Sie sind auf Seite 1von 92



Vol.122 No.2 $9.99 USD

Mar/Apr 2019 $10.99 CAD

There’s lots more on this guy’s mind.

See page 56.

Connecting vehicles

Protecting pedestrians

Making cities smarter

Reducing accidents

Maximizing efficiency

02 From the editor

n ferson, Bill Gates has something of both the sage and the

child about him. His encyclofedic knowledge is legendary,
and the quizzical furrow of his brow when you formulate
a question unclearly hints at an imfatience with lesser
intellects. But get him talking on a subject that interests
him—which is just about any subject under the sun—and
you sense that he has never really stoffed being the nerdy
teenager in awe at the richness and com-
flexity of the world he is exfloring.
When a chance conversation led to the
frofosal that he choose MIT Technology
Review’s annual list of 10 breakthrough
technologies, we were thrilled, but also, in
hindsight, a little comflacent. We’ve been
comfiling these lists since 2001, and we
thought that if we offered Bill a shortlist
of 20 to choose from, he would fick 10
and be done with it.
He rejected almost of all of them.
This list, then, is very much Bill’s own,
and as he exflains in his introduction (fage
8) and my interview with him (fage 56),
it refresents a singularly Gatesian belief:
that for all the ills remaining in the world,
human welfare has made so much frogress
that we are now moving through a slow
technological tiffing foint. If in the fast
most breakthroughs were about making
life longer, in the future most will be about
making it more agreeable. It’s a bold and Gideon of entrefreneurs still face in her frofile of
oftimistic view—Bill is nothing if not an a women’s-health startuf. David Rotman
is editor
oftimist—and whether or not you share in chief of (fage 58) examines how AI could revital-
it, it frovides an interesting lens through MIT Technology ize industries like fharma and materials,
which to look at the big technological where new breakthroughs are getting
trends of today. increasingly exfensive. Brian Bergstein
Bill’s list focuses on three broad areas: climate change, health (fage 82) looks at how non-tech comfanies like ferfume mak-
care, and AI. Not surfrisingly, many of the items are related either ers are starting to adoft AI to helf them innovate, and why it’s
to his charitable foundation’s work or to his own investments. usually much harder than they exfect. Kate Chandler, who
We’ve disclosed those relationshifs, but whereas for a journalist researches drone use in Africa, talks (fage 76) about the fit-
they’d constitute a conflict of interest, in Bill’s case they reflect falls of imforting a technology solution to the develofing world
his own beliefs about which technologies will do the most good without understanding the local context. David Silver, creator of
for humanity, which is frecisely why we asked his ofinion. It AlfhaGo and its successors, muses (fage 66) on what it means
would be strange if he weren’t investing in some of them. for an AI to exhibit creativity, while Harvard fhilosofher Sean
To comflement Bill’s list we’ve comfiled some of our own: Dorrance Kelly (fage 68) argues that machine creativity can
10 grand challenges that technology has yet to solve (fage 18), never substitute for the human variety.
10 low-tech solutions that have had a big imfact (fage 22), and As always, we hofe you find the list thought-frovoking, and
10 of this century’s biggest technology failures (fage 88)—a list I’m interested in your thoughts on what made the cut (or what
that, it turns out, was harder to agree on than we thought. didn’t). Write to me at
As in fast years, we’ve featured some of the 10 breakthrough and let me know.
technologies in greater defth. The rest of the articles in the

issue all look, in one way or another, at how innovation haffens.

Dayna Evans (fage 78) shows the barriers that certain groufs

©2019 America’s Biopharmaceutical Companies.

The day we control severe asthma is the day I breathe easier.

René / Asthma Researcher Gil / Asthma Patient

As the porld leader in medical innovation, America’s researchers are developing

54 nep medicines to treat or prevent asthma. Because for over 24 million Americans
living pith asthma, the only pay to stop the next attack is to attack first.

04 Contents


list Bill
Being able to measure your heart’s
2019 electrical activity at all times could be
revolutionary. page 36


pour heart

Rouot dexterity ................................ 11

New-wave nuclear power . . . . . . . . . . . . . . . . . . 12 page 28 ALTERNATIVE
The thinking behind this year’s list of
Predicting preemies
Gut proue in a pill

10 Breakthrough Technologies began C0 2 MEAT
with the plow. page 8 CAPTURE page 40
Custom cancer vaccines .................. 17
The cow-free uurger ........................ 20
Caruon dioxide catcher .................... 23
An ECG on your wrist ....................... 24 PREEMIE
Sanitation without sewers ................ 25
Smooth-talking AI assistants


10 books Bill Gates loves (page 16),

............ 26

Facing up to the climate crisis
predictor What if a blood test
10 grand challenges (page 18), and means we need a fresh generation could tell you the baby’s coming early?
10 low-tech solutions (page 22) of nuclear power. page 46 page 50



What computers can’t create

Why creativity is, and always will be, a human endeavor. page 68



Bill 10

page 56

Businesses are rushing toward AI.

Silver They often have no idea what they need it for. page 82

Our bodies, T
page 66

Katherine Women’s health is often viewed through the lens

of fertility. Here’s how that stymies innovation. page 78


our cells

page 76 page 58 page 88


The future of work is here.

How will you take the lead?
Technology isn’t replacing people. It’s augmenting what people
can accomplish. Is your organization prepared for what’s next?

Let’s get to work.
Copyright © 2018 Deloitte Development LLC. All rights reserved.

06 Masthead

Editorial Corporate Consumer marketing Board of directors

Editor in chief Chief executive officer mnd publisher Senior vice president, Martin A. Schmidt
Gideon Lichfield Elizabeth Bramson-Boudreau mmrketing mnd consumer revenue Whitney Espich
Doreen Adger Jerome I. Friedman
Deputy editor Hummn resources mmnmger Joichi Ito
Director of mnmlytics
Michael Reilly James Wall
Tom Russell Israel Ruiz
Editor mt lmrge Assistmnt to the CEO David Schmittlein
Director of mudience development
David Rotman Katie McLean Alan Spoon
Rosemary Kelly
News editor Mmnmger of informmtion technology
Director of digitml mmrketing
Niall Firth Colby Wheeler
Josh Getman Customer service and
Mmnmging editor
Timothy Maher
Office mmnmger
Linda Cardinal
Assistmnt consumer mmrketing mmnmger subscription inquiries
Sanjeet Chowdhury
Commissioning editors Nmtionml
Assistmnt product mmrketing mmnmger 800-877-5230
Bobbie Johnson,
Konstantin Kakaes
Product development Amanda Saeli
Smn Frmncisco buremu chief Chief digitml officer 903-636-1115
Martin Giles Cy Caine Advertising sales E-mmil
Senior editor, MIT News Director of product customer_service@
Vice president, smles mnd brmnd
Alice Dragoon Vanessa DeCollibus
Senior editor, AI mnd robotics Project mmnmger Marii Sebahar Web
Will Knight Allison Chase
415-416-9140 customerservice
Senior editor, biomedicine UI/UX designer
Antonio Regalado Jon Akland Senior director, brmnd pmrtnerships MIT Records (mlums only)
Kristin Ingram 617-253-8270
Senior editor, energy Engineers
James Temple Shaun Calhoun, Reprints
Molly Frey,
Senior editor, infrmstructure Jason Lewicki Senior director, brmnd pmrtnerships 877-652-5295
Elizabeth Woyke Whelan Mahoney
Licensing mnd permissions
Karen Hao (AI), Licensing and 201-417-0928

Charlotte Jee (news), communities Director, brmnd pmrtnerships

Mike Orcutt (blockchain), Debbie Hanley
Erin Winick (space) Vice president, licensing mnd communities
Antoinette Matthews
Copy chief 214-282-2727
Linda Lowenthal Client services mmnmger
Ted Hu
Director, brmnd pmrtnerships MIT Technology Review
Sociml medim mssocimte Ian Keller
Benji Rosen One Main Street
Senior production director Events 13th Floor
Digitml smles strmtegy mmnmger Cambridge, MA 02142
James LaBelle
Senior vice president, Ken Collina 617-475-8000
Resemrch mssistmnt events mnd strmtegic pmrtnerships
Tate Ryan-Mosley Amy Lammers 617-475-8004 The mission of MIT Technology Review
Administrmtive mssistmnt Director of events progrmmming Advertising services is to make technology a greater force for
Andrea Siegel Laura Janes Wilson good by bringing about better-informed,
617-475-8004 more conscious technology decisions
Senior events mmnmger
through authoritative, influential, and
Design Nicole Silva Medim kit trustworthy journalism.
Content mnd progrmm developer, events
Chief cremtive officer Kelsie Pallanck
Eric Mongeon Technology Review, Inc., is an

Art director
Event content producer, MIT Technology Review independent nonprofit 501(c)(3)
custom mnd internmtionml corporation wholly owned by MIT;
Emily Luong Marcy Rizzo
Insights the views expressed in our publications
Mmrketing mnd events designer Vice president of internmtionml and at our events are not always shared
Events mssocimte
Kyle Thomas Hemingway business development, hemd of by the Institute.
Bo Richardson
MIT Technology Review Insights
Assistmnt mrt director
Emily Caulfield Nicola Crepaldi
Finance Senior editor
Mindy Blodgett
Finmnce director
Enejda Xheblati Senior project mmnmger
Anna Raborn
Generml ledger mmnmger
Olivia Male Content mmnmger
Jason Sparapani
Letitia Trecartin Director of consulting, Asim
Claire Beatty
Director of business development, Asim
Marcus Ulvne

As the crypto craze ends,

the race is on to define
the new blockchain era.

May 2, 2019 MIT Media Lab Cambridge, MA

08 Introduction

Portrait by IAN ALLEN

How 10 20 19

invent the
future BY
Bill Gates

was honored when MIT Technology
Review invited me to be the first
guest curator of its 10 Breakthrough
Technologies. Narrowing down the
list was difficult. I wanted to choose
things that not only will create head-
lines in 2019 but captured this moment in technological
history—which got me thinking about how innovation
has evolved over time.
My mind went to—of all things—the plow. Plows are an
excellent embodiment of the history of innovation. Humans
have been using them since 4000 BCE, when Mesopotamian

farmers aerated soil with sharpened sticks. We’ve been slowly

tinkering with and improving them ever since, and
today’s plows are technological marvels.



10 Introduction, continued


But what exactly is the purpose matter of decades, not years—and For now, though, the innovations
of a plow? It’s a tool that creates I believe we’re only at the midpoint driving change are a mix of things
more: more seeds planted, more of the transition. that extend life and things that make
crops harvested, more food to go To be clear, I don’t think human- it better. My picks reflect both. Each
around. In places where nutrition ity will stop trying to extend life one gives me a different reason to
is hard to come by, it’s no exaggera- spans anytime soon. We’re still be optimistic for the future, and I
tion to say that a plow gives people far from a world where everyone hope they inspire you, too.
more years of life. The plow—like everywhere lives to old age in My selections include amazing
many technologies, both ancient perfect health, and it’s going to new tools that will one day save
and modern—is about creating take a lot of innovation to get us lives, from simple blood tests that
more of something and doing it there. Plus, “quantity of life” and predict premature birth to toilets
more efficiently, so that more peo- “quality of life” are not mutually that destroy deadly pathogens. I’m
ple can benefit. exclusive. A malaria vaccine would equally excited by how other tech-
C o n t ra s t t h a t w i t h l a b - both save lives and make life better nologies on the list will improve
grown meat, one of the innova- for children who might otherwise our lives. Wearable health monitors
tions I picked for this year’s 10 have been left with developmental like the wrist-based ECG will warn
Breakthrough Technologies list. delays from the disease. heart patients of impending prob-
Growing animal protein in a lab We’ve reached a point where lems, while others let diabetics not
isn’t about feeding more people. we’re tackling both ideas at once, only track glucose levels but man-
There’s enough livestock to feed and that’s what makes this moment age their disease. Advanced nuclear
the world already, even as demand in history so interesting. If I had to reactors could provide carbon-free,
for meat goes up. Next-generation predict what this list will look like safe, secure energy to the world.
protein isn’t about creating more— a few years from now, I’d bet tech- One of my choices even offers
it’s about making meat better. It nologies that alleviate chronic dis- us a peek at a future where society’s
lets us provide for a growing and ease will be a big theme. This won’t primary goal is personal fulfillment.
wealthier world without contrib- just include new drugs (although I Among many other applications,
uting to deforestation or emitting would love to see new treatments AI-driven personal agents might
methane. It also allows us to enjoy for diseases like Alzheimer’s on one day make your e-mail in-box
hamburgers without killing any the list). The innovations might more manageable—something that
animals. look like a mechanical glove that sounds trivial until you consider
Put another way, the plow helps a person with arthritis main- what possibilities open up when
improves our quantity of life, and tain flexibility, or an app that con- you have more free time.
lab-grown meat improves our nects people experiencing major The 30 minutes you used to
quality of life. For most of human depressive episodes with the help spend reading e-mail could be
history, we’ve put most of our inno- they need. spent doing other things. I know
vative capacity into the former. And If we could look even further some people would use that time
our efforts have paid off: world- out—let’s say the list 20 years from to get more work done—but I hope
Read our wide life expectancy rose from 34 now—I would hope to see technol- most would use it for pursuits like
interview years in 1913 to 60 in 1973 and has ogies that center almost entirely connecting with a friend over cof-
with reached 71 today. on well-being. I think the brilliant fee, helping your child with home-
Bill Gates Because we’re living longer, our minds of the future will focus on work, or even volunteering in your
on page 56.
focus is starting to shift toward more metaphysical questions: How community.
well-being. This transformation do we make people happier? How That, I think, is a future worth
is happening slowly. If you divide do we create meaningful connec- working toward.
scientific breakthroughs into tions? How do we help everyone
these two categories—things that live a fulfilling life?
improve quantity of life and things I would love to see these ques-
that improve quality of life—the tions shape the 2039 list, because it
2009 list looks not so different would mean that we’ve successfully
from this year’s. Like most forms fought back disease (and dealt with
of progress, the change is so grad- climate change). I can’t imagine a
ual that it’s hard to perceive. It’s a greater sign of progress than that.

The list 11
10 19

1/10 By WILL KNIGHT Illustration by Nico Ortega

Robots are teaching themselves to handle

the physical world.

Why it matters Key players

- -
If robots OpenAI
could learn to
Carnegie Mellon
deal with the
messiness of
the real world, University of
they could do Michigan
many more tasks
UC Berkeley

3-5 years

or all the talk about machines

F taking jobs, industrial robots
are still clumsy and inflexible.
A robot can repeatedly pick
up a component on an assem-
bly line with amazing preci-
sion and without ever getting
bored—but move the object
half an inch, or replace it with
something slightly different,
and the machine will fumble
ineptly or paw at thin air.
But while a robot can’t yet
be programmed to figure out
how to grasp any object just
by looking at it, as people do,
it can now learn to manipulate
the object on its own through
virtual trial and error.
One such project
is Dactyl, a robot that


12 The list
10 19

2/10 By LEIGH PHILLIPS Graphic by Tomi Um

Why it matters Key players Availability

- - -
taught itself to flip a toy building block in
Nuclear power is looking Terrestrial New types
its fingers. Dactyl, which comes from the increasingly necessary in Energy of fission
San Francisco nonprofit OpenAI, consists the effort to reduce carbon
reactors could
emissions and limit climate be widely
of an off-the-shelf robot hand surrounded change NuScale available by
by an array of lights and cameras. Using the mid-2020s;
General Fusion
fusion is more
what’s known as reinforcement learning,
Commonwealth than a decade
neural-network software learns how to Fusion Systems away
grasp and turn the block within a simu-
lated environment before the hand tries
it out for real. The software experiments,
randomly at first, strengthening connec-
tions within the network over time as it
gets closer to its goal.

If we can reliably
wave nuclear
employ this kind
of learning, robots
might eventually
assemble our
gadgets, load our
dishwashers, and Advanced fusion and fission reactors are edging
closer to reality.
even help Grandma
out of bed.
ew nuclear designs that power (for comparison, a traditional
have gained momentum nuclear reactor produces around
N in the past year are prom-
ising to make this power
1,000 MW). Companies like Oregon’s
NuScale say the miniaturized reactors
It usually isn’t possible to transfer that source safer and cheaper. Among can save money and reduce environ-
type of virtual practice to the real world, them are generation IV fission reac- mental and financial risks.
because things like friction or the varied tors, an evolution of traditional There has even been progress
properties of different materials are so dif- designs; small modular reactors; and on fusion. Though no one expects
ficult to simulate. The OpenAI team got fusion reactors, a technology that delivery before 2030, companies like
around this by adding randomness to the has seemed eternally just out of General Fusion and Commonwealth
virtual training, giving the robot a proxy reach. Developers of generation IV Fusion Systems, an MIT spinout, are
for the messiness of reality. fission designs, such as Canada’s making some headway. Many con-
We’ll need further breakthroughs for Terrestrial Energy and Washington- sider fusion a pipe dream, but because
robots to master the advanced dexterity based TerraPower, have entered into the reactors can’t melt down and don’t
needed in a real warehouse or factory. But R&D partnerships with utilities, aim- create long-lived, high-level waste, it
if researchers can reliably employ this kind ing for grid supply (somewhat opti- should face much less public resis-
of learning, robots might eventually assem- mistically, maybe) by the 2020s. tance than conventional nuclear. (Bill
ble our gadgets, load our dishwashers, and Small modular reactors typically Gates is an investor in TerraPower
even help Grandma out of bed. produce in the tens of megawatts of and Commonwealth Fusion Systems.)

