in Colorado from Fort Collins to Colorado Springs—while the truck’s only human driver sat in the sleeper berth at the back of the cab without touching the vehicle’s controls.

That commercial delivery, the first ever to be handled by an autonomous heavy truck, illustrated the potential of the tech- nology. But it also demonstrated the cur- rent limitations. The human driver piloted the truck to and from the highway the old- fashioned way, because the technology doesn’t drive on small rural roads or in cities. Even after it was on the highway, a car drove ahead of the truck to make sure the far right lane remained clear. Otto’s system is programmed to stay in that lane, because on many roads trucks are restricted to the far right and are generally considered safer there. And the truck was surrounded by several cars carrying Otto personnel and Colorado State Patrol staff.

In all other testing of Otto-equipped trucks, a professional driver like Greg Murphy sits in the driver’s seat, constantly ready to take the controls at a moment’s notice, even on the highway. Another Otto employee is in the cab as well. Murphy hits the Big Red Buttons when there’s debris on the road, or construction. “My hands are always on the wheel, and I have to concen- trate pretty hard to be ready,” says Murphy. “It’s actually harder than normal driving.” (I was invited to sit in on an Otto test ride, but shortly before I was due to show up I was told there had been a scheduling miscommunication and a truck wouldn’t be available. I suspect the cancellation had more to do with that morning’s heavy rain—which can throw off autonomous vehicles—but Otto stuck to its story.)

In fact, Otto insists it has no plans to release products intended to operate trucks without a driver in the cab. “We’re at least a decade away from having trucks

MA17_10_trucks.indd 68 2/8/17 2:16 PM

69

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW

VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

Otto says it has no intention of getting drivers out of the

cab entirely—at least for the next decade.

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW

VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

MA17_10_trucks.indd 69 2/8/17 12:43 PM

70

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

to nap and relax in the cab while Otto does the driving, says Berdinis, drivers could use the time away from the wheel to catch up on trucking’s heavy paperwork, locate a “backhaul” load that would pay for the return trip, chat with family and friends, learn a second trade, or run a business. “And while they’re doing it, the drivers are still getting paid for driving,” he says.

These potential benefits could help with recruiting and training truck driv- ers—a key concern, because there’s actually a big shortage of drivers in both the U.S. and Europe. The American Trucking Asso- ciations pegs the current U.S. shortage at about 50,000 drivers and predicts that a total of nearly 900,000 new drivers will be needed over the next eight years. “We have customers calling us up saying they’ll buy 10 new trucks from us if we can provide the drivers, too,” says Carl Johan Almqvist, who heads product safety at Volvo Trucks.

One endorsement of the potential ben- efits of autonomous trucks to both truck- ing companies and drivers has come from the state government of Ohio, a trucking hub that’s home to more than 70,000 driv- ers. The state has committed $15 million to set up a 35-mile stretch of highway outside Columbus for testing self-driving trucks. The heads of both the American Trucking Associations and the Ohio Trucking Asso- ciation have publicly suggested that auton- omous trucks will be good for truckers.

However, the technology is not just a way to make the job more attractive to human drivers; it’s potentially a way for trucking companies to fill in for drivers who aren’t available. And if self-driving systems someday become accepted as capable of standing in for drivers, why keep human drivers on at all? After all, drivers account for a third of the per-mile costs of operating a truck.

Even if, as is likely for the foresee- able future, drivers stay on in the cab of self-driving trucks, it’s not clear the eco- nomics will work out in their favor. That’s because there’s currently no regulation that would require companies to pay driv- ers for the time they spend in the back of the cab. What’s more, freight companies are likely to be forced to convert the cost savings from always-rolling trucks into lower hauling charges in order to compete. Those dropping fees could put pressure on truckers’ pay. “If load prices get pushed down with this technology, the company will say, ‘You didn’t do as much driving, so you don’t make as much,’” says Mugriyev.

Safety questions Is Otto’s technology up to safely pilot- ing 80,000 pounds of truck down a busy highway? Having a driver in the cab won’t do much to make up for any shortcomings

A key detail not seen in most images of the Budweiser delivery: Otto staff and police riding nearby in cars to ensure safety. Inset: Otto’s facility in San Francisco.

MA17_10_trucks.indd 70 2/8/17 12:43 PM

71

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW

VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

in the system, given that by Otto’s own reckoning it can take up to 30 seconds for a driver resting in the back to fully orient to the driver’s seat.

The extensive history racked up by Google’s self-driving cars is encouraging, with only 20 crashes over seven years and millions of miles. Only one of the crashes was found to be the fault of the car: a traf- fic merging situation of the sort that Otto hands off to the driver.

But that record doesn’t easily trans- late into a prediction for the safety of self- driving trucks. As Berdinis notes, trucks can’t swerve to avoid a hazard the way cars can. A fast, hard turn of the steering wheel at high speed would set the truck to fishtailing and possibly jackknifing. From the moment the brakes are applied in a truck going 55 miles per hour, it takes well over the length of a football field for the vehicle to stop. There are only six inches

of lane on either side of a truck, mean- ing even small hazards at the side of the lane can’t be avoided without leaving the lane. “Many avoidance algorithms for self- driving cars just don’t apply to trucks,” says Berdinis.

One advantage for trucks is that some of the sensors can be mounted at the top of the cab, providing a high-up view that can see over traffic far ahead. But even state- of-the-art sensors can struggle to provide accurate, unambiguous data. Bright sun- light can briefly blind cameras, computers can’t always differentiate between a car by the side of the road and a big sign, and systems can be thrown off by snow, ice, and sand. They also can’t interpret facial expressions and gestures of nearby driv- ers to predict the driving behavior of other vehicles. And few systems would be able to differentiate between a hitchhiker and a construction worker gesturing to pull over.

Self-driving cars have managed to do well in mostly city driving in spite of these limitations, but at highway speeds and with limited maneuverability, trucks may come up short more often. “We’re still having problems with these challenges,” says Volvo Trucks’ Almqvist. Heavy-truck drivers typically spend months in driv- ing school, and go through thousands of miles of supervised driving, before taking full charge of a big rig. Thus, matching a human driver’s skill is harder for a self- driving truck than it is for a self-driving car. Mugriyev wonders, for example, if an autonomous system would be able to do what he did: wrestle to a safe stop a truck with a blown front axle and a smashed-up car pasted to its front.

Because of such safety concerns, Volvo has no current plans to field its auton- omous trucks on public roads. Instead, it intends to limit them to private loca- tions such as mines and ports. “On public roads, we’ll use the technology to sup- port the driver, not to replace the driver,” says Almqvist. Volvo is still unsure about social acceptance of the technology. The company sometimes identifies the license plates of passing cars when testing its autonomous trucks, and then tracks the car owners down and surveys them about their perceptions.

Berdinis acknowledges the challenges, but he insists Otto’s technology is rapidly evolving to meet them. “We won’t ship until we’re confident there are no situa- tions where we’d need a human to imme- diately take control of the truck,” he says.

Otto will also have to convince regula- tors its systems are ready for the highway. Unlike Uber, which has relied on the con- sumer popularity of its passenger service to take to the roads first and wrestle with regulations later, Otto will do everything strictly by the book, notes Berdinis.

Even Volvo’s Almqvist thinks the tech- nology will make it to public roads in the not-too-distant future. But timing will be crucial, he adds: “If we do it too soon and have an accident, we’ll hurt the industry. And if you lose the public’s trust, it’s very difficult to regain it.”

MA17_10_trucks.indd 71 2/8/17 2:16 PM

Copyright of MIT Technology Review is the property of MIT Technology Review and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.

S P E C I A L R E P O R T

Top 10 Emerging Technologies of 2017 Which 10 disruptive solutions are now poised to change the world?

June 26, 2017 | Credit: World Economic Forum

10 Emerging Technologies to Watch Innovations that are on the verge of making a difference to society

June 26, 2017

Blood Tests Allow for Scalpel-Free Biopsies Ultrasensitive blood tests known as liquid biopsies promise to improve cancer diagnosis and care

June 26, 2017 — Apurv Mishra

E N G I N E E R I N G

B I O T E C H

Off-Grid Devices Draw Drinking Water from Dry Air Technologies that pull moisture from the air are now solar-powered

June 26, 2017 — Donna J. Nelson and Jeffrey Carbeck

P U B L I C H E A LT H

Scientific American is part of Springer Nature, which owns or has commercial relations with thousands of scientific publications (many of them can be found at www.springernature.com/us). Scientific American maintains a strict policy of editorial independence in reporting developments

in science to our readers.

© 2023 SCIENTIFIC AMERICAN, A DIVISION OF SPRINGER NATURE AMERICA, INC.

ALL RIGHTS RESERVED.

Deep-Learning Networks Rival Human Vision June 26, 2017 — Apurv Mishra

Artificial Leaf Turns Carbon Dioxide Into Liquid Fuel June 26, 2017 — Javier Garcia Martinez

Human Cell Atlas Opens a New Window to Health and Disease June 26, 2017 — Sang Yup Lee

Precision Farming Increases Crop Yields June 26, 2017 — Geoffrey Ling and Blake Bextine

Affordable Catalysts Give Green Vehicles a Push June 26, 2017 — Donna J. Nelson

Genomic Vaccines Fight Disease in Ways Not Possible Before June 26, 2017 — Geoffrey Ling

Sustainable Design of Communities Dramatically Reduces Waste June 26, 2017 — Daniel M. Kammen

Quantum Computing Becomes More Accessible June 26, 2017 — Dario Gil

A D V E R T I S E M E N T

Th e C

e l

l At

las

58

BREAKTHROUGH TECHNOLOGIES

MA17_10_cellatlas.indd 58 2/8/17 4:06 PM

Th e C

e l

l At

las

MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2TECHNOLOGYREVIEW.COM BREAKTHROUGH TECHNOLOGIES

Breakthrough A master catalog of every cell type in the human body.

Why It Matters Super-accurate mod- els of human physiol- ogy will speed up the discovery and testing of new drugs.

Availability 5 years

By Steve Connor

Biology’s next mega-project will find out what we’re really made of.

Key Players - Broad Institute - Sanger Institute - Chan Zuckerberg Biohub

FRED TOMASELLI Airborne Event 2003 Mixed media, acrylic paint, resin on woodC

O U

R T

E S

Y O

F F

R E

D T

O M

A S

E L

L I A

N D

J A

M E

S C

O H

A N

. N E

W Y

O R

K

MA17_10_cellatlas.indd 59 2/8/17 5:14 PM

60

TECHNOLOGYREVIEW.COM BREAKTHROUGH TECHNOLOGIES MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2

In 1665, Robert Hooke peered down his microscope at a piece of cork and discovered little boxes that reminded him of rooms in a monastery. Being the first scientist to describe cells,

Hooke would be amazed by biology’s next mega-project: a scheme to individually capture and scrutinize millions of cells using the most powerful tools in modern genomics and cell biology.

The objective is to construct the first comprehensive “cell atlas,” or map of human cells, a technological marvel that should comprehensively reveal, for the first time, what human bodies are actually made of and provide scientists a sophisticated new model of biology that could speed the search for drugs.

To perform the task of cataloguing the 37.2 trillion cells of the human body, an international consortium of scientists from the U.S., U.K., Sweden, Israel, the Netherlands, and Japan is being assem- bled to assign each a molecular signature and also give each type a zip code in the three-dimensional space of our bodies.

“We will see some things that we expect, things we know to exist, but I’m sure there will be completely novel things,” says Mike Stubbington, head of the cell atlas team at the Sanger Insti- tute in the U.K. “I think there will be surprises.”

Previous attempts at describing cells, from the hairy neurons that populate the brain and spinal cord to the glutinous fat cells of the skin, suggest there are about 300 variations in total. But the true figure is undoubtedly larger. Analyz- ing molecular differences between cells has already revealed, for example, two new types of retinal cells that escaped decades of investigation of the eye; a cell that forms the first line of defense against pathogens and makes up four in every 10,000 blood cells; and a newly spot- ted immune cell that uniquely produces a steroid that appears to suppress the immune response.

Three technologies are coming together to make this new type of map- ping possible. The first is known as “cel- lular microfluidics.” Individual cells are separated, tagged with tiny beads, and manipulated in droplets of oil that are shunted like cars down the narrow, one- way streets of artificial capillaries etched into a tiny chip, so they can be corralled, cracked open, and studied one by one.

The second is the ability to iden- tify the genes active in single cells by decoding them in superfast and efficient sequencing machines at a cost of just a few cents per cell. One scientist can now process 10,000 cells in a single day.

The third technology uses novel labeling and staining techniques that can locate each type of cell—on the basis of its gene activity—at a specific zip code in a human organ or tissue.

Behind the cell atlas are big-science powerhouses including Britain’s Sanger Institute, the Broad Institute of MIT and Harvard, and a new “Biohub” in Cali- fornia funded by Facebook CEO Mark Zuckerberg. In September Zuckerberg and his wife, Priscilla Chan, made the cell atlas the inaugural target of a $3 billion donation to medical research.

Fig. 1 Robert Hooke’s drawing of cork, as seen through a microscope (1665).

Fig. 2 Sperm containing a homunculus (Nicholas Hartsoeker, 1695). Fig. 3 Daguerreotypes of blood from humans, camels, and toads (A. Donné, 1845). Fig. 4 Plant cells (J. M. Schleiden, 1838).

Fig. 5 Sketches of animal cells (Theodor Schwann, 1839).

Fig. 6 A nerve (A. von Kolliker, 1852).

Sci en

tis ts

ar e b

uil din

g a n u

ltr a-

de tai

led

“h um

an ce

ll a tla

s” tha

t d efi

ne s l

ivi ng

ce lls

by w

ha t

g o

e s

o n

i n

s id

e t

h e

m .

MA17_10_cellatlas.indd 60 2/8/17 4:06 PM

61

MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2TECHNOLOGYREVIEW.COM BREAKTHROUGH TECHNOLOGIES

Fig. 1 Fig. 2

Fig. 4

Fig. 5

Fig. 6

Fig. 3

W E

L L

C O

M E

L IB

R A

R Y

, L O

N D

O N

MA17_10_cellatlas.indd 61 2/8/17 5:14 PM

Copyright of MIT Technology Review is the property of MIT Technology Review and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

“G o, go!” was the thought racing through Grégoire Courtine’s mind.

The French neuro- scientist was watching a

macaque monkey as it hunched aggres- sively at one end of a treadmill. His team had used a blade to slice halfway through the animal’s spinal cord, paralyzing its right leg. Now Courtine wanted to prove he could get the monkey walking again. To do it, he and colleagues had installed a recording device beneath its skull, touch- ing its motor cortex, and sutured a pad of flexible electrodes around the animal’s spinal cord, below the injury. A wire-

Scientists are making remarkable progress at using brain implants to restore the freedom of movement that spinal cord injuries take away.

REVERSING Paralysis

B y A

n t

o n

io R

e g

a l

a d

o

Breakthrough Wireless brain-body electronic interfaces to bypass damage to the nervous system.

Why It Matters Thousands of people suffer paralyzing injuries every year.

Key Players - École Polytechnique

Fédérale de Lausanne - Wyss Center for Bio and

Neuroengineering - University of Pittsburgh - Case Western Reserve

University

Availability 10 to 15 years

MA17_10_reversing_paralysis 82 2/8/17 11:14 AM

83

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW

VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

REVERSING Paralysis

An implant shown on a silicone model of a primate brain.

A L

A IN

H E

R Z

O G

/ E

P F

L

MA17_10_reversing_paralysis 83 2/8/17 11:14 AM

84

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

less connection joined the two electronic devices.