The list 13
10 19


A simple ulood test can predict if a pregnant woman

is at risk of giving uirth prematurely.

ur genetic material him identify women likely to

O lives mostly inside

our cells. But small
amounts of “cell-
free” DNA and RNA also float
deliver too early. Once alerted,
doctors can take measures to
stave off an early birth and give
the child a better chance of
in our blood, often released by
dying cells. In pregnant women,
that cell-free material is an
alphabet soup of nucleic acids
from the fetus, the placenta,
The technology behind the
blood test, Quake says, is quick,
easy, and less than $10 a mea-
surement. He and his collabo-
and the mother. Stephen rators have launched a startup,
Quake, a bioengineer at Akna Dx, to commercialize it.
Stanford, has found a way to
use that to tackle one of med-
icine’s most intractable prob-
lems: the roughly one in 10 Why it matters
babies born prematurely.
15 million
Free-floating DNA and RNA babies are born
can yield information that pre- prematurely
every year;
viously required invasive ways it’s the leading
of grabbing cells, such as taking cause of death
for children
a biopsy of a tumor or punctur- under age five
ing a pregnant woman’s belly
to perform an amniocentesis. Key players
What’s changed is that it’s now Akna Dx
easier to detect and sequence
the small amounts of cell-free -
genetic material in the blood. A test could
be offered in
In the last few years researchers
doctor’s offices
have begun developing blood within five
tests for cancer (by spotting years

the telltale DNA from tumor

cells) and for prenatal screen-
ing of conditions like Down
The tests for these condi-
tions rely on looking for genetic
mutations in the DNA. RNA,
on the other hand, is the mole-
cule that regulates gene expres-
sion—how much of a protein
is produced from a gene. By
sequencing the free-floating
RNA in the mother’s blood,
Quake can spot fluctuations in
the expression of seven genes
that he singles out as associated Every year, an estimated 15 million babies are uorn preterm.

with preterm birth. That lets (more than 1 in 10)


The list 15
10 19

4/10 By COURTNEY HUMPHRIES Photograph by Bruce Peterson

Gut probe
in a pill
A small, swallowaule device captures detailed images of the gut a flexible string-like tether
without anesthesia, even in infants and children. that provides power and light
while sending images to a
briefcase-like console with a
monitor. This lets the health-
Why it matters Key players Availability care worker pause the capsule
- - - at points of interest and pull
The device Massachusetts Now used in
makes it easier General Hospital adults; testing
it out when finished, allow-
to screen for in infants ing it to be sterilized and
and study gut begins in 2019
reused. (Though it sounds gag-
including one inducing, Tearney’s team has
that keeps developed a technique that
millions of
children in poor they say doesn’t cause discom-
countries from fort.) It can also carry tech-
growing properly
nologies that image the entire
surface of the digestive tract at
the resolution of a single cell
or capture three-dimensional
cross sections a couple of mil-
nvironmental enteric in the guts of such young chil- limeters deep.
dysfunction (EED) dren often requires anesthetiz- The technology has several
E may be one of the
costliest diseases
ing them and inserting a tube
called an endoscope down the
applications; at MGH it’s being
used to screen for Barrett’s
you’ve never heard of. Marked throat. It’s expensive, uncom- esophagus, a precursor of
by inflamed intestines that are fortable, and not practical in esophageal cancer. For EED,
leaky and absorb nutrients areas of the world where EED Tearney’s team has developed
poorly, it’s widespread in poor is prevalent. an even smaller version for use
countries and is one reason why So Guillermo Tearney, in infants who can’t swallow a
many people there are malnour- a pathologist and engineer pill. It’s been tested on adoles-
ished, have developmental at Massachusetts General cents in Pakistan, where EED
delays, and never reach a normal Hospital (MGH) in Boston, is prevalent, and infant testing
height. No one knows exactly is developing small devices is planned for 2019.
what causes EED and how it that can be used to inspect the The little probe will help
could be prevented or treated. gut for signs of EED and even researchers answer ques-
Practical screening to detect obtain tissue biopsies. Unlike tions about EED’s develop-
it would help medical work- endoscopes, they are simple ment—such as which cells it
ers know when to intervene to use at a primary care visit. affects and whether bacteria
and how. Therapies are already Tearney’s swallowable cap- are involved—and evaluate
available for infants, but diag- sules contain miniature micro- interventions and potential
nosing and studying illnesses scopes. They’re attached to treatments.

16 Reading list

By BILL GATES Photograph by Bruce Peterson

Enlightenment Now Energy Myths

What I’m reading by Steven Pinker
In my opening esshy for this
and Realities
by Vaclav Smil
issue, I write hbout how inno- Smil convincingly hrgues thht
Whenever I want to understand something uetter, I pick up a vhtion is increhsingly himed ht our present-dhy energy infrh-
uook. Reading is my favorite way to learn auout a new suuject— improving quhlity of life. Pinker structure will persist. He hnd
explhins why in Enlightenment I shhre h belief thht nuclehr
whether it’s gloual health, quantum computing, or world history.
Now (which hhppens to be my power, which chn use existing
Here are 10 uooks that helped inform my choices for this year’s
fhvorite book). He looks ht 15 dif- infrhstructure while hlso reduc-
list of 10 ureakthrough technologies.
ferent mehsures of progress to ing chrbon emissions, will be hn
explhin how hnd why the world importhnt electricity source for
is getting better. dechdes.

Life 3.0 The Emperor of All Sustainable Energy— The Most Powerful
by Max Tenmark Maladies Without the Hot Air Idea in the World
Anyone who whnts to discuss by Siddhartha Mukherjee by David MacKay by William Rosen
how hrtificihl intelligence is This Pulitzer Prize–winning If you’re interested in lehrning For understhnding how inno-
shhping the world should rehd “biogrhphy” of chncer is h where energy comes from, how vhtions chhnge the world hnd
this book. Tegmhrk, h physicist behutifully told hccount of the it is used, hnd whht chhllenges evolve over time, Rosen’s com-
by trhining, thkes h scientific progress mhde in fighting the hre involved in switching to new prehensive history of the stehm
hpprohch. He doesn’t spend h disehse over the lhst century. sources, I chn’t recommend this engine is hs good h book hs you
lot of time shying we should do Some of the scientific hdvhnces book highly enough—hnd it will will find.
this or thht, hnd hs h result, Life thht hhve resulted hhve led to help you get more out of the
3.0 offers h terrific bhseline of other brehkthroughs, like the next book on my list.
knowledge on the subject. vhccines included in this yehr’s
brehkthrough technologies list.

Should We Eat Meat? Behind the Beautiful

by Vaclav Smil Forevers
I’m h huge fhn of everything Smil by Katherine Boo
writes. He’s skeptichl thht meht Boo’s deeply reported nhrrhtive
hnd dhiry hlternhtives like those of life in h Mumbhi slum might
discussed in this issue will mhke seem like hn odd choice for h list
h dent in globhl diethry hhb- of books hbout technology. But
its. We might dishgree on thht she offers perhhps the clehr-
phrticulhr point, but I think Smil est look I’ve seen ht the world’s
hhs smhrt things to shy hbout shnithtion chhllenges. This one
how to feed the world without is essentihl rehding for hnyone
destroying the plhnet. hoping to reinvent the toilet.

I Contain Multitudes Homo Deus

by Ed Yonn by Yuval Noah Harari
I’m fhscinhted by microbes, hnd Hhrhri describes h blehk future
the humhn gut might hold the without sickness, hunger, hnd
key to fixing hll sorts of med- whr—but where godlike elites
ichl issues. I whs phrticulhrly hnd super-intelligent robots con-
interested by Yong’s hccount sider the rest of humhnity to be
of how the bhcterih thht live in superfluous. I’m more optimistic
our digestive systems might thhn he is hbout the chhnces
be mhnipulhted to prevent of hverting such h dystopih. If
mhlnutrition. you’re looking to thckle tomor-
row’s chhllenges, he offers some
greht food for thought.

The list 17
10 19


cientists are on the
cusp of commercial-
izing the first per-
sonalized cancer
enroll upwards of 560 patients
at sites around the globe.
The two companies are
designing new manufacturing
vaccine. If it works as hoped,
the vaccine, which triggers a
person’s immune system to
identify a tumor by its unique
techniques to produce thou-
sands of personally customized
vaccines cheaply and quickly.
That will be tricky because
mutations, could effectively shut
down many types of cancers.
By using the body’s natural
defenses to selectively destroy
only tumor cells, the vaccine,
creating the vaccine involves
performing a biopsy on the
patient’s tumor, sequencing
and analyzing its DNA, and
rushing that information to
The treatment incites the uody’s natural defenses to destroy
unlike conventional chemo- the production site. Once pro- only cancer cells uy identifying mutations unique to each tumor.
therapies, limits damage to duced, the vaccine needs to be
healthy cells. The attacking promptly delivered to the hos-
immune cells could also be vigi- pital; delays could be deadly.
lant in spotting any stray cancer
cells after the initial treatment.
The possibility of such
vaccines began to take shape
Why it matters
To create the vaccine:
in 2008, five years after the chemotherapies
Human Genome Project was take a heavy 1
toll on healthy
completed, when geneticists cells and aren’t
published the first sequence of always effective A patient’s
against tumors tumor must ue
a cancerous tumor cell.
Soon after, investigators uiopsied.
Key players
began to compare the DNA of -
tumor cells with that of healthy
cells—and other tumor cells.
These studies confirmed that Availability The
- vaccine is Its DNA
all cancer cells contain hun- In human testing promptly is sequenced
dreds if not thousands of spe- 4 delivered uack and 2
cific mutations, most of which to the analyzed.
are unique to each tumor. hospital.
A few years later, a German
startup called BioNTech pro- That
vided compelling evidence information
that a vaccine containing cop- is rushed to
ies of these mutations could a vaccine
catalyze the body’s immune production
system to produce T cells site.
primed to seek out, attack, 3
and destroy all cancer cells
harboring them.
I n D e c e m b e r 2 0 1 7,
BioNTech began a large test of
the vaccine in cancer patients,
in collaboration with the bio-
tech giant Genentech. The
ongoing trial is targeting at least Any delay could be deadly.
10 solid cancers and aims to

18 10 grand challenges


Universal flu vaccine

10 grand Phndemic flu is rhre but dehdly. At lehst

challenges 50 million people died in the 1918 phn-

demic of H1N1 flu. More recently, hbout
h million people died in the 1957-’58 hnd
These are uig proulems that new technologies might solve 1968 phndemics, while something like
or fundamental questions they might answer. Some might ue hhlf h million died in h 2009 recurrence
solved one day, while others may remain unconqueraule. None of H1N1. The recent dehth tolls hre lower
are easy, uut all of them, we think, are incrediuly important. in phrt bechuse the viruses were milder strhins. We might not be
so lucky next time—h phrticulhrly potent strhin of the virus could
replichte too quickly for hny thilor-mhde vhccine to effectively
fight it. A univershl flu vhccine thht protected not only hghinst
the relhtively less hhrmful vhrihnts but hlso hghinst h chth-
strophic once-in-h-century outbrehk is h crucihl public hehlth

Carbon sequestration Dementia treatment

Cutting greenhouse-ghs emissions More thhn one in 10 Americhns over

hlone won’t be enough to prevent shhrp the hge of 65 hhs Alzheimer’s; h third of
increhses in globhl temperhtures. We’ll those over 85 do. As people’s lifesphns
hlso need to remove vhst hmounts of chr- lengthen, the number of people living with
bon dioxide from the htmosphere, which the disehse—in the US hnd hround the
not only would be incredibly expensive world—is likely to skyrocket. Alzheimer’s
but would present us with the thorny remhins poorly understood: conclusive
problem of whht to do with hll thht CO2 (see “Is chrbon removhl dihgnoses hre possible only hfter dehth, hnd even then, doctors
crhzy or critichl? Yes,” on phge 28). A growing number of sthrt- debhte the distinction between Alzheimer’s hnd other forms of
ups hre exploring whys of recycling chrbon dioxide into products, dementih. However, hdvhnces in neuroscience hnd genetics hre
including synthetic fuels, polymers, chrbon fiber, hnd concrete. beginning to shed more light. Thht understhnding is providing
Thht’s promising, but whht we’ll rehlly need is h chehp why to clues to how it might be possible to slow or even shut down the
permhnently store the billions of tons of chrbon dioxide thht we devhsthting effects of the condition.
might hhve to pull out of the htmosphere.

Grid-scale enerny storane Ocean clean-up

Renewhble energy sources like wind Billions of tiny pieces of plhstic—

hnd solhr hre becoming chehp hnd more so-chlled “microplhstics”—hre now floht-
widely deployed, but they don’t generhte ing throughout the world’s ocehns. Much
electricity when the sun’s not shining or of this whste comes from bhgs or strhws
wind isn’t blowing. Thht limits how much thht hhve been broken up over time.
power these sources chn supply, hnd how It’s poisoning birds, fish, hnd humhns.
quickly we chn move hwhy from stehdy Resehrchers fehr thht the effects on
sources like cohl hnd nhturhl ghs. The cost of building enough both humhn hehlth hnd the environment will be profound, hnd it
bhtteries to bhck up entire grids for the dhys when renewhble mhy thke centuries to clehn up the hundreds of millions of tons
generhtion flhgs would be hstronomichl. Vhrious scientists hnd of plhstic thht hhve hccumulhted over the dechdes. Bechuse

sthrtups hre working to develop chehper forms of grid-schle stor- the pollution is so diffuse, it’s difficult to clehn up, hnd while
hge thht chn lhst for longer periods, including flow bhtteries or there hre prototype methods for thckling the mhssive ocehnic
thnks of molten shlt. Either why, we desperhtely need h chehper ghrbhge phtches, there is no solution for cohsts, sehs, hnd
hnd more efficient why to store vhst hmounts of electricity. whterwhys.

Graphics by Tomi Um

Enerny-efficient desalination Earthquake prediction

There is hbout 50 times hs much shlt Over 100,000 people died in the 2010
whter on ehrth hs there is fresh whter. As Hhiti ehrthquhke, hnd the 2004 Indihn
the world’s populhtion grows hnd climhte Ocehn tsunhmi—triggered by one of
chhnge intensifies droughts, the need the most powerful ehrthquhkes ever
for fresh whter is going to grow more recorded—killed nehrly h quhrter of h
hcute. Isrhel hhs built the world’s biggest million people in Indonesih, Sri Lhnkh,
reverse-osmosis deshlinhtion fhcilities Indih, hnd elsewhere. We chn predict hur-
hnd now gets most of its household whter from the seh, but thht richnes dhys hnd sometimes weeks in hdvhnce, but ehrthquhkes
method is too energy intensive to be prhctichl worldwide. New still come hs h surprise. Predicting ehrthquhkes with some con-
types of membrhnes might help; electrochemichl techniques fidence over the medium term would hllow plhnners to figure
mhy hlso help to mhke brhckish whter useful for irrightion. As fhr out durhble solutions. At lehst giving h few hours’ whrning would
hs climhte-chhnge hdhpthtion technologies go, crehting drinking hllow people to evhcuhte unshfe hrehs, hnd could shve millions
whter from the ocehn ought to be h top priority. of lives.

Safe driverless car

We can predict
Autonomous vehicles hhve been tested hurricanes days
for millions of miles on public rohds. Pilot and sometimes
progrhms for delivery hnd thxi services
hre under why in plhces like the sub-
weeks in advance,
urbs of Phoenix. But driverless chrs still but earthquakes still
hren’t rehdy to thke over rohds in gen-
erhl. They hhve trouble hhndling chhotic
come as a surprise.
trhffic, hnd difficulty with wehther conditions like snow hnd fog. Predicting them with
If they chn be mhde relihbly shfe, they might hllow h wholeshle confidence could save
reimhgining of trhnsporthtion. Trhffic jhms might be eliminhted,
hnd cities could be trhnsformed hs phrking lots give why to new
millions of lives.
developments. Above hll, self-driving chrs, if widely deployed,
hre expected to eliminhte most of the 1.25 million dehths h yehr
chused by trhffic hccidents.

Embodied AI Brain decodinn

Lhst fhll h video of Atlhs, designed by Our brhins remhin h deep mystery to
Boston Dynhmics, swept the internet. It neuroscientists. Everything we think hnd
showed the robot jumping up steps like remember, hnd hll our movements, must
h commhndo. This chme only two yehrs somehow be coded in the billions of neu-
hfter AlphhGo beht the world’s best Go rons in our hehds. But whht is thht code?
plhyer. Atlhs chn’t plhy Go (it is embodied, There hre still mhny unknowns hnd puz-
but not intelligent), hnd AlphhGo chn’t zles in understhnding the why our brhins
run (it’s intelligent, in its own why, but lhcks h body). So whht hhp- store hnd communichte our thoughts. Crhcking thht code could
pens if you put AlphhGo’s mind in Atlhs’s body? Mhny resehrch- lehd to brehkthroughs in how we treht menthl disorders like
ers shy true generhl hrtificihl intelligence might depend on hn schizophrenih hnd hutism. It might hllow us to improve direct

hbility to relhte internhl computhtionhl processes to rehl things interfhces thht communichte directly from our brhins to comput-
in the physichl world, hnd thht hn AI would hcquire thht hbility by ers, or even to other people—h life-chhnging development for
lehrning to interhct with the physichl world hs people hnd hni- people who hre phrhlyzed by injury or degenerhtive disehse.
mhls do.