The result: a system that read the monkey’s intention to move and then transmitted it immediately in the form of bursts of electrical stimulation to its spine. Soon enough, the monkey’s right leg began to move. Extend and flex. Extend and flex. It hobbled forward. “The mon- key was thinking, and then boom, it was walking,” recalls an exultant Courtine, a professor with Switzerland’s École Poly- technique Fédérale de Lausanne.

In recent years, lab animals and a few people have controlled computer cur-

sors or robotic arms with their thoughts, thanks to a brain implant wired to machines. Now researchers are taking a significant next step toward reversing paralysis once and for all. They are wire- lessly connecting the brain-reading tech- nology directly to electrical stimulators on the body, creating what Courtine calls a “neural bypass” so that people’s thoughts can again move their limbs.

At Case Western Reserve University, in Cleveland, a middle-aged quadriple- gic—he can’t move anything but his head and shoulder—agreed to let doctors place two recording implants in his brain, of the

Grégoire Courtine holds the two main parts of the brain-spine interface.

H IL

L A

R Y

S A

N C

T U

A R

Y /

E P

F L

MA17_10_reversing_paralysis 84 2/8/17 11:14 AM

85

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW

VOL. 120 | NO. 2BREAKTHROUGH TECHNOLOGIES

Top right: Flexible electrodes developed to simulate the spinal cord.

Above: A model of a wireless neurocommunication device sits on a skull.

Top left: A close-up of a brain-reading chip, bristling with electrodes.

Milestones in Neural Bypass

1961 Physician and inventor William F. House tests the first cochlear implant to restore hearing. The devices will go on to benefit more than 250,000 people.

1998 Doctors install a single electrode in the brain of a paralyzed man unable to speak. He uses it to communicate through a computer.

2008 A monkey’s brain signals are sent over the Internet from the U.S. to Japan, causing a robot to walk on a treadmill.

2013 U.S. regulators approve a “bionic eye” sold by the company Second Sight. It uses a chip sutured to the retina to bypass injured photoreceptors.

2014-2015 Ohio doctors launch efforts to “reanimate” the arms of two different paralyzed men. The thoughts of each are transmitted to electrodes on their arms, causing their hands to open and shut.

2016 28-year-old Nathan Copeland operates a robotic hand that, via a brain implant, allows him to “feel” the fingers. He fist-bumps Barack Obama during a presidential visit to a lab in Pittsburgh.A

L A

IN H

E R

Z O

G /

E P

F L

; C O

U R

T E

S Y

O F

T H

E W

Y S

S C

E N

T E

R

MA17_10_reversing_paralysis 85 2/8/17 11:14 AM

86

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

In these frames of a video made by EPFL researchers, a monkey with a spinal cord injury that paralyzed its right leg is able to walk again.

E P

F L

MA17_10_reversing_paralysis 86 2/8/17 11:14 AM

87

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW

VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

same type Courtine used in the monkeys. Made of silicon, and smaller than a post- age stamp, they bristle with a hundred hair-size metal probes that can “listen” as neurons fire off commands.

To complete the bypass, the Case team, led by Robert Kirsch and Bolu Ajiboye, also slid more than 16 fine elec- trodes into the muscles of the man’s arm and hand. In videos of the experiment, the volunteer can be seen slowly raising his arm with the help of a spring-loaded arm rest, and willing his hand to open and close. He even raises a cup with a straw to his lips. Without the system, he can’t do any of that.

Just try sitting on your hands for a day. That will give you an idea of the shat- tering consequences of spinal cord injury. You can’t scratch your nose or tousle a child’s hair. “But if you have this,” says Courtine, reaching for a red espresso cup and raising it to his mouth with an actor’s exaggerated motion, “it changes your life.”

The Case results, pending publica- tion in a medical journal, are a part of a broader effort to use implanted electron- ics to restore various senses and abili- ties. Besides treating paralysis, scientists hope to use so-called neural prosthetics to reverse blindness with chips placed in the eye, and maybe restore memo- ries lost to Alzheimer’s disease (see “10 Breakthrough Technologies 2013: Mem- ory Implants”).

And they know it could work. Con- sider cochlear implants, which use a microphone to relay signals directly to the auditory nerve, routing around non- working parts of the inner ear. Videos of wide-eyed deaf children hearing their mothers for the first time go viral on the Internet every month. More than 250,000 cases of deafness have been treated.

But it’s been harder to turn neural prosthetics into something that helps par- alyzed people. A patient first used a brain probe to move a computer cursor across a screen back in 1998. That and several

“P eo

ple w

ou ld

pr efe

r t o b

e r est

or ed

to th

eir

e v

e r

y d

a y

s e

l f

. T h

e y

w a

n t

t

o b

e r

e a

n im

a t

e d

.” other spectacular brain-control feats haven’t had any broader practical use. The technology remains too radical and too complex to get out of the lab. “Twenty years of work and nothing in the clinic!” Courtine exclaims, brushing his hair back. “We keep pushing the limits, but it is an important question if this entire field will ever have a product.”

Courtine’s laboratory is located in a vertiginous glass-and-steel building in Geneva that also houses a $100 million center that the Swiss billionaire Hansjörg Wyss funded specifically to solve the remaining technical obstacles to neuro- technologies like the spinal cord bypass. It’s hiring experts from medical-device makers and Swiss watch companies and has outfitted clean rooms where gold wires are printed onto rubbery electrodes that can stretch as our bodies do.

The head of the center is John Donoghue, an American who led the early development of brain implants in the U.S. (see “Implanting Hope,” March 2005) and who moved to Geneva two years ago. He is now trying to assemble in one place the enormous technical resources and talent—skilled neuroscientists, technol- ogists, clinicians—needed to create com- mercially viable systems.

Among Donoghue’s top priorities is a “neurocomm,” an ultra-compact wireless device that can collect data from the brain at Internet speed. “A radio inside your head,” Donoghue calls it, and “the most sophisticated brain communicator in the world.” The matchbox-size prototypes are made of biocompatible titanium with a sapphire window. Courtine used an ear- lier, bulkier version in his monkey tests.

As complex as they are, and as slow as progress has been, neural bypasses are worth pursuing because patients desire them, Donoghue says. “Ask someone if they would like to move their own arm,” he says. “People would prefer to be restored to their everyday self. They want to be reanimated.”

MA17_10_reversing_paralysis.indd 87 2/8/17 2:13 PM

Copyright of MIT Technology Review is the property of MIT Technology Review and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

By converting heat to focused beams of light, a new solar device could create cheap and continuous power.

S olar panels cover a growing num- ber of rooftops, but even decades after they were first developed, the slabs of silicon remain bulky, expensive, and inefficient. Funda-

mental limitations prevent these conven- tional photovoltaics from absorbing more than a fraction of the energy in sunlight.

But a team of MIT scientists has built a different sort of solar energy device that uses inventive engineering and advances in materials science to capture far more of the sun’s energy. The trick is to first turn sunlight into heat and then convert it back into light, but now focused within the spec- trum that solar cells can use. While various researchers have been working for years on

Hot Solar Cells

Breakthrough A solar power device that could theoretically double the efficiency of conventional solar cells.

Why It Matters The new design could lead to inexpensive solar power that keeps working after the sun sets.

B y J

a m

e s

T e

m p

l e

Key Players - David Bierman, Marin

Soljacic, and Evelyn Wang, MIT

- Vladimir Shalaev, Purdue University

Availability 10 to 15 years

P H

O T

O G

R A

P H

Y B

Y K

E N

R IC

H A

R D

S O

N

MA17_10_hot_solar_cells.indd 52 2/7/17 3:42 PM

53

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW

VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

Hot Solar Cells

A view of the solar device seen by looking through the equipment used to focus simulated sunlight on it.

MA17_10_hot_solar_cells 53 2/6/17 5:16 PM

54

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

so-called solar thermophotovoltaics, the MIT device is the first one to absorb more energy than its photovoltaic cell alone, demonstrating that the approach could dramatically increase efficiency.

Standard silicon solar cells mainly capture the visual light from violet to red. That and other factors mean that they can never turn more than around 32 percent of the energy in sunlight into electricity. The MIT device is still a crude prototype, operating at just 6.8 percent efficiency— but with various enhancements it could be roughly twice as efficient as conven- tional photovoltaics.

The key step in creating the device was the development of something called an absorber-emitter. It essentially acts as a light funnel above the solar cells. The absorbing layer is built from solid black carbon nanotubes that capture all the energy in sunlight and convert most of it into heat. As temperatures reach around 1,000 °C, the adjacent emitting layer radiates that energy back out as light, now mostly narrowed to bands that the photovoltaic cells can absorb. The emitter is made from a photonic crystal, a structure that can be designed at the nanoscale to control which wavelengths

of light flow through it. Another critical advance was the addition of a highly spe- cialized optical filter that transmits the tailored light while reflecting nearly all the unusable photons back. This “pho- ton recycling” produces more heat, which generates more of the light that the solar cell can absorb, improving the efficiency of the system.

There are some downsides to the MIT team’s approach, including the relatively high cost of certain components. It also currently works only in a vacuum. But the economics should improve as efficiency levels climb, and the researchers now have

Above: Black carbon nanotubes sit on top of the absorber-emitter layer, collecting energy across the solar spectrum and converting it to heat.

Facing page: The absorber-emitter layer is situated above an optical filter and photovoltaic cell, which is visible underneath.

MA17_10_hot_solar_cells 54 2/6/17 5:16 PM

55

MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

Th e d

ev ice

ev en

tua lly

co uld

be tw

ice as

e

f fi

c ie

n t

a s

c o

n v

e n

t io

n a

l

p h

o t

o v

o lt

a ic

s .

MA17_10_hot_solar_cells 55 2/6/17 5:16 PM

56

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

a clear path to achieving that. “We can fur- ther tailor the components now that we’ve improved our understanding of what we need to get to higher efficiencies,” says Evelyn Wang, an associate professor who helped lead the effort.

The researchers are also exploring ways to take advantage of another strength of solar thermophotovoltaics. Because heat

is easier to store than electricity, it should be possible to divert excess amounts gener- ated by the device to a thermal storage sys- tem, which could then be used to produce electricity even when the sun isn’t shining. If the researchers can incorporate a stor- age device and ratchet up efficiency levels, the system could one day deliver clean, cheap—and continuous—solar power.

MA17_10_hot_solar_cells 56 2/6/17 5:16 PM

57

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW

VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

Concentrated light from a solar simulator shines through the window of a vacuum chamber, where it reaches the solar thermophotovoltaic device and generates electricity.

MA17_10_hot_solar_cells 57 2/6/17 5:16 PM

Copyright of MIT Technology Review is the property of MIT Technology Review and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.

76

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

QuTe ch

re se

ar ch

er D

an ië

l B ou

m an

p ee

rs d

ow n

int o th

e ro om where delicate quantum experiments are happening at ultra-cold tem

peratures.

P r

a c

t ic

a l

Qu

an tum

Co mp

ute rs

MA17_10_quantum.indd 76 2/8/17 3:08 PM

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW

VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

By Russ Juskalian

Advances at Google, Intel, and several research groups indicate that computers with previously unimaginable power are finally within reach.

Breakthrough The fabrication of stable qubits, the basic unit of quantum computers.

Why It Matters Quantum computers could be exponentially faster at running artificial-intelligence programs and handling complex simulations and scheduling problems. They could even create uncrackable encryption.

Key Players - QuTech - Intel - Microsoft - Google - IBM

Availability 4 to 5 years

P r

a c

t ic

a l

Qu

an tum

Co mp

ute rs

M A

T H

IJ S

L A

B A

D IE

MA17_10_quantum.indd 77 2/8/17 3:08 PM

78

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

One of the labs at QuTech, a Dutch research institute, is responsi- ble for some of the world’s most advanced work on quantum computing, but it looks like an

HVAC testing facility. Tucked away in a quiet corner of the applied sciences build- ing at Delft University of Technology, the space is devoid of people. Buzzing with resonant waves as if occupied by a swarm of electric katydids, it is cluttered by tan- gles of insulated tubes, wires, and control hardware erupting from big blue cylinders on three and four legs.

Inside the blue cylinders—essen- tially supercharged refrigerators—spooky quantum- mechanical things are happen- ing where nanowires, semiconductors, and superconductors meet at just a hair above absolute zero. It’s here, down at the limits of physics, that solid materials give rise to so-called quasiparticles, whose unusual behavior gives them the potential to serve as the key components of quantum com- puters. And this lab in particular has taken big steps toward finally bringing those computers to fruition. In a few years they could rewrite encryption, materials sci- ence, pharmaceutical research, and arti- ficial intelligence.

Every year quantum computing comes up as a candidate for this Breakthrough Technologies list, and every year we reach the same conclusion: not yet. Indeed, for years qubits and quantum comput- ers existed mainly on paper, or in fragile experiments to determine their feasibil- ity. (The Canadian company D-Wave Sys- tems has been selling machines it calls quantum computers for a while, using a specialized technology called quantum annealing. The approach, skeptics say, is at best applicable to a very constrained set of computations and might offer no speed advantage over classical systems.) This year, however, a raft of previously theoretical designs are actually being built. Also new this year is the increased avail- ability of corporate funding—from Google,

Previously theoretical designs a

r e

a

c t

u -

a l

ly

b e

in g

b

u ilt

.

MA17_10_quantum.indd 78 2/8/17 3:08 PM

79

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW

VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES T

his blue refrigerator gets dow

n to just above absolute zero, m

aking quantum

experim ents possible

on tiny chips deep inside it. O

n subsequent pages are scenes from

the D

elft lab w here the

experim ents are

prepared.

MA17_10_quantum.indd 79 2/8/17 3:08 PM

80

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

What Is a Quantum Computer?

At the heart of quantum computing is the quantum bit, or qubit, a basic unit of information analogous to the 0s and 1s represented by transistors in your computer. Qubits have much more power than classical bits because of two unique properties: they can represent both 1 and 0 at the same time, and they can affect other qubits via a phenomenon known as quantum entanglement. That lets quantum computers take shortcuts to the right answers in certain types of calculations.

R U

S S

J U

S K

A L

IA N

MA17_10_quantum.indd 80 2/8/17 3:08 PM

81

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW

VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

IBM, Intel, and Microsoft, among others— for both research and the development of assorted technologies needed to actually build a working machine: microelectron- ics, complex circuits, and control software.

The project at Delft, led by Leo Kouwenhoven, a professor who was recently hired by Microsoft, aims to over- come one of the most long-standing obsta- cles to building quantum computers: the fact that qubits, the basic units of quantum information, are extremely susceptible to noise and therefore error. For qubits to be useful, they must achieve both quantum superposition (a property something like being in two physical states simultane- ously) and entanglement (a phenomenon where pairs of qubits are linked so that what happens to one can instantly affect the other, even when they’re physically separated). These delicate conditions are easily upset by the slightest disturbance, like vibrations or fluctuating electric fields.

People have long wrestled with this problem in efforts to build quantum com- puters, which could make it possible to solve problems so complex they exceed the reach of today’s best computers. But now Kouwenhoven and his colleagues believe the qubits they are creating could eventually be inherently protected—as sta- ble as knots in a rope. “Despite deform-

ing the rope, pulling on it, whatever,” says Kouwenhoven, the knots remain and “you don’t change the information.” Such stability would allow researchers to scale up quantum computers by substan- tially reducing the computational power required for error correction.

Kouwenhoven’s work relies on manipulating unique quasiparticles that weren’t even discovered until 2012. And it’s just one of several impressive steps being taken. In the same lab, Lieven Vandersypen, backed by Intel, is showing how quantum circuits can be manufac- tured on traditional silicon wafers.