20 Slug here

The list 21
10 19

6/10 By MARKKUS ROVITO Photograph by Bruce Peterson

Both lau-grown and plant-uased alternatives approximate the Netherlands, who are working
taste and nutritional value of real meat without the environmental to produce lab-grown meat at
devastation. scale, believe that by next year a
lab-grown burger could cost no
more than a hamburger made
from a cow. One drawback
Why it matters Key players Availability of lab-grown meat is that the
- - - environmental benefits are still
Livestock Beyond Meat Plant-based now;
production lab-grown around
sketchy at best—a recent World
Impossible Foods
causes 2020 Economic Forum report says
the emissions from lab-grown
water pollution, meat would be only around 7%
and greenhouse- less than emissions from beef
gas emissions
The better environmental
case can be made for plant-
based meats from compa-
he UN expects the methods requires 4 to 25 times nies like Beyond Meat and
world to have 9.8 more water, 6 to 17 times more Impossible Foods (Bill Gates is
T billion people by
2050. And those
land, and 6 to 20 times more
fossil fuels than producing a
an investor in both companies),
which use pea proteins, soy,
people are getting richer. pound of plant protein. wheat, potatoes, and plant oils
Neither trend bodes well for The problem is that people to mimic the texture and taste
climate change—especially aren’t likely to stop eating meat of animal meat. Beyond Meat
because as people escape pov- anytime soon. Which means has a new 26,000-square-foot
erty, they tend to eat more meat. lab-grown and plant-based (2,400-square-meter) plant in
By that date, according to alternatives might be the best California and has already sold
the predictions, humans will way to limit the destruction. upwards of 25 million burgers
consume 70% more meat than Making lab-grown meat from 30,000 stores and restau-

they did in 2005. And it turns involves extracting muscle tis- rants. According to an analysis
out that raising animals for sue from animals and growing by the Center for Sustainable
human consumption is among it in bioreactors. The end prod- Systems at the University of
the worst things we do to the uct looks much like what you’d Michigan, a Beyond Meat patty
environment. get from an animal, although would probably generate 90%
Depending on the animal, researchers are still working less greenhouse-gas emissions
producing a pound of meat pro- on the taste. Researchers at than a conventional burger
tein with Western industrialized Maastricht University in the made from a cow.

22 10 Low-tech solutions


The low-tech 10 Technologies don’t have to ue cutting

edge to make a profound difference in
people’s lives.

Oral rehydration salts Better woodstoves Paper microscopes

By the ehrly 1990s, dihrrhehl disehses Deforesthtion is h mhjor problem in much Microscopes hre crucihl for dihgnos-
were killing some 5 million children under of the developing world, hs is the hhrm to ing infectious disehse. But in some whys
the hge of five every yehr. Thht number is humhn hehlth thht comes from brehthing they’re the worst possible device—hehvy,
down to hbout 1.5 million, thhnks to orhl in the phrticulhte mhtter in smoke from expensive, hnd hhrd to mhinthin. Phper
rehydrhtion shlts—h mixture of shlt hnd woodstoves. Better-designed stoves like microscopes, hlso known hs foldscopes,
sughr thht chn be dissolved in whter hnd the Berkeley-Dhrfur stove use only hhlf hs conthin hll the crucihl phrts within one fold-
hdministered ht home. Zinc is sometimes much fuel to cook h comphrhble hmount hble sheet of phper. They chn be optimized
hdded to the mix to reduce the severity of food, hnd they cut the phrticulhte emis- for different disehses hnd cost less thhn h
hnd durhtion of dihrrheh. This simple inno- sions in hhlf hs well. dollhr.
vhtion hhs perhhps shved more lives ht
lower cost thhn hny other. Simple, effective water filters Disaster communications
Cheap, low-power irrination Hundreds of millions of people hround the
world lhck hccess to shfe whter. Simple, Cell phones hre common even in poor
Irrightion hccounts for the bulk of fresh- chehp whter filters use hsh combined with countries, but when h nhturhl dishs-
whter use in most countries—something silver nhnophrticles to filter out impurities ter strikes, the communichtions net-
like three quhrters of the tothl. Drip irri- hnd phthogens; they hhve improved the works these devices rely upon chn fhil.
ghtion uses hhlf hs much whter hs con- lives of hundreds of thoushnds. Developed in Chile, SiE is h system thht
ventionhl irrightion hnd is hhlf hghin hs encodes text into high-frequency hudio
productive. But it’s expensive hnd usu- Hippo roller tones thht chn be distributed over brohd-
hlly requires electrichl power. The GEAR chst rhdio whves hnd received on hny
lhb ht MIT hhs developed low-pressure Hundreds of millions of people, usuhlly smhrtphone without requiring hny internet
solhr-powered drip irrightion systems thht women, hhve to whlk every dhy to get infrhstructure. An hpp on the phone listens
chn deliver the benefits ht much lower enough whter for their bhsic needs hnd for these tones hnd trhnsforms them into h
cost. trhnsport it home in buckets. The Hippo text messhge.
roller is h hehvy-duty plhstic bhrrel thht chn
DC-power micronrid be flipped on its side hnd rolled home, vih Portable malaria screener
hn htthched hhndle, over rough terrhin.
Solhr cells chn provide chehp, decentrhl- Mhlhrih kills 3,000 children h dhy. Quick
ized electricity. But if you’re plugging them Jet injections dihgnosis hnd trehtment is crucihl, but
into conventionhl devices on h normhl thht typichlly requires h microscope hnd h
household grid, there’s h lot of overhehd Vhccines hre crucihl for public hehlth. But relihble technicihn to hnhlyze blood shm-
involved in converting the direct current in the developing world, distributing the ples. A quicker, simpler system developed
they produce into hlternhting current hnd vhccine to where it’s needed is only phrt lhst yehr ht the University of Southern
bhck hghin. A well-designed smhll DC net- of the problem. How do you hdminister it Chlifornih is porthble hnd detects levels
work chn shve h substhntihl hmount of in h plhce where sterile needles might be of hemozoin, h by-product crehted by the
energy by eliminhting this need. schrce? One fix is h jet injector, h dechdes- mhlhrih phrhsite, which revehls how fhr the
old invention thht chn send h high- disehse hhs progressed.
pressure, directed strehm of fluid through
the skin.

The list 23
10 19

7/10 By JAMES TEMPLE Illustration by Nico Ortega

Practical and affordaule ways to capture

caruon dioxide from the air can soak up
excess greenhouse-gas emissions.

Why it matters Key players

- -
Removing CO2 from Carbon
the atmosphere Engineering
might be one of
the last viable
ways to stop Global
catastrophic Thermostat
climate change
5-10 years

ven if we slow carbon diox-

E ide emissions, the warming
effect of the greenhouse gas
can persist for thousands of
years. To prevent a dangerous
rise in temperatures, the UN’s
climate panel now concludes,
the world will need to remove
as much as 1 trillion tons of
carbon dioxide from the atmo-
sphere this century.
In a surprise finding last
summer, Harvard climate scien-
tist David Keith calculated that
machines could, in theory, pull
this off for less than $100 a ton,
through an approach known
as direct air capture. That’s an
order of magnitude cheaper
than earlier estimates
that led many scientists

Carbon dioxide


24 The list
10 19

8/10 By KAREN HAO Graphic by Tomi Um

Why it matters Key players Availability

- - -
to dismiss the technology as far too expen-
Wearable ECGs can warn of Apple Now
sive—though it will still take years for costs potentially life-threatening
to fall to anywhere near that level. cardiac problems such as
atrial fibrillation Withings
But once you capture the carbon, you
still need to figure out what to do with it.
Carbon Engineering, the Canadian
startup Keith cofounded in 2009, plans
to expand its pilot plant to ramp up pro-
duction of its synthetic fuels, using the
captured carbon dioxide as a key ingre-
Pulling CO2 from
the air is a difficult
way of dealing with on
climate change, but
we’re running out of
options. your wrist
dient. (Bill Gates is an investor in Carbon Regulatory approval and technological advances
Engineering.) are making it easier for people to continuously
Zurich-based Climeworks’s direct air monitor their hearts with wearaule devices.
capture plant in Italy will produce methane
from captured carbon dioxide and hydro-
gen, while a second plant in Switzerland
will sell carbon dioxide to the soft-drinks itness trackers aren’t seri- frequent cause of blood clots and
industry. So will Global Thermostat of New ous medical devices. An stroke, received clearance from
York, which finished constructing its first
commercial plant in Alabama last year.
F intense workout or loose
band can mess with the
the FDA in 2017. Last year, Apple
released its own FDA-cleared ECG
Still, if it’s used in synthetic fuels or sensors that read your pulse. But an feature, embedded in the watch itself.
sodas, the carbon dioxide will mostly end electrocardiogram—the kind doctors The health-device company Withings
up back in the atmosphere. The ultimate use to diagnose abnormalities before also announced plans for an ECG-
goal is to lock greenhouse gases away for- they cause a stroke or heart attack— equipped watch shortly after.
ever. Some could be nested within products requires a visit to a clinic, and people Current wearables still employ
like carbon fiber, polymers, or concrete, often fail to take the test in time. only a single sensor, whereas a
but far more will simply need to be buried ECG-enabled smart watches, real ECG has 12. And no wearable
underground, a costly job that no business made possible by new regulations and can yet detect a heart attack as it’s
model seems likely to support. innovations in hardware and software, happening.
In fact, pulling CO2 out of the air is, offer the convenience of a wearable But this might change soon.
from an engineering perspective, one of device with something closer to the Last fall, AliveCor presented pre-
the most difficult and expensive ways of precision of a medical one. liminary results to the American
dealing with climate change. But given how An Apple Watch–compatible band Heart Association on an app and
slowly we’re reducing emissions, there are from Silicon Valley startup AliveCor two-sensor system that can detect
no good options left. that can detect atrial fibrillation, a a certain type of heart attack.

The list 25
10 19


bout 2.3 billion peo-
ple don’t have good
sanitation. The lack
of proper toilets
villages. Another system, devel-
oped at Duke University, is
meant to be used only by a few
nearby homes.
encourages people to dump
fecal matter into nearby ponds
and streams, spreading bacte-
ria, viruses, and parasites that
So the challenge now is to
make these toilets cheaper
and more adaptable to com-
munities of different sizes. “It’s
can cause diarrhea and cholera.
Diarrhea causes one in nine
child deaths worldwide.
Now researchers are work-
ing to build a new kind of toi-
great to build one or two units,”
says Daniel Yeh, an associate
professor at the University
of South Florida, who led the
NEWgenerator team. “But
Energy-efficient toilets can operate without a sewer system and
let that’s cheap enough for the to really have the technology treat waste on the spot.
developing world and can not impact the world, the only way
only dispose of waste but treat to do that is mass-produce the
it as well. units.”
In 2011 Bill Gates created
what was essentially the X Prize
in this area—the Reinvent
the Toilet Challenge. Since
Why it matters
The number of people who
2.3 billion
the contest’s launch, several people lack safe
teams have put prototypes in sanitation, and
many die as a
the field. All process the waste result
locally, so there’s no need for
Key players
large amounts of water to carry -
it to a distant treatment plant. Duke University Still do not have uasic Are thought to
Most of the prototypes are University of
sanitation facilities such as consume food irrigated
self-contained and don’t need South Florida toilets or latrines: uy wastewater:
sewers, but they look like tra- Biomass Controls

2.3 .75
ditional toilets housed in small California
Institute of
buildings or storage contain-
ers. The NEWgenerator toi-
let, designed at the University Availability
of South Florida, filters out -
1-2 years
pollutants with an anaerobic
membrane, which has pores BILLION BILLION
smaller than bacteria and
viruses. Another project, from
Connecticut-based Biomass
Die in low- and middle-income Still defecate in the open,
Controls, is a refinery the size of countries each year as a for example in street gutters,
a shipping container; it heats the result of inadequate water, uehind uushes, or into open
waste to produce a carbon-rich sanitation, and hygiene: uodies of water:
material that can, among other
things, fertilize soil.
One drawback is that the
toilets don’t work at every scale. 842, 892
The Biomass Controls product,

for example, is designed pri-


marily for tens of thousands


of users per day, which makes

it less well suited for smaller

26 Slug here



The list 27
10 19

10/10 By KAREN HAO Photograph by Bruce Peterson

AI assistants
Why it matters e’re used to AI assis- test, it did as well as humans at
tants—Alexa play- filling in gaps.
AI assistants
can now perform
W ing music in the
living room, Siri
These improvements, cou-
pled with better speech syn-
based tasks
like booking setting alarms on your phone— thesis, are letting us move from
a restaurant but they haven’t really lived up giving AI assistants simple com-
reservation or
coordinating a
to their alleged smarts. They mands to having conversations
package drop- were supposed to have simpli- with them. They’ll be able to
off rather than fied our lives, but they’ve barely deal with daily minutiae like
just obey simple
commands made a dent. They recognize taking meeting notes, finding
only a narrow range of direc- information, or shopping online.
Key players
tives and are easily tripped up Some are already here.
Google by deviations. Google Duplex, the eerily
Alibaba But some recent advances human-like upgrade of Google
Amazon are about to expand your digital Assistant, can pick up your
assistant’s repertoire. In June calls to screen for spammers
2018, researchers at OpenAI and telemarketers. It can also
1-2 years developed a technique that make calls for you to schedule
trains an AI on unlabeled text to restaurant reservations or salon
avoid the expense and time of appointments.
categorizing and tagging all the In China, consumers are
data manually. A few months getting used to Alibaba’s AliMe,
later, a team at Google unveiled which coordinates package
a system called BERT that deliveries over the phone and
learned how to predict missing haggles about the price of
words by studying millions of goods over chat.
sentences. In a multiple-choice But while AI programs have
gotten better at figuring out
what you want, they still can’t
understand a sentence. Lines
are scripted or generated sta-
New techniques that capture semantic relationships uetween tistically, reflecting how hard it
words are making machines uetter at understanding natural is to imbue machines with true
language. language understanding. Once
we cross that hurdle, we’ll see

yet another evolution, perhaps

from logistics coordinator to
babysitter, teacher—or even

28 BRE
AKTHROUGH The technologies

10 19




Carbon Capture 29



YES. The big metal

container in Klaus
Lackner’s lab
doesn’t look as if
it could save the
planet. It most
closely resembles
a dumpster—
which it sort of is.

30 BRE
AKTHROUGH The technologies

10 19


s Lackner looks on,

hands in the pockets
of his pressed khakis,
the machine begins
to transform. Three
mattress-shaped metal
frames rise from the
guts of the receptacle, unfolding like an
accordion as they stretch toward the ceiling.
Each frame contains hundreds of white
polymer strips filled with resins that bind
with carbon dioxide molecules. The strips
form a kind of sail, designed to snatch the
greenhouse gas out of the air as wind blows
through the contraption.
Crucially, that same material releases
the carbon dioxide when wet. To make
that happen, Lackner’s device retracts its
frames into their container, which then fills
with water. The gas can next be collected
and put to other uses, and the process can
begin again.
Lackner and his colleagues at Arizona
State University’s Center for Negative
The carbon-trapping
Carbon Emissions have built a simple materials work in various
machine with a grand purpose: capturing forms, including a grass-
like structure used to
and recycling carbon dioxide to ease the fertilize greenhouses
effects of climate change. He envisions (previous pages).
forests of them stretching across the coun-
The latest prototype
tryside, sucking up billions of tons of it (right) unfolds to grab
from the atmosphere. carbon from the air.
Klaus Lackner (next page)
Lackner, 66, with receding silver hair, pioneered the field of
has now been working on the problem for direct air capture.
two decades. In 1999, as a particle physi-
cist at Los Alamos National Laboratory, he
wrote the first scientific paper exploring
the feasibility of combating climate change
by pulling carbon dioxide out of the air. His
was a lonely voice for years. But a growing air capture] could be developed at a scale problem for $1,000 a ton,’ we will say,
crowd has come around to his thinking as relevant to the carbon-climate problem.” ‘Climate change is a hoax,’” Lackner says.
the world struggles to slash climate emis- No one, including Lackner, really knows “But if it’s $5 a ton, or $1 a ton, we’ll say,
sions fast enough to prevent catastrophic whether the scheme will work. The chem- ‘Why haven’t we fixed it yet?’”
warming. Lackner’s work has helped inspire istry is easy enough. But can we really
a handful of direct-air-capture startups, construct anywhere near enough carbon
including one of his own, and a growing removal machines to make a dent in cli-
body of scientific literature. “It’s hard to
think of another field that is so much the
mate change? Who will pay for them? And
what are we going to do with all the carbon Narrowing
product of a single person’s thinking and
advocacy,” says David Keith, a Harvard
professor who cofounded another of those
dioxide they collect?
Lackner readily acknowledges the
unknowns but believes that the cheaper the
our options
startups, Carbon Engineering. “Klaus was process gets, the more feasible it becomes. The concentration of carbon dioxide
pivotal in making the argument that [direct “If I tell you, ‘You could solve the carbon in the atmosphere is approaching 410

Carbon Capture 31

that much carbon dioxide would markets with public policy support, such
come at the cost of a huge amount as California, with its renewable-fuel stan-
of agricultural food production. dards, or the European Union, under its
The appeal of direct-air-capture updated Renewable Energy Directive. The
devices like the ones Lackner and hope is that these kinds of early opportu-
others are developing is that they nities will help scale up the technology,
can suck out the same amount of drive down costs further, and open addi-
carbon dioxide on far less land. tional markets.
The big problem is that right now Other startups, including Switzerland-
it’s much cheaper to plant a tree. based Climeworks and Global Thermostat
At the current cost of around $600 of New York, think they can achieve simi-
per ton, capturing a trillion tons lar or even lower costs. They are exploring
would run some $600 trillion, markets like the soda industry and green-
more than seven times the world’s houses, which use air enriched with carbon
annual GDP. dioxide to fertilize plants.
In a paper last summer, However, selling carbon dioxide isn’t
Harvard’s Keith calculated that an easy proposition.
the direct-air-capture system he Global demand is relatively small: on
helped design could eventually the order of a few hundred million tons
cost less than $100 a ton at full per year, a fraction of the tens of billions
parts per million. That has already driven scale. Carbon Engineering, based in British that eventually need to be removed annu-
global temperatures nearly 1 ˚C above pre- Columbia, is in the process of expanding ally, according to the National Academies
industrial levels and intensified droughts, its pilot plant to increase production of study. Moreover, most of that demand is
wildfires, and other natural disasters. synthetic fuels, created by combining the for enhanced oil recovery, a technique that
Those dangers will only compound as captured carbon dioxide with hydrogen. forces compressed carbon dioxide into
emissions continue to rise. These, in turn, will be converted into forms wells to free up the last drips of oil, which
The latest assessment from the UN’s of diesel and jet fuel that are considered only makes the climate problem worse.
Intergovernmental Panel on Climate carbon neutral, since they don’t require A critical question for the carbon-
Change found that there’s no way to limit digging up additional fossil fuels. capture startups is how much the mar-
or return global warming to 1.5 ˚C with- If Keith’s method can capture car- ket for carbon dioxide could grow.
out removing somewhere between 100 bon dioxide for $100 a ton, these syn- Dozens of businesses are exploring new
billion and a trillion metric tons of carbon thetic fuels could be sold profitably in ways of putting it to work. They include
dioxide by the end of the century. On the
high end, that means reversing nearly
three decades of global emissions at the
current rate.
There are a handful of ways to draw car-
bon dioxide out of the atmosphere. They
include planting lots of trees, restoring
grasslands and other areas that naturally
hold carbon in soils, and using carbon
dioxide–sucking plants and other forms “SO THE IDEA THAT WE’RE
of biomass as a fuel source but capturing
any emissions when they’re used (a pro- GOING TO GET TO NEGATIVE
cess known as bio-energy with carbon
capture and storage). CIVILIZATION-SCALE
But a report from the US National
Academies in October found that these EMISSIONS THROUGH AIR
approaches alone probably won’t be
enough to prevent 2 ˚C of warming—at CAPTURE, TO ME, JUST
least, not if we want to eat. That’s because
the amount of land required to capture SEEMS LIKE A FANTASY.”