Quantum computers will be particu- larly suited to factoring large numbers (making it easy to crack many of today’s encryption techniques and probably pro- viding uncrackable replacements), solv- ing complex optimization problems, and executing machine-learning algorithms. And there will be applications nobody has yet envisioned.

Soon, however, we might have a bet- ter idea of what they can do. Until now, researchers have built fully programma- ble five-qubit computers and more frag- ile 10- to 20-qubit test systems. Neither kind of machine is capable of much. But the head of Google’s quantum comput- ing effort, Harmut Neven, says his team

is on target to build a 49-qubit system by as soon as a year from now. The target of around 50 qubits isn’t an arbitrary one. It’s a threshold, known as quantum suprem- acy, beyond which no classical supercom- puter would be capable of handling the exponential growth in memory and com- munications bandwidth needed to sim- ulate its quantum counterpart. In other words, the top supercomputer systems can currently do all the same things that five- to 20-qubit quantum computers can, but at around 50 qubits this becomes physi- cally impossible.

All the academic and corporate quan- tum researchers I spoke with agreed that somewhere between 30 and 100 qubits— particularly qubits stable enough to per- form a wide range of computations for longer durations—is where quantum com- puters start to have commercial value. And as soon as two to five years from now, such systems are likely to be for sale. Eventually, expect 100,000-qubit systems, which will disrupt the materials, chemistry, and drug industries by making accurate molecular- scale models possible for the discovery of new materials and drugs. And a million- physical-qubit system, whose general computing applications are still difficult to even fathom? It’s conceivable, says Neven, “on the inside of 10 years.”

MA17_10_quantum.indd 81 2/8/17 3:08 PM

Copyright of MIT Technology Review is the property of MIT Technology Review and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.

72

BREAKTHROUGH TECHNOLOGIES

Pa yin

g w ith

Y o

u r

F a

c e

MA17_10_paying_with_your_face.indd 72 2/8/17 10:48 AM

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW

VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

Pa yin

g w ith

Y o

u r

F a

c eBreakthrough

Face recognition technology that is finally accurate enough to be widely used in financial transactions and other everyday applications.

Why It Matters The technology offers a secure and extremely convenient method of payment but could raise privacy concerns.

By Will Knight

Face-detecting systems in China now authorize payments, provide access to facilities, and track down criminals. Will other countries follow?

Key Players - Face++ - Baidu - Alibaba

Availability Now

Y O

S H

I S O

D E

O K

A

MA17_10_paying_with_your_face.indd 73 2/8/17 10:48 AM

74

TECHNOLOGYREVIEW.COM BREAKTHROUGH TECHNOLOGIES MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2

S hortly after walking through the door at Face++, a Chinese startup valued at roughly a billion dollars, I see my face, unshaven and look- ing a bit jet-lagged, flash up on a

large screen near the entrance. Having been added to a database, my

face now provides automatic access to the building. It can also be used to monitor my movements through each room inside. As I tour the offices of Face++ (pronounced “face plus plus”), located in a suburb of Beijing, I see it appear on several more screens, automatically captured from countless angles by the company’s soft- ware. On one screen a video shows the soft- ware tracking 83 different points on my face simultaneously. It’s a little creepy, but undeniably impressive.

Over the past few years, computers have become incredibly good at recogniz- ing faces, and the technology is expanding quickly in China in the interest of both surveillance and convenience. Face recog- nition might transform everything from policing to the way people interact every day with banks, stores, and transporta- tion services.

Face++ pinpoints 83 points on a face. The distance between them provides a means of identification.

Technology from Face++ is already being used in several popular apps. It is possible to transfer money through Alipay, a mobile payment app used by more than 120 million people in China, using only your face as credentials. Meanwhile, Didi, China’s dominant ride-hailing company, uses the Face++ software to let passengers confirm that the person behind the wheel is a legitimate driver. (A “liveness” test, designed to prevent anyone from duping the system with a photo, requires people being scanned to move their head or speak while the app scans them.)

The technology figures to take off in China first because of the country’s atti- tudes toward surveillance and privacy. Unlike, say, the United States, China has a large centralized database of ID card pho- tos. During my time at Face++, I saw how local governments are using its software to identify suspected criminals in video from surveillance cameras, which are omni- present in the country. This is especially impressive—albeit somewhat dystopian— because the footage analyzed is far from perfect, and because mug shots or other images on file may be several years old.

Facial recognition has existed for decades, but only now is it accurate enough to be used in secure financial transactions. The new versions use deep learning, an artificial-intelligence technique that is especially effective for image recognition because it makes a computer zero in on the facial features that will most reliably identify a person (see “10 Breakthrough Technologies 2013: Deep Learning”).

“The face recognition market is huge,” says Shiliang Zhang, an assistant profes- sor at Peking University who specializes in machine learning and image process- ing. Zhang heads a lab not far from the offices of Face++. When I arrived, his stu- dents were working away furiously in a dozen or so cubicles. “In China security is very important, and we also have lots of people,” he says. “Lots of companies are working on it.”

One such company is Baidu, which operates China’s most popular search engine, along with other services. Baidu researchers have published papers show- ing that their software rivals most humans in its ability to recognize a face. In Janu- ary, the company proved this by taking

C O

U R

T E

S Y

O F

F A

C E

+ +

MA17_10_paying_with_your_face.indd 74 2/8/17 10:48 AM

75

TECHNOLOGYREVIEW.COM BREAKTHROUGH TECHNOLOGIES MIT TECHNOLOGY REVIEW

VOL. 120 | NO. 2

part in a TV show featuring people who are remarkably good at identifying adults from their baby photos. Baidu’s system outshined them.

Now Baidu is developing a system that lets people pick up rail tickets by showing their face. The company is already working with the government of Wuzhen, a historic tourist destination, to provide access to many of its attractions without a ticket. This involves scanning millions of faces in a database to find a match, which Baidu says it can do with 99 percent accuracy.

Jie Tang, an associate professor at Tsinghua University who advised the founders of Face++ as students, says the convenience of the technology is what appeals most to people in China. Some apartment complexes use facial recogni- tion to provide access, and shops and res- taurants are looking to the technology to make the customer experience smoother. Not only can he pay for things this way, he says, but the staff in some coffee shops are now alerted by a facial recognition system when he walks in: “They say, ‘Hello, Mr. Tang.’”

Lo ca

l g ov

ern me

nts ar

e u sin

g t he

so ftw

ar e t

o i de

nti fy

su sp

ect ed

cr im

ina ls

in vid

eo fr

om su

rv eil

lan ce

c a

m e

r a

s , w

h ic

h a

r e

o

m n

ip r

e s

e n

t i

n C

h in

a .

Employees simply show their face to gain entry to the company’s headquarters.

The system captured MIT Technology Review’s

Will Knight as he moved through Face++’s offices.

C O

U R

T E

S Y

O F

F A

C E

+ +

; W IL

L K

N IG

H T

MA17_10_paying_with_your_face.indd 75 2/8/17 10:49 AM

Copyright of MIT Technology Review is the property of MIT Technology Review and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.

B o

t n

e t

s of

Th ing

s

88

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

MA17_10_botnets.indd 88 2/7/17 3:27 PM

B o

t n

e t

s of

Th ing

s

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW

VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

By Bruce Schneier

The relentless push to add connectivity to home gadgets is creating dangerous side effects that figure to get even worse.

Breakthrough Malware that takes control of webcams, video recorders, and other consumer devices to cause widespread Internet outages.

Why It Matters Botnets based on this software are disrupting larger and larger swaths of the Internet—and getting harder to stop.

Key Players - Whoever created the

Mirai botnet software - Anyone who runs a poorly

secured device online— including you?

Availability Now

R O

B E

R T

B E

A T

T Y

MA17_10_botnets.indd 89 2/8/17 1:22 PM

90

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

B otnets have existed for at least a decade. As early as 2000, hack- ers were breaking into com- puters over the Internet and controlling them en masse from

centralized systems. Among other things, the hackers used the combined comput- ing power of these botnets to launch dis- tributed denial-of-service attacks, which flood websites with traffic to take them down.

But now the problem is getting worse, thanks to a flood of cheap webcams, dig- ital video recorders, and other gadgets in the “Internet of things.” Because these devices typically have little or no secu- rity, hackers can take them over with lit- tle effort. And that makes it easier than ever to build huge botnets that take down much more than one site at a time.

In October, a botnet made up of 100,000 compromised gadgets knocked

an Internet infrastructure provider par- tially offline. Taking down that provider, Dyn, resulted in a cascade of effects that ultimately caused a long list of high- profile websites, including Twitter and Netflix, to temporarily disappear from the Internet. More attacks are sure to follow: the botnet that attacked Dyn was created with publicly available malware called Mirai that largely automates the process of coöpting computers.

The best defense would be for every- thing online to run only secure software, so botnets couldn’t be created in the first place. This isn’t going to happen anytime soon. Internet of things devices are not designed with security in mind and often have no way of being patched. The things that have become part of Mirai botnets, for example, will be vulnerable until their owners throw them away. Botnets will get larger and more powerful simply because

the number of vulnerable devices will go up by orders of magnitude over the next few years.

What do hackers do with them? Many things.

Botnets are used to commit click fraud. Click fraud is a scheme to fool advertisers into thinking that people are clicking on, or viewing, their ads. There are lots of ways to commit click fraud, but the easiest is probably for the attacker to embed a Google ad in a Web page he owns. Google ads pay a site owner accord- ing to the number of people who click on them. The attacker instructs all the com- puters on his botnet to repeatedly visit the Web page and click on the ad. Dot, dot, dot, PROFIT! If the botnet makers figure out more effective ways to siphon revenue from big companies online, we could see the whole advertising model of the Internet crumble.

This map shows the extent of some of the Internet outages caused by denial-of-service attacks on Dyn on October 21, 2016. Dyn operates domain-name servers that connect end users to websites.

O U

T A

G E

D A

T A

F R

O M

D O

W N

D E

T E

C T

O R

MA17_10_botnets.indd 90 2/8/17 1:22 PM

91

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW

VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

Similarly, botnets can be used to evade spam filters, which work partly by know- ing which computers are sending millions of e-mails. They can speed up password guessing to break into online accounts, mine bitcoins, and do anything else that requires a large network of computers. This is why botnets are big businesses. Criminal organizations rent time on them.

But the botnet activities that most often make headlines are denial-of- service attacks. Dyn seems to have been the vic- tim of some angry hackers, but more financially motivated groups use these attacks as a form of extortion. Political groups use them to silence websites they don’t like. Such attacks will certainly be a tactic in any future cyberwar.

Once you know a botnet exists, you can attack its command-and-control sys- tem. When botnets were rare, this tactic was effective. As they get more common, this piecemeal defense will become less so. You can also secure yourself against the effects of botnets. For example, several companies sell defenses against denial-of- service attacks. Their effectiveness varies, depending on the severity of the attack and the type of service.

But overall, the trends favor the attacker. Expect more attacks like the one against Dyn in the coming year.

Bruce Schneier, chief technology officer at IBM Resilient, is the author of 13 books on cryptography and data security.

Bo tne

ts wi

ll g et

lar ge

r an

d m or

e p ow

erf ul

sim ply

be

ca us

e t he

nu mb

er of

vu lne

ra ble

de vic

es wi

ll g o

u p

b y

o r

d e

r s

o

f m

a g

n it

u d

e

o v

e r

t h

e n

e x

t

f e

w y

e a

r s

.

Worldwide number of “Internet- connectable” devices

D A

T A

F R

O M

IH S

M A

R K

IT

2011 8.0 billion

2012 9.3 billion

2013 11.1 billion

2015 15.2 billion

2016 17.4 billion

2014 13.1 billion

MA17_10_botnets.indd 91 2/7/17 3:27 PM

Copyright of MIT Technology Review is the property of MIT Technology Review and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.

48

BREAKTHROUGH TECHNOLOGIES

MA17_10_genetherapy2.0.indd 48 2/8/17 12:38 PM

MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2TECHNOLOGYREVIEW.COM BREAKTHROUGH TECHNOLOGIES

Ge ne

Th era

py 2

.0

By Emily Mullin

Breakthrough First gene therapies on track for approval in the U.S. More are on the way.

Why It Matters Thousands of diseases stem from an error in a single gene. New treatments could cure them.

Key Players - Spark Therapeutics - BioMarin - BlueBird Bio - GenSight Biologics - UniQure

Availability Now

Scientists have solved fundamental problems that were holding back cures for rare hereditary disorders. Next we’ll see if the same approach can take on cancer, heart disease, and other common illnesses.

A R

M A

N D

O V

E V

E

MA17_10_genetherapy2.0.indd 49 2/8/17 4:38 PM

50

TECHNOLOGYREVIEW.COM BREAKTHROUGH TECHNOLOGIES MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2

W hen Kala Looks gave birth to fraternal twin boys in January 2015, she and her husband, Philip, had no idea that one of them was

harboring a deadly mutation in his genes. At three months old, their son Levi

was diagnosed with severe combined immune deficiency, or SCID, which ren- ders the body defenseless against infec- tions. Levi’s blood had only a few immune cells essential to fighting disease. Soon he would lose them and have no immune system at all.

Kala and Philip frantically began sanitizing their home to keep Levi alive. They got rid of the family cat, sprayed every surface with Lysol, and boiled the twins’ toys in hot water. Philip would strap on a surgical mask when he came home from work.

At first, Kala and Philip thought their only option was to get Levi a bone mar- row transplant, but they couldn’t find a match for him. Then they learned about an experimental gene therapy at Boston Children’s Hospital. It was attempting to treat children like Levi by replacing the gene responsible for destroying his immune system.

“I thought, this isn’t real,” Kala says. “There’s no way this could work.”

Nonetheless, the Lookses flew from their home in Michigan to Boston in May 2015. Days later, Levi got an infu- sion of the therapy into his veins. He has been a normal boy ever since—and he has even grown larger than his twin brother. Babies born with SCID typically didn’t survive past two years old. Now, a one- time treatment offers a cure for patients like Levi Looks.

Researchers have been chasing the dream of gene therapy for decades. The idea is elegant: use an engineered virus to deliver healthy copies of a gene into patients with defective versions. But until recently it had produced more disappointments than successes. The entire field was slowed in 1999 when an 18-year-old patient with a liver disease, Jesse Gelsinger, died in a gene-therapy experiment.

But now, crucial puzzles have been solved and gene therapies are on the verge of curing devastating genetic dis- orders. Two gene therapies for inherited diseases—Strimvelis for a form of SCID and Glybera for a disorder that makes fat build up in the bloodstream—have won regulatory approval in Europe. In the United States, Spark Therapeutics could be the first to market; it has a treatment

1960s The idea of gene therapy arises when scientists discover enzymes that can be used to cut DNA sequences and stitch them together in test tubes.

1970s Scientists experiment with using viruses to introduce new genes into animals.

1990 A four-year-old girl (pictured at lower right in 1992) is treated for SCID, a genetic disease that would have left her defenseless against infections. But other children with the disease will later develop leukemia from a different gene therapy.

1999 Jesse Gelsinger, 18, becomes the first patient to die in a clinical trial for gene therapy.

Gene-Therapy Time Line

C H

R IS

W A

R E

/ G

E T

T Y

; F O

X P

H O

T O

S /

G E

T T

Y ; C

O U

R T

E S

Y O

F T

H E

IM M

U N

E

D E

F IC

IE N

C Y

F O

U N

D A

T IO

N ; C

O U

R T

E S

Y O

F T

H E

G E

L S

IN G

E R

F A

M IL

Y

MA17_10_genetherapy2.0.indd 50 2/8/17 12:38 PM

51

TECHNOLOGYREVIEW.COM BREAKTHROUGH TECHNOLOGIES MIT TECHNOLOGY REVIEW

VOL. 120 | NO. 2

for a progressive form of blindness. Other gene therapies in development point to a cure for hemophilia and relief from an incapacitating skin disorder called epider- molysis bullosa.