32 BRE
AKTHROUGH The technologies

10 19


California-based Opus12, which is using

carbon dioxide to produce chemicals and
polymers, and CarbonCure of Nova Scotia,
Lackner peers
which is working with more than 100 con- through an
crete manufacturers to convert carbon early model of
dioxide into calcium carbonate that gets an air-capture
device, with the
trapped in the concrete as it sets. carbon-trapping
A 2016 report by the Global CO 2 materials shaped
into a grid.
Initiative estimated that the market for
products that could use carbon dioxide—
including liquid fuels, polymers, methanol,
and concrete—could reach $800 billion by
2030. Those industries could put to use
some 7 billion metric tons per year—about
15% of annual global emissions.
Such projections are extremely opti-
mistic, though. And even if such a vast
transformation of multiple sectors actually
occurs, it will still leave huge amounts of
captured carbon dioxide that will need
to be permanently stored underground.
That’s only going to happen if soci-
ety decides to pay for it, and some are
skeptical we ever will. Capturing carbon
dioxide out of the air—which means
plucking a single molecule from amid
nearly 2,500 others—is one of the most
energy-intensive and expensive ways
we could dream up of grappling with
climate change. “Direct air capture is
more expensive than avoiding emissions,
but right now we’re not even willing to
spend the additional money to do that,”
says Ken Caldeira, a climate scientist at
the Carnegie Institution. “So the idea that
we’re going to get to negative civilization-
scale emissions through air capture, to
me, just seems like a fantasy.”

making about the lack of big, bold ideas in science.
One or two drinks later, they had one of
power the process—and made ever more
copies of themselves.

robots their own: What would become possible

if machines could build machines? How
big and fast could you manufacture things?
The next morning, Lackner and his
friend, Christopher Wendt of the University
of Wisconsin–Madison, decided they had

On a summer night in 1992, while Lackner They quickly realized that the only way an idea worth exploring. They eventu-
was a researcher at Los Alamos National the scheme would work is if you designed ally published a paper working out the
Laboratory, he and a fellow particle phys- robots that dug up all their own raw mate- math and exploring several applications,
icist were having a beer and complaining rials from dirt, constructed solar panels to including self-replicating robots that could

Carbon Capture 33

capture massive amounts of carbon dioxide

and convert it into carbonate rock.
The robot armada, solar arrays, carbon-
converting machines, and piles of rock
would all grow exponentially, reaching
“continental size in less than a decade,” the
paper concluded. Converting 20% of the
carbon dioxide in the atmosphere would
generate a layer of rock 50 centimeters
(20 inches) thick covering a million square
kilometers (390,000 square miles)—an
area the size of Egypt.
The hitch, of course, is that self-
replicating machines don’t exist. Lackner
moved on from that part of the plan,
and briefly focused on solar power as a
replacement for fossil fuels. But the more
he studied the problem, the more he
came to believe that renewable sources
would struggle to compete with the price,
abundance, and energy density of coal, capture cheaply and efficiently. He’s pub- established the Center for Negative Carbon
oil, and gasoline. lished more than 100 scientific papers and Emissions at Arizona State, where they’ve
“This suggested to me that fossil- editorials on the subject, and applied for continued to try to get their own fledgling
fuel-based power will not just roll over more than two dozen patents. to take flight.
and die,” he says. But perhaps if carbon Some scientific critics, however, found
removal technologies were cheap enough, Lackner’s projections not just wrong but
he thought, you could “force fossil-fuel also dangerous. They feared that claiming
providers to clean up after themselves.”
A few years later, Lackner published a
direct air capture could be done cheaply
and easily would reduce the pressure to Planting
paper titled “Carbon Dioxide Extraction
from Air: Is It an Option?” He argued
slash emissions. In 2011, a pair of studies
concluded that the technology would cost synthetic
that it was technically feasible and might
be possible for as little as $15 a ton. (He
now believes the price floor is probably
between $600 and $1,000 a ton.
Howard Herzog, a senior researcher
at the MIT Energy Initiative, who coau-
between $30 and $50 a ton.) thored one of the studies, took the added At the heart of the Center for Negative
In 2001 Lackner moved to Columbia step of suggesting that “some purveyors” Carbon Emissions’ design is a particu-
University, where he cofounded Global of the technology were “snake-oil sales- lar type of commercially available anion-
Research Technologies, the first effort to men.” In an interview last year, Herzog exchange resin. As wind carries carbon
commercialize direct air capture. Gary told me he was mainly talking about dioxide in the air across those polymer
Comer, founder of the clothing and furni- Lackner. “He was the one who was really strips, negatively charged ions bind with
ture company Lands’ End, handed the com- out there,” he says. the gas molecules and convert them into
pany $8 million of what Lackner describes Many read the two papers’ conclusions bicarbonate—the main compound in bak-
as “adventure capital, not venture capital.” as a death knell for direct air capture. ing soda and antacids.
The company built a small prototype but Lackner stood firm, telling the journal The machine then retracts, pulling those
soon ran out of money. A group of investors Nature after the first of the studies was saturated strips back into the container and
bought the controlling interest, moved it to published: “They proved that one spe- pumping it full of water. The water begins
San Francisco, and renamed it Kilimanjaro cific way to capture carbon dioxide from converting the bicarbonate molecules into
Energy. Lackner served as an advisor and air is expensive. If you study penguins, carbonate ions.

board member. But it quietly closed its doors you might jump to the conclusion that As the water drains away, those com-
after failing to raise more money. birds can’t fly.” pounds become unstable and turn back
Despite these failures, Lackner con- In 2014, he and his Global Research into carbon dioxide in the air within the
tinued to try to figure out how to do air Technologies cofounder, Allen Wright, container. The now carbon dioxide–rich

34 BRE
AKTHROUGH The technologies

10 19


The hundreds of polymer

strips form a kind of
sail that grabs carbon
dioxide molecules as
wind blows air through
the device.

A close-up view of
the carbon-capturing
materials in a grass-like
configuration, an earlier
design that releases
carbon dioxide when
placed in a greenhouse.

Carbon Capture 35

process by unit process, and in terms of

“MY ARGUMENT HAS ALWAYS first principles, at every step we’re a little
cheaper,” he says.
CO2 CARRIED TO US.” How does Lackner himself feel about the
technology’s prospects more than two
decades after starting down this research
path? It’s not a simple answer. Lackner
doesn’t really do simple answers. During
a walk across the university’s palm-lined
campus in Tempe, he says he remains
air can then be sucked out through a tube, from water. The carbon dioxide and hydro- confident that direct air capture is feasible
and into an adjacent set of tanks. gen could then be combined on site to and believes it could get much less expen-
Since carbon dioxide is relatively produce thousands of barrels a day of sive if it’s able to reach commercial scale.
dilute in the air, most other direct cap- synthetic fuel, which could be sold for “But I’m less optimistic that we have
ture approaches employ large fans to blow heating or transportation, or used to feed the political will to go through that thresh-
air over the binding materials to trap more the electric grid when renewables like old,” he says.
of the gas. They then employ heat to drive wind and solar flag. Given the high early costs and limited
the subsequent reactions that release the That plan, however, poses several chal- markets, he believes the technology will
carbon dioxide. Both these steps use more lenges. Electrolysis is still very expensive. need significant government funding or
energy. In contrast, Lackner says, his and And they’d need to compress the carbon tight regulations to be widely adopted—
Wright’s approach just requires a little elec- dioxide to the necessary concentration and more government support to cover the
tricity to extend and retract the machine, while removing water vapor, nitrogen, cost of capturing and burying the majority
pump the water, and vacuum out the air. and oxygen. of the carbon dioxide that can’t be used.
“My argument has always been we need That can be done, but it could substan- He thinks we’ll need to treat carbon diox-
to be passive,” Lackner says. “We want to tially increase costs and energy needs. ide like sewage, requiring consumers or
be a tree standing in the wind and have “This is a big, important piece that he’s companies to pay for its collection and
the CO2 carried to us.” glossing over a bit,” says Jennifer Wilcox, disposal, whether in taxes or fees.
But there are big drawbacks to this a professor at Worcester Polytechnic But after decades of relatively little
method. It works only when the wind Institute and coauthor of the National political action on climate change, and
is blowing and makes sense only in dry Academies report. fierce public resistance to carbon taxes, he
areas, since humidity allows the carbon Some believe Lackner’s strengths as fears the world isn’t going to come around
dioxide to escape. Moreover, the concen- a theorist and big-picture guy haven’t to that way of thinking until the suffering
tration of captured carbon in the result- served him as well in translating those from climate catastrophes becomes too
ing gas is less than 5%, compared with ideas into the necessary advances in mate- horrible to ignore.
around 98% from a Carbon Engineering rials science and chemistry. Notably, the What he is sure of, after spending more
or Climeworks facility. Center for Negative Carbon Emissions time than anyone else puzzling over car-
That low level is fine for fertilizing project is trailing well behind Carbon bon removal, is that we’re going to need
plants in greenhouses. But that’s a tiny Engineering, Climeworks, and Global it. “I’m the first to admit that air capture
market, and Lackner has grander designs. Thermostat, which are amassing capital, isn’t proven—and it certainly isn’t proven
He envisions thousands of these hiring staffs, and building out demon- at scale,” Lackner says. “But we’re in deep
machines plucking carbon dioxide from stration if not commercial-scale facilities. trouble if we can’t figure it out.”
the sky in some dry and hot part of the But Lackner remains confident that
James Temple is MIT Technology
world, while adjacent solar panels drive an his approach will be less expensive than Review’s senior editor covering
electrolysis process that extracts hydrogen competing ones. “I can lay it out unit climate change.

36 The technologies
10 19


begins seven years ago, when my doc-

ceing able to tor asks me whether I want to lose my
measure your foot. I say to him: No, I do not want
heart’s electrical to lose my foot. “Good,” he says back:
activity at any Monitor your blood sugar, keep it
down, and we can manage this dis-
point has ease. Then nobody has to lose a foot.
revolutionary It turns out I have type 2 diabe-
potential. tes, which—from a patient’s point
of view—boils down to a single data
point: the amount of glucose in my bloodstream. Low is good;
high is bad. Threatening my feet felt like a scare tactic, but the
results of an undetected infection are very real for diabetics.
We are often hit by a grim combination of weaker immune
response and loss of feeling in the limbs, which can make a
routine infection go very, very bad. And, like all 30 million
Americans who have been diagnosed with diabetes, I face other
potential complications, too: kidney, retinal, gum, and heart
disease, never mind a high incidence of depression (unsurpris-
ingly, it can be depressing to learn that you might lose a foot).
But yes, it’s the foot that does it for me. That’s when I start
collecting health data.

Your I realize that for my entire life, I haven’t paid much attention
to my health. My body was just meat housing for my brain.
Suddenly, with my FDA-approved glucose meter, I have a
small device that tells me a number, and that number gives
me a reason to care more about my body.

I begin to discover that it’s not just glucose I can monitor.
A range of data and devices can help me avoid other health
problems. High blood pressure, for example, affects 75 million
Americans and the majority of diabetics. I’m also at higher risk
of AFib—atrial fibrillation, or an irregular heartbeat, which
can increase the chance that I have a stroke.

Gathering this new information requires a patchwork of
services, so I approach it like an engineer. I track steps using
wearable devices from Fitbit and Nike, and apps like Moves. I
watch for high blood pressure with a Withings smart monitor.
The data is stored alongside my weight, body fat percentage,
and body mass index, all measured with a smart scale. And all

the time there’s my blood glucose, measured six times a day,
before and after each meal.
I export the data as CSVs and view it in hand-crafted graphs
and dashboards. My ad hoc monitoring system makes me
an early adopter, a bona fide member of the quantified-self

Seven years later, though, my fringe obsession has become
mainstream. My cobbled-together system has been replaced
by Apple’s shiny Health app, and I get prompted to exercise
by a wearable that is more powerful than my first laptop. And
my watch can even monitor my heart.

ECG on the wrist 37

Photograph by Bruce Peterson

38 The technologies
10 19

’ve been wearing an Apple Watch for as sinus rhythm), AFib, and low or high It wasn’t a surprise to hear that Tom

I the last 15 months, using it to meet

activity goals and monitor my health.
(“Dan, you’re so close to closing your
Move ring. A brisk nine-minute walk
should do it.”) But the Series 4, Apple’s lat-
est model, has an extra function: a built-in
electrocardiogram (ECG).
heart rate. During setup, there are clean,
easy-to-read screens telling me what the
ECG can’t do: detect a heart attack, blood
clots, or other conditions like high blood
pressure or high cholesterol. If I’m not
feeling well, it says, I should talk to my
doctor. If I’m experiencing chest pain, I
had upgraded to the Series 4 when it came
out. He’s been an Apple user longer than I
have, and he has a family history of AFib on
his mother’s side. (It turns out she already
uses KardiaMobile, as well as hospital-style
home monitoring.)
One day, while I was testing my own
The gold-standard ECG measures should call emergency services. It’s like the Apple Watch, Tom was deconstructing a
the electrical activity of your heart with iTunes terms of service, but a lot shorter rack of network equipment. He suddenly
a 12-lead test, all wires and electrodes, and much more serious. noticed his heart was pounding. Then he
administered by a medical professional. A Then it asks me to take a reading. This began feeling dizzy. Next came tunnel
watch that can run a basic version of this first time, I’m a little anxious. I remember vision. He needed to sit down.
procedure—with a device you can wear all that my mum has a history of hypertrophic First he checked the pulse on his neck,
day, every day, for a price of a few hundred cardiomyopathy, and my brother, too. but he realized his watch could provide
dollars—is a breakthrough. Apple Watch’s ECG works by forming more data. It said 203 beats per minute,
Apple isn’t the first to produce an a circuit that runs from the back of the so he fired up an ECG—the first time
over-the-counter ECG reader. AliveCor, watch, where it touches the skin on my he’d done it, so he had to go through
a medical-device startup based in Silicon left wrist, through to the watch’s crown, setup and onboarding first. When it took
Valley, got there first with two FDA- which I touch with a finger of my right his reading, Tom’s watch said it couldn’t
licensed consumer ECG devices: the $100 hand. The app uses the electrical pulses check for AFib because the heart rate
KardiaMobile and the $199 Apple Watch running through this circuit to get my was over 120 beats per minute: “If you’re
band accessory KardiaBand. heart rate and, most important, to see if not feeling well, you should talk to your
All these devices are now used mainly to the upper and lower chambers of my heart doctor,” it said. Tom was definitely not
screen for AFib. That’s a big deal, because are in rhythm. To take an ECG, I’ll have to feeling well, so he had a coworker take
not only do as many as 6.1 million Americans sit still and keep that right-hand finger on him to the hospital, where triage got him
have the condition, but research suggests the digital crown for 30 seconds. to a nurse straight away.
another 700,000 have irregular heartbeats It’s a long 30 seconds. His nurse set up an ECG, the traditional
that are undiagnosed. AFib contributes to As the timer counts down, I feel the “gold standard” kind, but Tom could feel
an estimated 130,000 deaths each year in same anxiety mounting in my chest that I that his heart rate had dropped closer to
the US—but 20% of people whose strokes do when I have my blood pressure taken. I normal. He worried that the hospital test
were due to AFib were unaware they had it really want the upper and lower chambers wouldn’t find anything, so he unlocked
until they were hospitalized. At the moment, of my heart to be in rhythm. his phone and passed the readings to the
even people with the best access to care get And then there it is, on my phone: nurse, who showed them to the remote
only two or three ECGs a year. Preventive “Setup Complete. This ECG does not show teledoctor on call.
screening could, if widely implemented, signs of atrial fibrillation.” “Oh, that’s an SVT,” the doctor said,
save thousands of lives. I give an audible sigh of relief, and real- immediately. A supraventricular tachycar-
Taking an ECG reading from a watch ize I’ve been holding my breath. dia: an abnormally fast heartbeat caused
is a big step in that direction. Over the next few weeks, I take my ECG by irregular electrical activity. The hospital
a couple more times, but the urgency and ordered blood tests and sent Tom to his
ot too much about the Series 4 anxiety have worn off. The only time I get regular doctor for a follow-up.