Fixing rare diseases, impressive in its own right, could be just the start. Researchers are studying gene therapy in clinical trials for about 40 to 50 different diseases, says Maria-Grazia Roncarolo, a pediatrician and scientist at Stanford Uni- versity who led early gene-therapy experi- ments in Italy that laid the foundation for Strimvelis. That’s up from just a few con- ditions 10 years ago. And in addition to treating disorders caused by malfunctions in single genes, researchers are looking to engineer these therapies for more com- mon diseases, like Alzheimer’s, diabetes, heart failure, and cancer. Harvard geneti- cist George Church has said that someday,

everyone may be able to take gene therapy to combat the effects of aging.

Early gene therapies failed in part because of the delivery mechanism. In 1990, a four-year-old girl with a form of SCID was treated by scientists at the National Institutes of Health, who extracted white blood cells from her, inserted normal copies of her faulty gene into them, then injected her with the cor- rected cells. But patients later treated for a different type of SCID went on to develop leukemia. The new genetic material and the virus used to carry it into cells were delivered to the wrong part of the genome, which switched on cancer- causing genes in some patients. In Gelsinger’s case, the virus used to transport functioning genes into his cells made his immune system go into overdrive, leading to multiple organ failure and brain death.

Gene-therapy researchers have sur- mounted many of those early problems by using viruses that are more efficient at transporting new genetic material into cells.

But several challenges remain. While gene therapies have been developed for several relatively rare diseases, creating such treatments for more common dis- eases that have complex genetic causes will be far more difficult. In diseases like SCID and hemophilia, scientists know the precise genetic mutation that is to blame. But diseases like Alzheimer’s, diabetes, and heart failure involve multiple genes— and the same ones aren’t all involved in all people with those conditions.

Nonetheless, for Kala and Philip Looks, the success of gene therapy is already real. A treatment they had never heard of rid their child of a horrific disease.

2017 or 2018 A gene therapy for an inherited disease could be approved in the U.S. for the first time.

May 2016 European regulators approve Strimvelis, the second gene therapy for an inherited disease, to treat a type of SCID.

2007-2008 Patients with an inherited retinal disease called Leber’s congenital amaurosis appear to have improved vision after treatment with a gene therapy. However, years later, researchers will report in the New England Journal of Medicine that some patients’ eyesight has begun to wane.

2012 The European Medicines Agency approves the first gene therapy for an inherited disease. Called Glybera, the drug treats lipoprotein lipase deficiency, which causes fat to build up in the blood.

M IK

A E

L H

Ä G

G S

T R

Ö M

; C O

U R

T E

S Y

O F

U N

IQ U

R E

; M Y

S T

È R

E M

A R

T IN

/ W

IK IM

E D

IA ; E

R IC

P IE

R M

O N

T /

G E

T T

Y

MA17_10_genetherapy2.0.indd 51 2/8/17 12:38 PM

Copyright of MIT Technology Review is the property of MIT Technology Review and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.

BREAKTHROUGH TECHNOLOGIESTECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2

By experimenting, computers are figuring out how to do things that no programmer could teach them.

I nside a simple computer simulation, a group of self-driving cars are per- forming a crazy-looking maneuver on a four-lane virtual highway. Half are trying to move from the right-hand

lanes just as the other half try to merge from the left. It seems like just the sort of tricky thing that might flummox a robot vehicle, but they manage it with precision.

I’m watching the driving simulation at the biggest artificial-intelligence con- ference of the year, held in Barcelona this past December. What’s most amazing is that the software governing the cars’ behavior wasn’t programmed in the con- ventional sense at all. It learned how to merge, slickly and safely, simply by practic-

Reinforcement Learning

Breakthrough An approach to artificial intelligence that gets computers to learn like people, without explicit instruction.

Why It Matters Progress in self- driving cars and other forms of automation will slow dramatically unless machines can hone skills through experience.

B y W

il l

K n

ig h

t

Key Players - DeepMind - Mobileye - OpenAI - Google - Uber

Availability 1 to 2 years

S E

Y M

O U

R C

H W

A S

T

MA17_10_reinforcement_learning.indd 32 2/8/17 10:32 AM

33

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW

VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

Reinforcement Learning

MA17_10_reinforcement_learning.indd 33 2/8/17 10:32 AM

ing. During training, the control software performed the maneuver over and over, altering its instructions a little with each attempt. Most of the time the merging happened way too slowly and cars inter- fered with each other. But whenever the merge went smoothly, the system would learn to favor the behavior that led up to it.

This approach, known as reinforce- ment learning, is largely how AlphaGo, a computer developed by a subsidiary of Alphabet called DeepMind, mastered the impossibly complex board game Go and beat one of the best human players in the world in a high-profile match last year. Now reinforcement learning may soon inject greater intelligence into much more than games. In addition to improv- ing self-driving cars, the technology can get a robot to grasp objects it has never seen before, and it can figure out the opti- mal configuration for the equipment in a data center.

These images are from the Mobileye vision system for cars, which will benefit from reinforcement learning.

T h

e t

e c

h n

o l

o g

y

ca n g

et a r

ob ot

to gr

as p o

bje cts

it

ha s n

ev er

see n b

efo re.

Reinforcement learning copies a very simple principle from nature. The psy- chologist Edward Thorndike documented it more than 100 years ago. Thorndike placed cats inside boxes from which they could escape only by pressing a lever. After a considerable amount of pacing around and meowing, the animals would eventu- ally step on the lever by chance. After they learned to associate this behavior with the desired outcome, they eventually escaped with increasing speed.

Some of the very earliest artificial- intelligence researchers believed that this process might be usefully reproduced in machines. In 1951, Marvin Minsky, a stu- dent at Harvard who would become one of the founding fathers of AI as a professor at MIT, built a machine that used a simple form of reinforcement learning to mimic a rat learning to navigate a maze. Minsky’s Stochastic Neural Analogy Reinforcement Computer, or SNARC, consisted of dozens of tubes, motors, and clutches that simu- lated the behavior of 40 neurons and syn- apses. As a simulated rat made its way out of a virtual maze, the strength of some syn- aptic connections would increase, thereby reinforcing the underlying behavior.

There were few successes over the next few decades. In 1992, Gerald Tesauro, a researcher at IBM, demonstrated a pro- gram that used the technique to play backgammon. It became skilled enough to rival the best human players, a landmark achievement in AI. But reinforcement learning proved difficult to scale to more complex problems. “People thought it was a cool idea that didn’t really work,” says David Silver, a researcher at DeepMind in the U.K. and a leading proponent of reinforcement learning today.

That view changed dramatically in March 2016, however. That’s when AlphaGo, a program trained using rein- forcement learning, destroyed one of the best Go players of all time, South Korea’s Lee Sedol. The feat was astonishing, because it is virtually impossible to build

a good Go-playing program with conven- tional programming. Not only is the game extremely complex, but even accomplished Go players may struggle to say why certain moves are good or bad, so the principles of the game are difficult to write into code. Most AI researchers had expected that it would take a decade for a computer to play the game as well as an expert human.

Jostling for position Silver, a mild-mannered Brit who became fascinated with artificial intelligence as an undergraduate at the University of Cam- bridge, explains why reinforcement learn- C

O U

R T

E S

Y O

F M

O B

IL E

Y E

34

TECHNOLOGYREVIEW.COM MIT TECHNOLOGY REVIEW VOL. 120 | NO. 2 BREAKTHROUGH TECHNOLOGIES

MA17_10_reinforcement_learning.indd 34 2/8/17 10:32 AM

ing has recently become so formidable. He says that the key is combining it with deep learning, a technique that involves using a very large simulated neural net- work to recognize patterns in data (see “10 Breakthrough Technologies 2013: Deep Learning”).