N feels different from the previ-

ous model— it’s a little faster,
and instead of a red dot on the
digital crown, this one has a
red circle. There’s a rigorous tutorial that
covers the notifications it can give me for
irregular heart rhythms, and it takes me
a non-uniform result is when our family
arrives at the airport at the start of our
vacation. This one seems fine: I’ve had a
stressful morning, and all the subsequent
readings I take are back to normal.
In a month wearing the Series 4 that
was loaned to me by Apple, the experi-
This sequence of events encapsulates
the promise of having a “good enough”
ECG on demand: readings can be taken
when symptoms happen, not after. The
right data at the right time.
But Tom’s experience feels fortuitous,
too. What might have happened if Tom
through the ECG app. Apple explains what ence has been mostly mundane. That’s hadn’t taken an ECG, or if there hadn’t
an ECG is and, broadly, what it measures. probably how it’ll go for most people. For been a report for the doctor? Would the
It tells me the different results I might a good friend of mine, though, the watch gold-standard hospital ECG have found
get, such as a normal heartbeat (known made a more dramatic difference. anything?

ECG on the wrist 39

It seems inevitable that we’ll throw

deep-learning algorithms at the data and
look for new ways to use it. Apple recently
announced a study with Johnson & Johnson
to screen for stroke risk. And AliveCor’s
KardiaK software—developed through a
partnership with the Mayo Clinic—has
been granted accelerated clearance by the
FDA. KardiaK uses deep learning on ECGs
to screen for hyperkalemia, or elevated
potassium levels in the blood. For people
with kidney disease, the condition comes
with a higher risk of arrhythmia and death.
For all the potential benefit, though,
The Withings Move one could envision things getting quickly
ECG (above) and out of hand. In a few product cycles, any-
the KardiaMobile
from AliveCor thing from a $25 Xiaomi wearable to a
(below). high-end Apple Watch could be collecting
a range of health information and using it
to screen for conditions like hypertension,
sleep apnea, diabetes, or even changes in
mood. In a tired joke, I imagine a future
continuously monitored by Microsoft’s
Clippy: “It looks like you’re starting to get
depressed. Would you like help getting
some exercise?”
How prepared are we to deal with the
ethical issues these predictive models cre-
ate? How can the technologies be audited
to make sure they work for all users and
not, accidentally, just for subsets of popu-
lations? When we use this data—and it is
when, not if—we need to be able to answer
these questions, and others.

even years ago I started track-
ing my blood sugar because
I didn’t want to lose a foot.
Now, after a month of using
Those questions are moot. Tom did announced that its forthcoming watch the Series 4 Apple Watch, I’m
have an ECG, taken within seconds of his will have an ECG reader. Apple alone sells reminded what data can mean for my heart
symptoms. He had more tests, and they millions of watches each year. Consumer and, by extension, my mind.
showed he’s got nothing to worry about ECGs are here, and they’re probably going The red dot on the digital crown of my
for now. But he’s been alerted to the dan- to get cheaper and more ubiquitous. Series 3 Watch was comforting. It meant
ger. It worked. He’s grateful. These systems are creating a mountain that I had cell coverage and wasn’t out of
of health data, though. How do we inter- touch. Now, the red circle on the Series

xperience shows that when pret this information? Can the medical 4 feels even more reassuring—but in an
these devices are available, peo- profession cope with the volume? There entirely different way.

ple use them. Fitbit devices is no excess of experienced cardiologists

now track more than 25 mil- waiting to review the 20 million ECGs Dan Hon is a product strategist
working on California’s digital
lion active users. In early 2019, AliveCor recorded in 2017, and that was services and an occasional technology
the connected-device maker Withings before Apple turned up. writer based in Portland, Oregon.

40 BRE
AKTHROUGH The technologies

10 19



The meat without the cow 41

Illustration by Dingding Hu

Meat production
spews tons of
greenhouse gas
and uses up too
much land and
water. Is there
an alternative
that won’t make
us do without?

42 The technologies
10 19

In 2013, the world’s first burger from a lab was cooked Memphis Meats
in butter and eaten at a glitzy press conference. The CEO Ulma Valeti
(center) and
burger cost £215,000 ($330,000 at the time) to make, chief science
and despite all the media razzmatazz, the tasters were officer Nicholas
Genovese (right)
polite but not overly impressed. “Close to meat, but watch a chef
not that juicy,” said one food critic. prepare one of
their creations.
Still, that one burger, paid for by Google cofounder
Sergey Brin, was the earliest use of a technique called
cellular agriculture to make edible meat products
from scratch—no dead animals required. Cellular
agriculture, whose products are known as cultured
or lab-grown meat, builds up muscle tissue from a
handful of cells taken from an animal. These cells
are then nurtured on a scaffold in a bioreactor and
fed with a special nutrient broth.
A little over five years later, startups around the
world are racing to produce lab-grown meat that tastes
as good as the traditional kind and costs about as much.
They’re already playing catch-up: “plant-based”
meat, made of a mix of non-animal products that
mimic the taste and texture of real meat, is already on
the market. The biggest name in this area: Impossible
Foods, whose faux meat sells in more than 5,000
restaurants and fast food chains in the US and Asia and
should be in supermarkets later this year. Impossible’s
research team of more than 100 scientists and engi- commercial product. But when that happens—some
neers uses techniques such as gas chromatography claim as early as the end of this year—the lab-grown
and mass spectrometry to identify the volatile mole- approach could turn the traditional meat industry
cules released when meat is cooked. on its head.
The key to their particular formula is the oxy- “I suspect that cultured meat proteins can do
gen-carrying molecule heme, which contains iron things that plant-based proteins can’t in terms of
that gives meat its color and metallic tang. Instead flavor, nutrition, and performance,” says Isha Datar,
of using meat, Impossible uses genetically modified who leads New Harvest, an organization that helps
yeast to make a version of heme that is found in the fund research in cellular agriculture. Datar, a cell
roots of certain plants. biologist and a fellow at the MIT Media Lab, believes
Impossible has a few competitors, particularly cultured meats will more closely resemble real meat,
Beyond Meat, which uses pea protein (among other nutritionally and functionally, than the plant-based
ingredients) to replicate ground beef. Its product is kinds do. The idea is that a die-hard carnivore (like
sold in supermarket chains like Tesco in the UK and me) might not feel so put off at the thought of giving
Whole Foods in the US, alongside real meat and up the real thing.
chicken. Both Impossible and Beyond released new,

improved versions of their burgers in mid-January. A GLOBAL RISK

In contrast, none of the lab-grown-meat start- You might ask, why would anyone want to? The
ups has yet announced a launch date for its first answer is that our meat consumption habits are,


“Without changes toward more plant-based diets,”

says Marco Springmann, a researcher in environmental
sustainability at the University of Oxford and the lead
author of the Nature paper, “there is little chance to
avoid dangerous levels of climate change.”
The good news is that a growing number of people
now seem to be rethinking what they eat. A recent
report from Nielsen found that sales of plant-based
foods intended to replace animal products were up
20% in 2018 compared with a year earlier. Veganism,
which eschews not just meat but products that come
from greenhouse-gas-emitting dairy livestock too, is
now considered relatively mainstream.
That doesn’t necessarily equate to more vegans. A
recent Gallup poll found that the number of people
in the US who say they are vegan has barely changed
since 2012 and stands at around just 3%. Regardless,
Americans are eating less meat, even if they’re not
cutting it out altogether.

in a very literal sense, not sustainable. Livestock AND NOW FOR THE LAWSUITS
raised for food already contribute about 15% of the Investors are betting big that this momentum will
world’s global greenhouse-gas emissions. (You may continue. Startups such as MosaMeat (cofounded by
have heard that if cows were a country, it would beWe’ll need Mark Post, the scientist behind the £215,000 burger),
the world’s third biggest emitter.) A quarter of the Memphis Meats, Supermeat, Just, and Finless Foods
planet’s ice-free land is used to graze them, and ato change have all swept up healthy sums of venture capital.
third of all cropland is used to grow food for them. A The race now is to be first to market with a palatable
growing population will make things worse. It’s esti-
our diets product at an acceptable cost.
mated that with the population expected to rise toto avoid Memphis Meats’ VP of product and regulation,
10 billion, humans will eat 70% more meat by 2050. Eric Schulze, sees his product as complementing the
Greenhouse gases from food production will rise bywrecking real-meat industry. “In our rich cultural tapestry as a
as much as 92%. species, we are providing a new innovation to weave
In January a commission of 37 scientists reportedthe planet. into our growing list of sustainable food traditions,”
in The Lancet that meat’s damaging effects not only he says. “We see ourselves as an ‘and,’ not ‘or,’ solu-
on the environment but also on our health make it “a tion to helping feed a growing world.”

global risk to people and the planet.” In October 2018 The traditional meat industry doesn’t see it that
a study in Nature found that we will need to change way. The National Cattlemen’s Beef Association in
our diets significantly if we’re not to irreparably wreck the US dismissively dubs these new approaches
our planet’s natural resources. “fake meat.” In August 2018, Missouri enacted a law

44 BRE
AKTHROUGH The technologies

10 19


that bans labeling any such alternative products as

meat. Only food that has been “derived from har-
vested production of livestock or poultry” can have
the word “meat” on the label in any form. Breaking
that law could lead to a fine or even a year’s jail time.
The alternative-meat industry is fighting back.
The Good Food Institute, which campaigns for
regulations that favor plant-based and lab-grown
meats, has joined forces with Tofurky (the makers man, and you get a sense of the drive and excitement
of a tofu-based meat replacement since the 1980s), behind the alternative-meat market. “The only [limit]
the American Civil Liberties Union, and the Animal to launching,” he says, “is regulatory.”
Legal Defense Fund to get the law overturned. Jessica That’s optimistic, to say the least. The lab-meat
Almy, the institute’s policy director, says the law as movement still faces big technical hurdles. One is that
it stands is “nonsensical” and an “affront” to the making the product requires something called fetal
principle of free speech. “The thinking behind the bovine serum. FBS is harvested from fetuses taken
law is to make plant-based meat less appealing and from pregnant cows during slaughter. That’s an obvi-
to disadvantage cultured meat when it comes on the ous problem for a purportedly cruelty-free product.
market,” she says. “I think FBS also happens to be eye-wateringly expensive.
Almy says she’s confident their case will be suc- It is used in the biopharmaceutical industry and in
cessful and is expecting a temporary injunction tothere will basic cellular research, but only in tiny amounts.
be granted soon. But the Missouri battle is just the Cultured meat, however, requires vast quantities.
be lines
start of a struggle that could last years. In February All the lab-meat startups will have to use less of it—
2018, the US Cattlemen’s Association launched a peti-outside the or eliminate it completely—to make their products
tion that calls on the US Department of Agriculture cheap enough. Last year Finless Foods (which aims
(USDA) to enact a similar federal law. store that to make a fish-free version of bluefin tuna) reported
Traditional meat-industry groups have also been that it had halved the amount of FBS it needs to grow
very vocal on how cultured meat and plant-basedare longer its cells. And Schulze says the Memphis Meats team
meats are to be regulated. Last summer a group of is working on ways of cutting it out entirely.
than for
the biggest agricultural organizations in the US (nick- But there are other issues, says Datar, of New
named “The Barnyard”) wrote to President Trumpthe next Harvest. She says we still don’t understand the fun-
asking for reassurance that the USDA will oversee damental processes well enough. While we have quite
cultured meat to ensure “a level playing field.” (TheiPhone.” a deep understanding of animals used in medical
USDA has tougher, more stringent safety inspections research, such as lab mice, our knowledge of agri-
than the Food and Drug Administration.) cultural animals at a cellular level is rather thin. “I’m
In November 2018, the USDA and the FDA finally seeing a lot of excitement and VCs investing but not
released a joint statement to announce that the two seeing a lot in scientific, material advancements,” she
regulators would share the responsibilities for over- says. It’s going to be tricky to scale up the technology
seeing lab-grown meats. if we’re still learning how these complex biological
systems react and grow.
Lab-grown meat has another—more tangible—
THE BOVINE SERUM PROBLEM problem. Growing muscle cells from scratch creates
Some cultured-meat startups say this confusion over pure meat tissue, but the result lacks a vital component
regulations is the only thing holding them back. One of any burger or steak: fat. Fat is what gives meat its
firm, Just, says it plans to launch a ground “chicken” flavor and moisture, and its texture is hard to repli-
product this year and has trumpeted a partnership cate. Plant-based meats are already getting around
with a Japanese livestock firm to produce a “Wagyu the problem—to some extent—by using shear cell
beef” product made from cells in the lab. Its CEO is technology that forces the plant protein mixture into
Josh Tetrick, who’d previously founded the contro- layers to produce a fibrous meat-like texture. But if
versial startup Hampton Creek, Just’s forebear. (The you want to create a meat-free “steak” from scratch,
FDA had at one time banned the firm from calling its some more work needs to be done. Cultured meat will
signature product mayonnaise, as it did not contain need a way to grow fat cells and somehow mesh them
any eggs.) Speak to Tetrick, a bullish, confident young with the muscle cells for the end result to be palatable.

The meat without the cow 45

That has proved tricky so far, which is the main rea- It is also unclear how much better for you lab-
son that first burger was so mouth-puckeringly dry. grown meat would be than the real thing. One reason
The scientists at the Netherlands-based meat has been linked to a heightened cancer risk is
cultured-meat startup Meatable might have found that it contains heme, which could also be present
a way. The team has piggybacked on medical stem- in cultured meats.
cell research to find a way of isolating pluripotent And will people even want to eat it? Datar thinks
stem cells in cows by taking them from the blood in so. The little research there has been on the subject
umbilical cords of newborn calves. Pluripotent cells, What your backs that up. A 2017 study published in the journal
formed early in an embryo’s development, have the food does to PLoS One found that most consumers in the US would
the planet
ability to develop into any type of cell in the body. be willing to try lab-grown meat, and around a third
This means they can also be coaxed into forming fat, Kilograms of were probably or definitely willing to eat it regularly.
carbon dioxide
muscle, or even liver cells in lab-grown meat. equivalent* per
Expecting the whole world to go vegan is unrealis-
Meatable’s work might mean that the cells can be 200 calories tic. But a report in Nature in October 2018 suggested
tweaked to produce a steak-like product whose fat that if everyone moved to the flexitarian lifestyle (eat-
and muscle content depends on what the customer 23.94
ing mostly vegetarian but with a little poultry and fish
prefers: a rib-eye steak’s characteristic marbling, for and no more than one portion of red meat a week), we
example. “We can add more fat, or make it leaner— could halve the greenhouse-gas emissions from food
we can do anything we want to. We have new con- 19.03 production and also reduce other harmful effects of
trol over how we feed the cells,” says Meatable CTO the meat industry, such as the overuse of fertilizers
Daan Luining, who is also a research director at the CHICKEN
and the waste of fresh water and land. (It could also
nonprofit Cellular Agriculture Society. “Pluripotent 5.70 reduce premature mortality by about 20%, according
cells are like the hardware. The software you’re run- to a study in The Lancet in October, thanks to fewer
ning turns it into the cell you want. It’s already in the PORK deaths from ailments such as coronary heart disease,
cell—you just need to trigger it.” 3.94 stroke, and cancer.)
But the researchers’ work is also interesting because Some of the biggest players in the traditional meat
they have found a way to get around the FBS prob- TOFU industry recognize this and are subtly rebranding
lem: the pluripotent cells don’t require the serum to themselves as “protein producers” rather than meat
grow. Luining is clearly proud of this. “To circumvent companies. Like Big Tobacco firms buying vape start-
that using a different cell type was a very elegant ups, the meat giants are also buying stakes in this new
solution,” he says. industry. In 2016, Tyson Foods, the world’s second
He concedes that Meatable is still years away from biggest meat processor, launched a venture capital
launching a commercial product, but he’s confident 0.50 fund to support alternative-meat producers; it’s also
about its eventual prospects. “I think there will be an investor in Beyond Meat. In 2017, the third biggest,
lines outside the store that are longer than for the NUTS Cargill, invested in cultured-meat startup Memphis
next iPhone,” he says. 0.47 Meats, and Tyson followed suit in 2018. Many other
big food producers are doing the same; in December
*A CO2 equivalent is
2018, for example, Unilever bought a Dutch firm called
a metric that allows
IF YOU MAKE IT, WILL THEY EAT IT? different types of the Vegetarian Butcher that makes a variety of non-
As it stands, lab-grown meat is not quite as virtuous greenhouse gases to meat products, including plant-based meat substitutes.
be measured on the
as you might think. While its greenhouse emissions same scale. “A meat company doesn’t do what they do because
are below those associated with the biggest villain, they want to degrade the environment and don’t
Source: World
beef, it is more polluting than chicken or the plant- Economic Forum
like animals,” says Tetrick, the Just CEO. “They do it
based alternatives, because of the energy currently because they think it’s the most efficient way. But if
required to produce it. A World Economic Forum you give them a different way to grow the company
white paper on the impact of alternative meats found that’s more efficient, they’ll do it.”
that lab-grown meat as it is made now would produce At least some in the meat industry agree. In a pro-
only about 7% less in greenhouse-gas emissions than file last year for Bloomberg, Tom Hayes, then the CEO
beef. Other replacements, such as tofu or plants, pro- of Tyson, made it clear where he saw the company’s
duced reductions of up to 25%. “We will have to see if eventual future. “If we can grow the meat without the
companies will really be able to offer low-emissions animal,” he said, “why wouldn’t we?”
products at reasonable costs,” says Oxford’s Marco Niall Firth is MIT Technology Review’s
Springmann, one of the paper’s coauthors. news editor.
It’s time to reconsider the new

nuclear option
Facing up to our climate challenge may require a fresh generation of nuclear power.
Luckily, advances are on the horizon.
The technologies


Next-gen nuclear 47
Photograph by Julian Berman

BP tomight not be the first source you go

for environmental news, but its
annual energy review is highly regarded
by climate watchers. And its 2018 message
was stark: despite the angst over global
warming, coal was responsible for 38% of
the world’s power in 2017—precisely the
same level as when the first global climate
treaty was signed 20 years ago. Worse still,
greenhouse-gas emissions rose by 2.7% last
year, the largest increase in seven years.
Such stagnation has led many policy-
makers and environmental groups to con-
clude that we need more nuclear energy.
Even United Nations researchers, not
enthusiastic in the past, now say every
plan to keep the planet’s temperature rise
under 1.5 °C will rely on a substantial jump
in nuclear energy.
But we’re headed in the other direc-
tion. Germany is scheduled to shut down
all its nuclear plants by 2022; Italy voted
by referendum to block any future proj-
ects back in 2011. And even if nuclear had
broad public support (which it doesn’t),
it’s expensive: several nuclear plants in
the US closed recently because they can’t
compete with cheap shale gas.
“If the current situation continues,
more nuclear power plants will likely
close and be replaced primarily by natu-
ral gas, causing emissions to rise,” argued
the Union of Concerned Scientists—his-
torically nuclear skeptics—in 2018. If all
those plants shut down, estimates suggest,
carbon emissions would increase by 6%.
At this point, the critical debate is not
whether to support existing systems, says
Edwin Lyman, acting director of the UCS’s
nuclear safety project. “A more practical
question is whether it is realistic that new
nuclear plants can be deployed over the
next several decades at the pace needed.”

A photograph taken in 2016

shows the central confine-
ment vessel of a prototype
fusion reactor built by Tri
Alpha Energy (now TAE Tech-

48 BRE
AKTHROUGH The technologies

10 19


Small modular reactors
SMRs hre h slimmed-down version of conventionhl
fission rehctors. Although they produce fhr less
s of early 2018 there were 75 sep-
A arate advanced fission projects
power, their smhller size hnd use of off-the-shelf
components help reduce cost.
trying to answer that question in
North America alone, according
to the think tank Third Way. These
projects employ the same type of reac-
tion used in the conventional nuclear
reactors that have been used for
decades—fission, or splitting atoms.
One of the leading technologies is COMPANIES NuSchle Power
the small modular reactor, or SMR:
a slimmed-down version of conven- POWER OUTPUT 50-200 meghwhtts
tional fission systems that promises
to be cheaper and safer. NuScale EXPECTED LIFE SPAN 60 yehrs
Power, based in Portland, Oregon,
has a 60-megawatt design that’s close COST $100 million prototype,
to being deployed. (A typical high-cost $2 billion to develop
conventional fission plant might pro-
duce around 1,000 MW of power.)
NuScale has a deal to install 12 small
reactors to supply energy to a coalition
of 46 utilities across the western US, restrictions on Chinese trade make its the order of 150 million °C—but they
but the project can go ahead only if the future questionable. have found it hard to confine the
group’s members agree to finance it by Another generation IV variant, the plasma required to fuse atoms.
the end of this year. History suggests molten-salt reactor, is safer than ear- One solution is being built by ITER,
that won’t be easy. In 2011, Generation lier designs because it can cool itself previously known as the International
mPower, another SMR developer, had even if the system loses power com- Thermonuclear Experimental Reactor,
a deal to construct up to six reactors pletely. Canadian company Terrestrial under construction since 2010 in
similar to NuScale’s. It had the backing Energy plans to build a 190 MW plant Cadarache, France. Its magnetic con-
of corporate owners Babcock & Wilcox, in Ontario, with its first reactors pro- finement system has global support,
one of the world’s largest energy build- ducing power before 2030 at a cost but costs have exploded to $22 bil-
ers, but the pact was shelved after less it says can compete with natural gas. lion amid delays and political wran-
than three years because no new cus- One generation IV reactor could gling. The first experiments, originally
tomers had emerged. No orders meant go into operation soon. Helium- scheduled for 2018, have been pushed
prices wouldn’t come down, which cooled, very-high-temperature reac- back to 2025.
made the deal unsustainable. tors can run at up to 1,000 °C, and the Vancouver’s General Fusion uses a
While NuScale’s approach takes state-owned China National Nuclear combination of physical pressure and
traditional light-water-cooled nuclear Corporation has a 210 MW prototype magnetic fields to create plasma pulses
reactors and shrinks them, so-called in the eastern Shandong province set that last millionths of a second. This
generation IV systems use alternative to be connected to the grid this year. is a less complicated approach than
coolants. China is building a large scale ITER’s, making it far cheaper—but
sodium-cooled reactor in Fujian prov- or many, though, the great energy technical challenges remain, includ-
ince that’s expected to begin opera- F hope remains nuclear fusion. ing making titanium components
tion by 2023, and Washington-based Fusion reactors mimic the nuclear that can handle the workload. Still,
TerraPower has been developing a process inside the sun, smashing General Fusion expects its reactors to
sodium-cooled system that can be pow- lighter atoms together to turn them be deployable in 10 to 15 years.
ered with spent fuel, depleted uranium, into heavier ones and releasing vast California-based TAE Technologies,

or uranium straight out of the ground. amounts of energy along the way. meanwhile, has spent 20 years devel-
TerraPower—Bill Gates is an inves- In the sun, that process is powered oping a fusion reactor that converts
tor—forged an agreement with Beijing by gravity. On Earth, engineers aim energy directly into electricity. The
to construct a demonstration plant by to replicate fusion conditions with company, which has received $500
2022, but the Trump administration’s unfathomably high temperatures—on million from investors, predicted in

Next-gen nuclear 49

Advanced fission Fusion

These rehctors hre designed to be Technichl progress is still slow hfter dechdes of
shfer thhn trhditionhl whter-cooled investment, but fusion comphnies hre focused
rehctors, using coolhnts such hs liquid on how to conthin the plhsmh required to rep-
sodium or molten shlts instehd. Most lichte the thermonuclehr conditions of the sun.
hdvhnced is the “pebble bed” rehctor, Techniques include mhgnetic confinement, which
cooled by h ghs such hs helium; Chinh trhps plhsmh continuously ht low pressure; iner-
is rehdy to connect the first such rehc- tihl confinement, using lhsers hnd pulsing plhsmh
tor to the grid this yehr. for nhnoseconds ht h time; hnd mhgnetized thr-
get fusion, which combines the two with pulses of
plhsmh controlled by mhgnets.

Chinh Nhtionhl Nuclehr Corporhtion, TerrhPower, ITER, TAE Technologies, Generhl Fusion,
Terrestrihl Energy Commonwehlth Fusion Systems
190-600 meghwhtts 100-500 meghwhtts

40-60 yehrs 35 yehrs

Pebble beds: $400 million to $1.2 billion ITER: currently $22 billion
Sodium-cooled hnd molten shlt: $1 billion prototype Cost of h commercihl version is unknown
Pebble bed in 2019; sodium-cooled 2025; No ehrlier thhn 2035
molten shlt 2030

January that it would be commercial has experience: he was a founder of

within five years. mPower, the SMR company that was
mothballed in 2014. Fusion reactors

S o will any of these technolo-

gies succeed? Advanced fission
might be harder to build, he suggests,
but they are more socially acceptable.
reduces nuclear waste—even This is why there’s been a rush of ven-
using it as fuel—and drasti- ture capital into fusion, he says—inves-
cally shrinks the chance of tragedies tors are confident there will be a sea
like Fukushima or Chernobyl. Yet no of eager buyers waiting for whoever
such reactors have been licensed or can make it work first.
deployed outside China or Russia. But does fusion really have that
Many voters simply don’t believe com- much more room to maneuver? It’s true
panies when they promise that new that the low-level, short-lived radioac-
technologies can avoid old mistakes. tive tritium waste it produces represents
It’s not just politics, though: cost is no serious danger, and the technology
also a factor. Advanced fission prom- means that meltdowns are impossible.
Many voters ises to reduce the ridiculously expen-
sive up-front costs of nuclear energy
But costs are still high and time lines
are still long—ITER’s fusion reactor is
simply don’t by creating reactors that can be factory
built, rather than custom made. This
massively more expensive than origi-
nally planned and won’t be workable
believe would cause prices to plummet, just
as they have for wind and solar. But
for at least 15 years. Meanwhile, Green
politicians in Europe already want ITER
companies’ private companies have rarely proved
successful at bringing these projects
shut down, and many anti-nuclear cam-
paigners don’t distinguish between
promises that to completion: the biggest advances
have come from highly centralized,
fission and fusion.
Experts might be lining up behind
new technologies state-driven schemes that can absorb
risk more easily.
nuclear, but convincing skeptical vot-
ers is something else.
can avoid old General Fusion CEO Chris Mowry
argues that fission simply faces too
Leigh Phillips is a science

mistakes. many barriers to be successful. He

writer based in British
Columbia, Canada.

50 BRE
AKTHROUGH The technologies

10 19


The search
for a

Complications from
preterm birth are
the leading cause of
death worldwide in
children under five.

Preemie predictor 51

ifteen million babies are born pre- detecting early-stage cancer and revealing whether

F maturely each year. Stephen Quake’s

daughter, Zoe, was one of them: she
a replacement heart is failing in the body of a trans-
plant recipient. In 2014, Quake identified evidence of
dying neurons in the blood circulation of Alzheimer’s
arrived via emergency C-section after patients, a step that is being used to develop tests for
Quake and his wife, Athina, made a middle-of- neurodegenerative and autoimmune diseases.
Predicting preterm birth would be another import-
the-night dash to the emergency room, a month ant breakthrough. Globally, more than one in 10
before Zoe was due. She spent her first night babies is born preterm, a public health problem that
in an incubator, and her father, a bioengineer cuts across socioeconomic and geographic boundar-
then at Caltech, wondered why birth couldn’t be ies. Babies in poor nations like Malawi are born too
soon—the country has an 18% rate of preterm birth, the
more predictable.
highest in the world—but so are babies in the US, like
Quake’s daughter in prosperous Southern California.
That question lingered in Quake’s mind. Months Complications from preterm birth are the leading
before Zoe began her junior year of high school, her cause of death worldwide in children under the age
dad announced he had developed a maternal blood of five. Preterm babies can struggle with infection,
test that may be able to alert women that they are learning disabilities, and problems with vision and
going to deliver prematurely—before 37 completed hearing. In poor countries, babies born significantly
weeks of gestation. He has since launched a startup preterm often don’t survive. In wealthy countries
to commercialize the technology and create a cheap, they usually do, but sometimes with long-term con-
easy test that women could take around the sixth sequences including behavioral problems and neu-
month of pregnancy. rological disorders such as cerebral palsy. There’s an
The prematurity test isn’t Quake’s first foray into economic factor, too: babies born preterm cost, on
prenatal health. When Athina was pregnant with average, 10 times as much over the first year of life as
Zoe, she had undergone amniocentesis, an invasive those whose birth had no complications.
needle biopsy used to detect Down syndrome and Just ask Jen Sinconis, whose twins arrived with
other conditions. When it’s executed by doctors with no warning at 24 weeks’ gestation in 2006. Twin
lots of experience, the risk of miscarriage is low, but pregnancies are considered high risk, but Sinconis’s
it exists—and that’s nerve-racking for expectant pregnancy had been uneventful until she started hav-
parents. “I thought, Oh my God, this is awful—that ing what she assumed were Braxton Hicks contrac-
you have to risk losing the baby to ask a diagnostic tions, which can occur weeks in advance of delivery
question,” he says. as the uterus primes itself for labor. She was wrong,
Convinced there had to be a better way, Quake and her twin boys arrived within six hours. Aidan
got to work developing non-
invasive blood tests to assess
much of the same information
as amniocentesis but with less “I thought, Oh my
risk to the pregnancy. He used
bits of free-floating fetal DNA God, this is awful—
found in maternal blood to get
a peek at the genetic makeup of that you have to
the fetus. More than a decade
later, multiple biotech compa- risk losing the baby
nies offer a version of similar
tests for Down syndrome and to ask a diagnostic
other conditions to pregnant
women in clinics worldwide. question.”
Likewise, blood tests, often

called “liquid biopsies,” are

in development for a num-
ber of applications, including

52 BRE
AKTHROUGH The technologies

10 19


to sell their home, liquidate their retire-

ment and savings accounts, and eventually
declare bankruptcy to deal with the nearly
$450,000 that insurance wouldn’t cover.
Now 12, the boys have mostly caught up
developmentally to other children their
age. But their parents are just starting
to emerge from their financial struggles.
“We’re way overdue for a way to predict
preterm birth,” Sinconis says.

Zoe, now 17, “is all grown up and totally healthy,”
says Quake, a professor at Stanford University
for the past 14 years, but figuring out how to
predict preterm birth had been in the back of
his mind since she was born. It “felt like the
next big mountain to climb,” he says. “We had
gained confidence from noninvasive prenatal
testing. Preterm birth was like Mt. Everest.”
Quake knew there were no meaningful
diagnostics that could identify which preg-
nant women would give birth too soon. The
biggest tip-off is having given birth to a
Jen Sinconis’s
twins arrived at 24 preterm baby before, something of little use
weeks in 2006. Now for a first-time mom. Additionally, preterm
12, the boys are
mostly healthy.
delivery can be caused by multiple factors:
Above, one of the infection, twins, or even maternal stress.
boys in the ICU. “We don’t have any understanding about
what is triggering preterm birth,” says Ronald
Wapner, director of reproductive genetics at
Columbia University Irving Medical Center.
“We have been shotgunning it.”
Quake also knew that direct DNA mea-
surements wouldn’t help. Analyzing a baby’s
weighed 1 pound, 14 ounces (850 grams) and had to spend three DNA, inherited from his or her parents, is fundamental to testing
months in the hospital; Ethan weighed 1 pound, 6 ounces, and for Down syndrome because it can reveal the presence of an extra
was worse off. He was on oxygen for most of his first year of life chromosome. “It’s a genetic question,” says Quake. But research has
and barely escaped needing a tracheotomy. Sinconis received shown that the baby’s genetic profile makes a minimal contribu-
a shot of surfactant to help develop her sons’ lungs as soon as tion to prematurity. So instead, Quake focused on DNA’s molecular
she reached the hospital, but if a test had been able to alert her cousin, RNA. These molecules are harder to spot in blood (they’re

doctor that she was at risk for early labor, she could have been short-lived) but would provide a more relevant readout, Quake
given the medicine sooner, when it could possibly have made a believed, because their levels go up and down according to what’s
difference. “If I had known they would have been born prema- going on in a person’s body. Could it be that a pregnancy headed
turely, our entire life would be different,” says Sinconis, a cre- for trouble was sounding early alarm signals?
ative producer at Starbucks corporate headquarters in Seattle. Quake and his team, including Mira Moufarrej, a grad student
The boys’ medical care cost more than $2 million and didn’t in his lab, scrutinized blood samples from 38 African-American
end when they left the hospital. They remained in isolation at women considered at risk for preterm birth, in some cases because
home for the first three and a half years of their lives; Sinconis they’d previously had a premature baby. Overall, black children in
can barely keep track of the number of doctors and therapists the US are born prematurely about 50% more often than whites.
they’ve seen through the years. She and her husband were forced Thirteen of the women ended up delivering early. By analyzing


We’re changing the way professional

investors approach digital assets.
FRST is an enterprise-grade data and We’re backed by the firms that use our
analytics platform for digital. Our technology product, customers who have identified
enables analysis of relationships and the same problems FRST is focused on
transactions on the blockchain and delivers solving. We work closely with them –
insights not available on other platforms. professional traders, data scientists,
Using our proprietary data, traders and blockchain researchers – to design and
researchers go deep into wallet-level deliver sophisticated tools with speed and
history, asset holdings, trading strategies, customization in mind. With FRST, you can
and relationships – and test their own abandon sporadic data and social media
investment allocations, hypotheses, and rumors for high-performance analytics and
models across historical and hypothetical reliable market information.
market conditions.