Reinforcement learning works because researchers figured out how to get a com- puter to calculate the value that should be assigned to, say, each right or wrong turn that a rat might make on its way out of its maze. Each value is stored in a large table, and the computer updates all these values as it learns. For large and compli- cated tasks, this becomes computationally impractical. In recent years, however, deep learning has proved an extremely efficient way to recognize patterns in data, whether the data refers to the turns in a maze, the positions on a Go board, or the pixels shown on screen during a computer game.

In fact, it was in games that Deep- Mind made its name. In 2013 it published details of a program capable of learning to play various Atari video games at a super- human level, leading Google to acquire the company for more than $500 million in 2014. These and other feats have in turn inspired other researchers and com- panies to turn to reinforcement learning. A number of industrial-robot makers are testing the approach as a way to train their machines to perform new tasks without manual programming. And researchers at Google, also an Alphabet subsidiary, worked with DeepMind to use deep rein- forcement learning to make its data cen- ters more energy efficient. It is difficult to figure out how all the elements in a data center will affect energy usage, but a reinforcement- learning algorithm can learn from collated data and experiment in simulation to suggest, say, how and when to operate the cooling systems.

But the setting where you will proba- bly most notice this software’s remarkably humanlike behavior is in self-driving cars. Today’s driverless vehicles often falter in

Reinforcement learning led to AlphaGo’s stunning victory over a human Go champion last year.

complex situations that involve interact- ing with human drivers, such as traffic circles or four-way stops. If we don’t want them to take unnecessary risks, or to clog the roads by being overly hesitant, they will need to acquire more nuanced driving skills, like jostling for position in a crowd of other cars.

The highway merging software was demoed in Barcelona by Mobileye, an Israeli automotive company that makes vehicle safety systems used by dozens of carmakers, including Tesla Motors (see “50 Smartest Companies 2016”). After screen- ing the merging clip, Shai Shalev-Shwartz, Mobileye’s vice president for technology, shows some of the challenges self-driv- ing cars will face: a bustling roundabout in Jerusalem; a frenetic intersection in Paris; and a hellishly chaotic scene from a road in India. “If a self-driving car follows the law precisely, then during rush hour I might wait in a merge situation for an hour,” Shalev- Shwartz says.

Mobileye plans to test the software on a fleet of vehicles in collaboration with BMW and Intel later this year. Both Google and Uber say they are also test- ing reinforcement learning for their self- driving vehicles.

Reinforcement learning is being applied in a growing number of areas, says Emma Brunskill, an assistant professor at Stanford University who specializes in the approach. But she says it is well suited to automated driving because it enables “good sequences of decisions.” Progress would proceed much more slowly if pro- grammers had to encode all such deci- sions into cars in advance.

But there are challenges to overcome, too. Andrew Ng, chief scientist at the Chinese company Baidu, warns that the approach requires a huge amount of data, and that many of its successes have come when a computer could practice relent- lessly in simulations. Indeed, research- ers are still figuring out just how to make reinforcement learning work in complex situations in which there is more than one objective. Mobileye has had to tweak its protocols so a self-driving car that is adept at avoiding accidents won’t be more likely to cause one for someone else.

When you watch the outlandish merg- ing demo, it looks as though the company has succeeded, at least so far. But later this year, perhaps on a highway near you, reinforcement learning will get its most dramatic and important tests to date. C

O U

R T

E S

Y O

F D

E E

P M

IN D

35

BREAKTHROUGH TECHNOLOGIES

MA17_10_reinforcement_learning.indd 35 2/8/17 10:32 AM

Copyright of MIT Technology Review is the property of MIT Technology Review and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.

FORBES INNOVATION ENTERPRISE TECH

Forrester's Top 15 Emerging

Technologies To Watch, 2017

- 2021

Louis Columbus Former Contributor

Sep 25, 2016, 02:07pm EDT

This article is more than 6 years old.

Customers’ expectations of excellent experiences are accelerating

driven by the real-time responsiveness of the latest social, mobile,

analytics and cloud applications. This is especially the case for B2B

buyers, who expect the same contextually intelligent, data-rich

interactions across any channel they choose to purchase through,

anytime.

Salesforce's Einstein is an example of how Artificial

Intelligence (AI) has the potential to revolutionize CRM and

customer-facing systems.

Augmented reality has immediate implications for

streamlining the Configure, Price, Quote (CPQ) and Quote-

To-Cash (QTC) strategies of many companies.

74% of B2B buyers research half or more of their work

purchases online before buying according to a recent

Forrester Study.

74% of B2B buyers research half or more of their work purchases

online before buying according to a recent Forrester Study. 30%

make half or more of their work purchases online today, and 56%

expect to make half or more of their work purchases online in 3

years. Forrester’s recently released top 15 emerging technologies to

watch, 2017 – 2021 reflect the accelerating expectations of

customers who rely on digital channels the majority of the time to

evaluate, buy and get support for products and services. B2B buyers

are collaborators in the creation and continual refining of these

technologies as their buying and service requirements force the

issues scalability, speed, and service quickly.

Defining 15 Technologies That Win And Keep Customers

Forrester’s methodology of defining the top 15 emerging

technologies is based on an assessment of which have the greatest

potential to win, serve and retain customers. Looking at which

technologies can make the greatest contribution to business growth

in the next five years, Forrester didn’t include blockchain or 3D

printing, however. Instead, the focus is on technologies that

support systems of engagement, systems of insight and supporting

technologies that enable greater speed, scalability and real-time

integration fo diverse systems, apps, and platforms. The following

graphic provides an overview of the 15 emerging technologies to

watch:

Forbes Daily: Our best stories, exclusive reporting and Forbes perspectives on the day’s top news, plus the inside scoop on the world's most important entrepreneurs.

Email address Sign Up

You may opt out any time. By signing up for this newsletter, you agree to the Terms and Conditions and Privacy Policy

Forrester sees the Internet of Things (IoT) creating

more valuable customer interactions and improving

the customer experience. Making the most use of

actuator, sensor, and network-captured data through the use

of advanced analytics techniques including Hadoop and R,

organizations will better be able to understand and predict

customer buying cycles. The most immediate benefit for B2B

buyers will be in the supply chain they rely on, as IoT data

will help alleviate stock-outs and inventory allocation

problems

Intelligent agents are already accelerating Artificial

Intelligence (AI) and machine learning in CRM and

customer-facing systems. Salesforce will demonstrate

and heavily promote Salesforce Einstein at this year's

Dreamforce 2016 happening next month, October 4 - 7th in

San Francisco. It's a great example of how AI has the

potential to revolutionize CRM and customer-facing

systems. Expect to see several announcements out of the

Salesforce partner community regarding intelligent agents,

AI and machine learning, further making this area one of the

main themes of Dreamforce 2016.

Augmented reality has immediate implications for

streamlining the Configure, Price, Quote (CPQ) and

Quote-To-Cash (QTC) strategies of many

companies. One of the most valuable lessons learned from

being on hundreds of sales calls where CPQ systems were

discussed is the need manufacturers have to be able to

visualize the complete, finished product they are designing.

The greater the product complexity the slower the sales

cycles go. With augmented reality and 3D visualization,

being able to walk through a proposed oil refinery, around a

complex engine, and through a proposed jet would be

invaluable.

Cloud native application platforms will continue to

enable rapid advances in enterprise accounting

Source: Forrester's Top Emerging Technologies To Watch: 2017-

2021

Louis Columbus

I am currently serving as Principal, IQMS, part of Dassault Systèmes. Previous

positions include product management at Ingram Cloud, product... Read More

software. Creating more scalable, secure and agile general

ledgers is going to be increasingly possible with the

continued growth of cloud-native applications

platforms. SelectHub's latest update on Enterprise

Accounting Software Recommendations provides an

overview of where this area of enterprise software is today

and where it is going.

ADVERTISEMENT

Editorial Standards Reprints & Permissions