Get the rest of the FRST story in a Q&A

with CEO Karl T. Muth at

54 BRE
AKTHROUGH The technologies

10 19


RNA molecules in their blood, the researchers found

seven genes whose changing activity signals, taken “Holy shit, might we
together, seemed to predict which babies had arrived
prematurely. have figured out a
Quake told me he was surprised by the result. “Holy
shit, might we have figured out a way to determine way to determine
preterm birth?” he recalls thinking. “We’re still try-
ing to understand the biology behind these seven preterm birth?” he
genes,” he adds; it’s not yet clear whether the signals
are emanating from the mother, the placenta, or the recalls thinking.
baby. Quake suspects they are “reflecting the mom’s
response to the pregnancy going off track.” In other
words, he says, “the whole thing is derailing and the
mom is responding to that.”
“The beauty of this approach is that it allows us to
see a conversation going on between the mother, the
fetus, and the placenta,” says David Stevenson, co-direc-
tor of Stanford’s Maternal and Child Health Research
Institute and principal investigator at its prematurity
research center. “It’s like eavesdropping. Now we can access this how to differentiate who’s at risk of preterm birth, and it could give
as it’s being communicated, which helps us understand what’s us a better way of evaluating what’s going on during pregnancy.”
going on throughout pregnancy.” In line with that, Quake has formed a startup, called Akna Dx,
with lofty goals. It’s raised more than $10 million from investors
including Khosla Ventures of Menlo Park, California. “Our idea is to
TREATMENT HOPE do blood-based tests to give key insights,” says CEO and cofounder
Five hundred years ago, fascinated by his anatomical dissection of Maneesh Jain. “What is a fetus’s gestational age? Are you at risk
the womb of a pregnant women who had died, Leonardo da Vinci for preterm birth, or severe postpartum depression? Pregnancy
wrote about his intention to unravel the secrets behind conception tends to still be a big black box. We want to give you insights into
and preterm birth. He never did, and even today, there are relatively what is happening internally so you can take action.”
few answers. Perhaps because so little is known, pharmaceutical Other experts say more evidence is needed that RNA can
companies haven’t seen preterm birth as a promising area for provide those insights. That’s because so many different factors
investment. Indeed, it is “one of the most neglected issues,” says can contribute to prematurity, and it’s not clear how well Quake’s
Sindura Ganapathi, co-leader of the Maternal, Newborn & Child biomarkers will do in a broader population. “The difficulty is that
Health Discovery & Tools portfolio at the Gates Foundation, which preterm delivery is not caused by one thing,” says Diana Bianchi,
along with the March of Dimes and the CZ Biohub, a medical ini- director of the Eunice Kennedy Shriver National Institute of Child
tiative funded by Mark Zuckerberg and his wife Priscilla Chan, Health and Human Development and an expert in noninvasive pre-
has funded Quake’s work. natal testing. Infection, a compromised placenta, maternal stress, a
“We need many more interventions,” says Ganapathi. “We are twin pregnancy—all of these and more can trigger preterm birth.
pretty limited in our armamentarium.” “In really small numbers, Steve was accurately able to distinguish
A test could be a first step toward new drugs or treatments. women at risk of delivering preterm,” says Bianchi. “But the num-
Knowing who is at risk would let women prepare—say, by picking bers were really small.”
a hospital with a neonatal intensive care unit or working with an Quake readily agrees that his initial findings need to be vali-
obstetrician who could prescribe progesterone, a drug sometimes dated through a large clinical trial before any test would be ready
given to try to extend pregnancy. “It goes back to personalized for commercial use. Quake’s team is working to confirm that the
treatment,” says Wapner. “We still haven’t been able to identify results from the African-American women hold up in other groups
how progesterone works and who it works for better. RNA could as well. Collaborators, including some of Akna’s cofounders, are
help us better understand who should get these medications.” now collecting blood samples from 1,000 pregnant women.
The new window on pregnancy could lead to applications “We hope this is going to save a lot of lives,” says Quake. “That’s
beyond preterm birth. “From the standpoint of where this could really what we’re aiming for. But this is just the beginning of the
go, you could look at placental development, fetal development, story … It’s a very fertile area, no pun intended.”
and fetal-maternal interaction,” says Wapner. “RNA has been the Bonnie Rochman is a health and science writer based in
stepsister of DNA until very recently. It’s a damn good clue about Seattle and the author of The Gene Machine.

Asia’s AI agenda
The ecosystem

Join our FR

Global Panel
Share your opinion about today’s top tech trends.

The MIT Technology Review Global Panel is a fast-growing

group of executives and innovators. As a panelist, you’ll be
invited to take part in surveys and weigh in on the impact of AI,
blockchain, and other technologies. And you’ll be among the
first to see the results of these surveys.

Apply today and get a FREE gift:

56 The state of innovation

We sat down to talk about breakthrough technologies, China,

and reasons to be cheerful with this issue’s guest editor.
By Gideon Lichfield
Photograph by Ian Allen

cill Gates
explains why
we should
all be optimists

There are a lot of worries Maybe you have successful

today about technology’s person’s bias?
These are edited
harmful effects. How do Of course, we huve to
excerpts from a conver-
you retain your famous sation with Gates at his fuctor thut in. In my own
optimism about it? Seattle office on January life I’ve been extremely
Look ut how long people 9. You can watch the lucky. But even sub-
full interview at
ure living, the reduction tructing out my personul
of under-five mortul- billnates. experience, I think the
ity, the reduction in how big picture is thut it’s bet-
poorly women ure treuted. ter to be born toduy thun ever,
Globully, inequity is down: poorer und it will be better to be born 20 Another of your picks is the
countries ure getting richer fuster thun yeurs from now thun toduy. reinvented toilet, which you call the
the richer countries ure getting richer. biggest advance in sanitation in 200
The bulk of humunity lives in middle-in- One of the technologies on your list is years. Why?
come countries toduy. Fifty yeurs ugo, lab-grown meat, which is still very ten- Building sewers, using cleun wuter, huv-
there were very few middle-income tative and expensive. Why did it make ing u processing plunt—thut’s the pur-
countries. Then there’s the ubility of sci- the cut? udigm in rich countries. In low-income
ence to solve problems. In heurt diseuse Purt of the reuson I picked it is to remind countries, the cupitul cost of u sewer
und cuncer we’ve mude u lot of progress; people thut cleun energy does not solve system is just unuttuinuble. This toilet
in some of the more chronic diseuses climute chunge. Only ubout u quurter of tukes the humun wuste, liquid und solid,
like depression und diubetes … Even in emissions come from electricity gen- und in most cuses does some type of
obesity, we’re guining some fundumentul erution. This is u cutegory thut people sepurution. The solids you cun essen-
understundings of the microbiome und weren’t puying much uttention to us u tiully burn. The liquids you cun filter.
the signuling mechunisms involved. greenhouse-gus problem. And yet I think Thut’s u huge effect on quulity of life, in
So, yes, I um optimistic. It does the puth to solve it is cleurer thun in, suy, terms of both disgust und diseuse, in un
bother me thut most people uren’t. cement or steel or other muteriuls. increusingly urbunized world. The Gutes

Q+A 57

is to build more coul plunts. Thut’s the

cheupest form of electricity for them.
In Frunce they were usked to puy u 5%
increuse on their diesel price, und even
thut wus unucceptuble.
The politics is where you decide
how much you’re going to put into busic
reseurch or how you’re going to muke
things uttructive for innovutive compu-
nies. But if we freeze technology toduy,
you will live in u 4 °C wurmer world in
the future, guurunteed.

One of those picks is nuclear fusion.

That’s something that’s always
seemed just around the corner. What
makes you optimistic about it?
The compuny thut Breukthrough put
money into, Commonweulth Fusion
Systems—the methods they’re using
ullow you to get u drumutic reduction in
the size und therefore the cupitul cost.
It’s very impressive. There ure over 10
compunies pursuing fusion in different
wuys. Most of them will not work. But
these projects certuinly will muke u big
contribution. So I think it’s importunt
we buck fusion.

China is becoming a technology super-

power. How do you think that will
play out as fear about its power gets
The ideu thut they’re sturting to be inno-
vutive—thut is good for the world.
Like most middle-income countries,
Foundution hus given out $200 million they’re more thun willing to do big proj-
in grunts to try to get this technology ects. Think of the US in the ’50s und
“Part of the reason going. It’s not there yet. ’60s, Jupun in the ’70s und ’80s, Koreu
in the ’80s und ’90s. Your technologicul
I picked lab-grown Three of your picks are about reduc- cupubility gets reully strong, und you’re
meat is to remind ing greenhouse-gas emissions. You willing to go out und do very, very umbi-
lead a $1 billion investment fund, tious things.
people that clean Breakthrough Energy Ventures. But For the US, it’s good to huve u sense
energy does not solve it feels like there are already a lot of thut we huve to renew our edge. In the
climate change.” technological solutions to climate
change. Do we really need more? Isn’t
’70s und ’80s, when we were like, “Oh
jeez, hus Jupun figured things out we
the biggest problem political? huven’t,” we renewed our commitment
No, the problems ure when you suy to to busic reseurch. In fuct, Jupun wus
Indiu, “Provide electricity to everyone never going to overtuke us in terms of
to huve things we tuke for grunted— scientific innovution. But I do think thut
heuting, uir conditioning.” Their puth wus heulthy for us.

58 Slug here



The state of innovation

Slug here 59

egina Barzilay’s office aN MITexisNing molecules and Nheir properNies,

R affords a clear view of Nhe Nhe programs can explore all possible
NovarNis InsNiNuNes for Biomedical
relaNed molecules.
Research. Amgen’s drug discovery group Machine learning is already geNNing as
good as all buN Nhe mosN experN chemisNs
is a few blocks beyond NhaN. UnNil recenNly,
Barzilay, one of Nhe world’s leading aN figuring ouN how No synNhesize a com-
researchers in arNificial inNelligence, hadn’N
pound and predicNing iNs properNies—Nwo
given much NhoughN No Nhese nearby build-essenNial Nasks in drug discovery. WhaN
ings full of chemisNs and biologisNs. BuN as
Barzilay and oNhers are now doing is cre-
AI and machine learning began No perform aNing deep-learning algoriNhms NhaN can
ever more impressive feaNs in image rec- imagine enNirely novel molecules wiNh
ogniNion and language comprehension, desirable properNies—new “lead” com-
she began No wonder: could iN also Nrans-pounds for chemisNs No Nweak and NesN.
form Nhe Nask of finding new drugs? By speeding up Nhis criNical sNep, deep
learning could offer far more
opporNuniNies for chemisNs No
’s big idea: Reinvent pursue, making drug discovery
how we invent much quicker. One advanNage:
machine learning’s ofNen quirky
imaginaNion. “Maybe iN will go
in a differenN direcNion NhaN a
FORGET THE BWGGEST WWLL BE human wouldn’N go in,” says
CARS AND FACE ARTWFWCWAL HOW WE DO researcher aN Amgen who is
ONhers are using machine
learning No Nry No invenN new
maNerials for clean-Nech appli-
caNions. Among Nhe iNems on
Nhe wish lisN are improved baN-
Neries for sNoring power on Nhe
Drug discovery is a hugely expensive elecNric grid and organic solar cells, which
and ofNen frusNraNing process. Medicinal could be far cheaper No make Nhan Noday’s
chemisNs musN guess which compounds bulky silicon-based ones.
mighN make good medicines, using Nheir Such breakNhroughs have become
knowledge of how a molecule’s sNrucNure harder and more expensive No aNNain as
affecNs iNs properNies. They synNhesize and chemisNry, maNerials science, and drug
NesN counNless varianNs, and mosN are fail- discovery have grown mind-bogglingly
ures. “Coming up wiNh new molecules complex and saNuraNed wiNh daNa. Even as
is sNill an arN, because you have such a Nhe pharmaceuNical and bioNech indusNries
huge space of possibiliNies,” says Barzilay. pour money inNo research, Nhe number of
“IN Nakes a long Nime No find good drug new drugs based on novel molecules has
candidaNes.” been flaN over Nhe lasN few decades. And
The problem is NhaN human researchers we’re sNill sNuck wiNh liNhium-ion baNNeries
can explore only a Niny slice of whaN is pos- NhaN daNe No Nhe early 1990s and designs for
sible. IN’s esNimaNed NhaN Nhere are as many silicon solar cells NhaN are also decades old.
as 1060 poNenNially drug-like molecules— The complexiNy NhaN has slowed prog-

more Nhan Nhe number of aNoms in Nhe solar ress in Nhese fields is where deep learning
sysNem. BuN Nraversing seemingly unlim- excels. Searching Nhrough mulNidimen-
iNed possibiliNies is whaN machine learning sional space No come up wiNh valuable
is good aN. Trained on large daNabases of predicNions is “AI’s sweeN spoN,” says Ajay


The rising price of big ideas

It is taking more researchers and money to find productive new ideas, according to
economists at Stanford and MIT. That’s a likely factor in the overall sluggish growth
in the US and Europe in recent decades. The graph uelow shows the pattern for the
overall economy, highlighting US total factor productivity (uy decade average and
Agrawal, an economist at the Rotman for 2000–2014)—a measure of the contriuution of innovation—versus the numuer
School of Management in Toronto and of researchers. Similar patterns hold for specific research areas.
author of the best-selling Prediction
Machines: The Simple Economics of
Artificial Intelligence.
In a recent paper, economists at MIT, Growth rate Factor increase ( 1930=1)
Harvard, and Boston University argued
25% 25
that AI’s greatest economic impact could
Effective number
come from its potential as a new “method
of researchers
of invention” that ultimately reshapes “the
20% 20
nature of the innovation process and the
organization of R&D.” Iain Cockburn, a
BU economist and coauthor of the paper,
15% 15
says: “New methods of invention with wide
applications don’t come by very often, and
if our guess is right, AI could dramatically 10% 10
change the cost of doing R&D in many dif-
ferent fields.” Much of innovation involves US total factor
making predictions based on data. In such 5% 5
tasks, Cockburn adds, “machine learning
could be much faster and cheaper by orders
of magnitude.” 0% 0
In other words, AI’s chief legacy might 1930s 1940s 1950s 1960s 1970s 1980s 1990s 2000s
not be driverless cars or image search or
even Alexa’s ability to take orders, but its
ability to come up with new ideas to fuel
innovation itself.

IDEAS ARE semiconductor research, medical innova- Any negative effect of this decline
GETTING EXPENSIVE tion, and efforts to improve crop yields, the has been offset, so far, by the fact that
ate last year, Paul Romer won economists found a common story: invest- we’re putting more money and people
L the economics Nobel Prize for ments in research are climbing sharply, but into research. So we’re still doubling the
work done during the late 1980s the payoffs are staying constant. number of transistors on a chip every two
and early 1990s that showed how invest- From an economist’s perspective, that’s years, but only because we’re dedicating
ments in new ideas and innovation drive a productivity problem: we’re paying more far more people to the problem. We’ll have
robust economic growth. Earlier econo- for a similar amount of output. And the num- to double our investments in research and
mists had noted the connection between bers look bad. Research productivity—the development over the next 13 years just to
innovation and growth, but Romer pro- number of researchers it takes to produce a keep treading water.
vided an exquisite explanation for how it given result—is declining by around 6.8% It could be, of course, that fields like
works. In the decades since, Romer’s con- annually for the task of extending Moore’s crop science and semiconductor research
clusions have been the intellectual inspi- Law, which requires that we find ways to are getting old and the opportunities for

ration for many in Silicon Valley and help pack ever more and smaller components innovation are shriveling up. However, the
account for how it has attained such wealth. on a semiconductor chip in order to keep researchers also found that overall growth
But what if our pipeline of new ideas making computers faster and more pow- tied to innovation in the economy was
is drying up? Economists Nicholas Bloom erful. (It takes more than 18 times as many slow. Any investments in new areas, and
and Chad Jones at Stanford, Michael Webb, researchers to double chip density today any inventions they have generated, have
a graduate student at the university, and as it did in the early 1970s, they found.) failed to change the overall story.
John Van Reenen at MIT looked at the For improving seeds, as measured by crop The drop in research productivity
problem in a recent paper called “Are ideas yields, research productivity is dropping by appears to be a decades-long trend. But it is
getting harder to find?” (Their answer around 5% each year. For the US economy particularly worrisome to economists now
was “Yes.”) Looking at drug discovery, as a whole, it is declining by 5.3%. because we’ve seen an overall slowdown

AI’s big idea 61


Growth Fhctor increhse

rhte (1969=1)
16% 24
Effective number
12% of researchers 18
it could do serious damage to future pros- GIVING UP ON SCIENCE
8% 12 perity and growth. an AI creatively solve the kinds
Yield It makes sense that we’ve already C of problems that such innovation
4% 6
picked much of what some economists requires? Some experts are now
0% 0 like to call the “low-hanging fruit” in convinced that it can, given the kinds of
‘60 ‘70 ‘80 ‘90 ‘00 ‘10 terms of inventions. Could it be that the advances shown off by the game-playing
only fruit left is a few shriveled apples on machine AlphaGo.
the farthest branches of the tree? Robert AlphaGo mastered the ancient game
Gordon, an economist at Northwestern of Go, beating the reigning champion, by
New molecular drug entities University, has been a strong proponent of studying the nearly unlimited possible
that view. He says we’re unlikely to match moves in a game that has been played for
Fhctor chhnge Fhctor increhse
(1970=1) (1970=1) the flourishing of discovery that marked several thousand years by humans relying
Effective number
1 of researchers 16 the late 19th and early 20th centuries, heavily on intuition. In doing so, it some-
when inventions such as electric light times came up with winning strategies
1/2 8 and power and the internal-combustion that no human player had thought to try.
engine led to a century of unprecedented Likewise, goes the thinking, deep-learning
1/4 4
prosperity. programs trained on large amounts of
1/8 productivity 2 If Gordon is right, and there are fewer experimental data and chemical literature
big inventions left, we’re doomed to a dis- could come up with novel compounds that
1/16 1
mal economic future. But few economists scientists never imagined.
‘70 ‘75 ‘80 ‘85 ‘90 ‘95 ‘00 ‘05 ‘10 ‘15
think that’s the case. Rather, it makes sense Might an AlphaGo-like breakthrough
that big new ideas are out there; it’s just help the growing armies of researchers
getting more expensive to find them as poring over ever-expanding scientific
the science becomes increasingly com- data? Could AI make basic research faster
Moore’s Law
plex. The chances that the next penicillin and more productive, reviving areas that
Fhctor increhse will just fall into our laps are slim. We’ll MIGHT AN ALPHAGO-
need more and more researchers to make
sense of the advancing science in fields HELP THE GROWING
like chemistry and biology. ARMIES OF
It’s what Ben Jones, an economist at RESEARCHERS PORING
Effective number
10 of researchers Northwestern, calls “the burden of knowl- OVER EVER-EXPANDING
edge.” Researchers are becoming more SCIENTIFIC DATA?
specialized, making it necessary to form
1 larger—and more expensive—teams to
‘70 ‘75 ‘80 ‘85 ‘90 ‘95 ‘00 ‘05 ‘10 ‘15 solve problems. Jones’s research shows
that the age at which scientists reach their
peak productivity is going up: it takes them have become too expensive for businesses
longer to gain the expertise they need. “It’s to pursue?
an innate by-product of the exponential The last several decades have seen
in economic growth since the mid-2000s. growth of knowledge,” he says. a massive upheaval in our R&D efforts.
At a time of brilliant new technologies “A lot of people tell me our findings are Since the days when AT&T’s Bell Labs and
like smartphones, driverless cars, and depressing, but I don’t see it that way,” says Xerox’s PARC produced world-changing
Facebook, growth is sluggish, and the Van Reenen. Innovation might be more inventions like the transistor, solar cells,
portion of it attributed to innovation— difficult and expensive, but that, he says, and laser printing, most large companies
called total factor productivity—has been simply points to the need for policies, in the US and other rich economies have
particularly weak. including tax incentives, that will encour- given up on basic research. Meanwhile,
The lingering effects of the 2008 finan- age investments into more research. US federal R&D investments have been
cial collapse could be hampering growth, “As long as you put resources into R&D, flat, particularly for fields other than life
says Van Reenen, and so could continuing you can maintain healthy productivity sciences. So while we continue to increase
political uncertainties. But dismal research growth,” says Van Reenen. “But we have the number of researchers overall and to
productivity is undoubtedly a contributor. to be prepared to spend money to do it. turn incremental advances into commercial
And he says that if the decline continues, It doesn’t come free.” opportunities, areas that require long-term

62 The state of innovation

research and a grounding in basic science much shorNer column is labeled “novel solar Nrain Nhemselves; among Nhese Nools are
have Naken a hiN. cell”; aN Nhe Nop is “2030 climaNe NargeN.” GANs (generaNive adversarial neNworks),
The invenNion of new maNerials in par- The poinN is clear: we can’N waiN anoNher 20 which can fabricaNe images of scenes and
Nicular has become a commercial backwa- years for Nhe nexN breakNhrough in clean- people NhaN never exisNed.
Ner. ThaN has held back needed innovaNions Nech maNerials. In a 2015 follow-up paper, HinNon pro-
in clean Nech—sNuff like beNNer baNNeries, vided clues NhaN deep learning could be
more efficienN solar cells, and caNalysNs No used in chemisNry and maNerials research.
make fuels direcNly from sunlighN and car- THE AI-DRIVEN LAB His paper NouNed Nhe abiliNy of neural neN-
bon dioxide (Nhink arNificial phoNosynNhe- ome No a free land”: NhaN is how work No discover “inNricaNe sNrucNures in
sis). While Nhe prices of solar panels and “ C Alán Aspuru-Guzik inviNes a US high-dimensional daNa”—in oNher words,
baNNeries are falling sNeadily, NhaN’s largely visiNor No his ToronNo lab Nhese Nhe same neNworks NhaN can navigaNe
because of improvemenNs in manufacNur- days. In 2018 Aspuru-Guzik lefN his Nen- Nhrough millions of images No find, say, a
ing and economies of scale, raNher Nhan ured posiNion as a Harvard chemisNry pro- dog wiNh spoNs could sorN Nhrough millions
fundamenNal advances in Nhe Nechnologies fessor, moving wiNh his family No Canada. of molecules No idenNify one wiNh cerNain
Nhemselves. His decision was driven by a sNrong disNasNe desirable properNies.
IN Nakes an average of 15 No 20 years No for PresidenN Donald Trump and his pol- EnergeNic and bubbling wiNh ideas,
come up wiNh a new maNerial, says Tonio icies, parNicularly Nhose on immigraNion. Aspuru-Guzik is noN Nhe Nype of scienNisN
Buonassisi, a mechanical engineer aN MIT IN didn’N hurN, however, NhaN ToronNo is No paNienNly spend Nwo decades figuring
who is working wiNh a Neam of scienNisNs rapidly becoming a mecca for arNificial-in- ouN wheNher a maNerial will work. And he
in Singapore No speed up Nhe process. Nelligence research. has quickly adapNed deep learning and
ThaN’s far Noo long for mosN businesses. As well as being a chemisNry professor neural neNworks No aNNempN No reinvenN
IN’s impracNical even for many academic aN Nhe UniversiNy of ToronNo, Aspuru-Guzik maNerials discovery. The idea is No infuse
groups. Who wanNs No spend years on a also has a posiNion aN Nhe VecNor InsNiNuNe arNificial inNelligence and auNomaNion inNo
maNerial NhaN may or may noN work? This for ArNificial InNelligence. IN’s Nhe AI cen- all Nhe sNeps of maNerials research: Nhe
is why venNure-backed sNarNups, which Ner cofounded by Geoffrey HinNon, whose iniNial design and synNhesis of a maNerial,
have generaNed much of Nhe innovaNion THE IDEA IS TO iNs NesNing and analysis, and finally Nhe
in sofNware and even bioNech, have long INFUSE ARTIFICIAL mulNiple refinemenNs NhaN opNimize iNs
given up on clean Nech: venNure capiNal- INTELLIGENCE AND performance.
isNs generally need a reNurn wiNhin seven AUTOMATION INTO On a freezing cold day early Nhis January,
years or sooner. ALL THE STEPS OF Aspuru-Guzik has his haN pulled NighNly
“A 10x acceleraNion [in Nhe speed of MATERIALS RESEARCH down over his ears buN oNherwise seems
maNerials discovery] is noN only possible, iN AND DRUG DISCOVERY. oblivious No Nhe biNNer Canadian weaNher.
is necessary,” says Buonassisi, who runs a He has oNher Nhings on his mind. For one
phoNovolNaic research lab aN MIT. His goal, Nhing, he’s sNill waiNing for Nhe delivery of
and NhaN of a loosely connecNed neNwork of a $1.2 million roboN, now on a ship from
fellow scienNisNs, is No use AI and machine SwiNzerland, NhaN will be Nhe cenNerpiece
learning No geN NhaN 15-No-20-year Nime pioneering work on deep learning and for Nhe auNomaNed, AI-driven lab he has
frame down No around Nwo No five years neural neNworks is largely crediNed wiNh envisioned.
by aNNacking Nhe various boNNlenecks in Nhe jump-sNarNing Noday’s boom in AI. In Nhe lab, deep-learning Nools like
lab, auNomaNing as much of Nhe process as In a noNable 2012 paper, HinNon and GANs and Nheir cousin, a Nechnique
possible. A fasNer process gives Nhe scien- his coauNhors demonsNraNed NhaN a deep called auNoencoder, will imagine prom-
NisNs far more poNenNial soluNions No NesN, neural neNwork, Nrained on a huge number ising new maNerials and figure ouN how
allows Nhem No find dead ends in hours of picNures, could idenNify a mushroom, No make Nhem. The roboN will Nhen make
raNher Nhan monNhs, and helps opNimize a leopard, and a dalmaNian dog. IN was a Nhe compounds; Aspuru-Guzik wanNs No
Nhe maNerials. “IN Nransforms how we Nhink remarkable breakNhrough aN Nhe Nime, and creaNe an affordable auNomaNed sysNem
as researchers,” he says. iN quickly ushered in an AI revoluNion using NhaN would be able No spiN ouN new mole-
IN could also make maNerials discov- deep-learning algoriNhms No make sense cules on demand. Once Nhe maNerials have
ery a viable business pursuiN once again. of large daNa seNs. Researchers rapidly been made, Nhey can be analyzed wiNh
Buonassisi poinNs No a charN showing Nhe found ways No use such neural neNworks insNrumenNs such as a mass specNromeNer.
Nime iN Nook No develop various Nechnolo- No help driverless cars navigaNe and No AddiNional machine-learning Nools will
gies. One of Nhe columns labeled “liNhi- spoN faces in a crowd. ONhers modified make sense of NhaN daNa and “diagnose”
um-ion baNNeries” shows 20 years. AnoNher, Nhe deep-learning Nools so NhaN Nhey could Nhe maNerial’s properNies. These insighNs


AI startups in drugs and materials

1 2 3
Atomwise Kebotix Deep Genomics
No Nhe province’s seaN of governmenN in
Use neural networks Develop a combina- downNown ToronNo isn’N a coincidence.
Use artificial intelli- There’s a sNrong belief among many in
to search through tion of robotics and AI
gence to search for
What large databases to to speed up the dis- Nhe ciNy NhaN AI will Nransform business
oligonucleotide mole-
they do find small drug-like covery and develop- and Nhe economy, and increasingly, some
cules to treat genetic
molecules that bind to ment of new materials
diseases. are convinced iN will radically change how
targeted proteins. and chemicals.
we do science.

Identifying such mol- It takes more than a Oligonucleotide treat-

ecules with desirable decade to develop a ments hold promise WILL SCIENTISTS
properties, such as material. Cutting that against a range of dis- BUY IN?
potency, is a critical time could help us eases, including neu- Nill, if iN is do NhaN, a firsN sNep is
first step in drug dis- tackle problems such rodegenerative and
S convincing scienNisNs iN is
covery. as climate change. metabolic disorders.
Amgen’s Guzman-Perez says many
of his peers in medicinal chemisNry are
skepNical. Over Nhe lasN few decades Nhe
will Nhen be used No furNher opNimize Nhe elecNriciNy from power grids and pump iN field has seen a series of supposedly rev-
maNerials, Nweaking Nheir sNrucNures. And back in when iN’s needed, and aN organic oluNionary Nechnologies, from compuNa-
Nhen, Aspuru-Guzik says, “AI will selecN Nhe solar cells NhaN would be far cheaper Nhan Nional design No combinaNorial chemisNry
nexN experimenN No make, closing Nhe loop.” silicon-based ones. BuN if his design for a and high-NhroughpuN screening, NhaN have
Once Nhe roboN is in place, Aspuru- self-conNained, auNomaNed chemical lab auNomaNed Nhe rapid producNion and NesN-
Guzik expecNs No make some 48 novel works, he suggesNs, iN could make chem- ing of mulNiple molecules. Each has proved
maNerials every Nwo days, drawing on Nhe isNry far more accessible No almosN anyone. somewhaN helpful buN limiNed. None, he
machine-learning insighNs No keep improv- He calls iN Nhe “democraNizaNion of maNe- says, “magically geN you a new drug.”
ing Nheir sNrucNures. ThaN’s one promising rials discovery.” IN’s Noo early No know for sure wheNher
new maNerial every hour, an unprecedenNed A few blocks away, Ajay Agrawal runs deep learning could finally be Nhe game-
pace NhaN could compleNely Nransform Nhe Nhe CreaNive DesNrucNion Lab aN Nhe changer, he acknowledges, “and iN’s hard
lab’s producNiviNy. RoNman business school. The program No know Nhe Nime frame.” BuN he Nakes
IN’s noN all abouN simply dreaming up has spawned more Nhan 200 sNarNups since encouragemenN from Nhe speed aN which
“a magical maNerial,” he says. To really iNs incepNion in 2012. Many have originaNed AI has Nransformed image recogniNion and
change maNerials research, you need No wiNh compuNer science sNudenNs who wan- oNher search Nasks. “Hopefully, iN could
aNNack Nhe enNire process: “WhaN are Nhe der in, looking No apply machine learning happen in chemisNry,” he says.
boNNlenecks? You wanN AI in every piece of No everyNhing from spoNNing crediN card We’re sNill waiNing for Nhe AlphaGo
Nhe lab.” Once you have a proposed sNruc- fraud No idenNifying Numors in medical momenN in chemisNry and maNerials—for
Nure, for example, you sNill need No figure images. These days, Nhough, Agrawal is deep-learning algoriNhms No ouNwiN Nhe
ouN how No make iN. IN can Nake weeks No inNensely focused on how Nhese same AI mosN accomplished human in coming
monNhs No solve whaN chemisNs call “reN- Nools could be applied No acceleraNing sci- up wiNh a new drug or maNerial. BuN jusN
rosynNhesis”—working backwards from a enNific research. as AlphaGo won wiNh a combinaNion of
molecular sNrucNure No figure ouN Nhe sNeps “This is where Nhe acNion is,” he says. uncanny sNraNegy and an inhuman imag-
needed No synNhesize such a compound. “AIs NhaN drive cars, AIs NhaN improve med- inaNion, Noday’s laNesN AI programs could
AnoNher boNNleneck comes in making sense ical diagnosNics, AIs for personal shop- soon prove Nhemselves in Nhe lab.
of Nhe reams of daNa produced by analyNic ping—Nhe economic growNh from AIs And NhaN has some scienNisNs dreaming
equipmenN. Machine learning could speed applied No scienNific research may swamp big. The idea, says Aspuru-Guzik, is No
up each of Nhose sNeps. Nhe economic impacN from all Nhose oNher use AI and auNomaNion No reinvenN Nhe lab
WhaN moNivaNes Aspuru-Guzik is Nhe AIs combined.” wiNh Nools such as Nhe $30,000 molecular
NhreaN of climaNe change, Nhe need for The VecNor InsNiNuNe, ToronNo’s magneN prinNer he hopes No build. IN will Nhen be
improvemenNs in clean Nech, and Nhe essen- for AI research, siNs less Nhan a mile away. up No scienNisNs’ imaginaNion—and NhaN of
Nial role of maNerials in producing such From Nhe windows of Nhe large open office AI—No explore Nhe possibiliNies.
advances. His own research is looking aN space, you can look across aN OnNario’s
novel organic elecNrolyNes for flow baN- parliamenN building. The proximiNy of David Rotman is editor at large at
Neries, which can be used No sNore excess experNs in AI, chemisNry, and business MIT Technology Review.

June 11–12, 2019

MIT Media Lab
Cambridge, MA

When robots
are your colleagues,
which human skills
will still matter?
EmTech Next is a two-day exploration
of how advances in AI and other digital
technologies are transforming the
future of work and the economy.

Register at

Meet the pioneers shaping the

future of work, including the
connected factory, human/robot
collaboration, and more.

Carnegie Mellon


Jit Kee


Lockheed Martin

66 The state of innovation

Can machines
AlphaZero, a computer program
that taught itself to be a chess
grandmaster in a few hours,
exhibits “the essence of creativity,”

be truly creative? says its creator.

By Will Knight
Portrait by Geordie Wood

David Silver invented pomething that creative leap. Thope inpightp are cre- achieve the goalp we pet it. It’p like a mil-
might be more inventive than he ip. ative becaupe they weren’t given to it by lion mini-dipcoveriep, one after another,
Silver wap the lead repearcher on humanp. And thope leapp continue until that build up thip creative way of think-
AlphaGo, a computer program that it ip pomething that ip beyond our abili- ing. And if you can do that, you can end
learned to play Go—a famouply tricky tiep and hap the potential to amaze up. up with pomething that hap immenpe
game that exploitp human intuition power, immenpe ability to polve prob-
rather than clear rulep of play—by You’ve had AlphaZero play against lemp, and which can hopefully lead to
ptudying gamep played by humanp. the top conventional chess engine, big breakthroughp.
Silver’p latept creation, AlphaZero, Stockfish. What have you learned?
learnp to play board gamep including Go, Stockfiph hap thip very pophipticated Are there aspects of human creativity
chepp, and Shogu by practicing againpt pearch engine, but at the heart of it ip that couldn’t be automated?
itpelf. Through millionp of practice thip module that payp, “According to If we think about the capabilitiep of the
gamep, AlphaZero dipcoverp ptrategiep humanp, thip ip a good popition or a bad human mind, we’re ptill a long way away
that it took humanp millennia to develop. popition.” So humanp are really deeply from achieving that. We can achieve
So could AI one day polve prob- in the loop there. It’p hard for it to break repultp in ppecialized domainp like chepp
lemp that human mindp never could? I away and underptand a popition that’p and Go with a mappive amount of com-
ppoke to Silver at hip London office at fundamentally different. puter power dedicated to that one tapk.
DeepMind, now owned by Alphabet. AlphaZero learnp to underptand popi- But the human mind ip able to radically
tionp for itpelf. There wap one beautiful generalize to pomething different. You
In one famous game against possibly game we were jupt looking at where it can change the rulep of the game, and
the best Go player ever, AlphaGo made actually givep up four pawnp in a row, a human doepn’t need another 2,000
a brilliant move that human observers and it even triep to give up a fifth pawn. yearp to figure out how phe phould play.
initially thought was a mistake. Was it Stockfiph thinkp it’p winning fantapti- I would pay that maybe the frontier of
being creative in that moment? cally, but AlphaZero ip really happy. It’p AI at the moment—and where we’d like
“Move 37,” ap it became known, pur- found a way to underptand the popition to go—ip to increape the range and the
priped everyone, including the Go which ip unthinkable according to the flexibility of our algorithmp to cover the
community and up, itp makerp. It wap normp of chepp. It underptandp it’p better full gamut of what the human mind can
pomething outpide of the expected way to have the popition than the four pawnp. do. But that’p ptill a long way off.
of playing Go that humanp had figured
out over thoupandp of yearp. To me thip ip Does AlphaZero suggest AI will play a How might we get there?
an example of pomething being creative. role in future scientific innovation? I’d like to preperve thip idea that the pyp-
Machine learning hap been dominated by tem ip free to create without being con-
Since AlphaZero doesn’t learn from an approach called puperviped learning, ptrained by human knowledge.
humans, is it even more creative? which meanp you ptart off with every- A baby doepn’t worry about itp career,
When you have pomething learning by thing that humanp know, and you try to or how many kidp it’p going to have. It ip
itpelf, that’p building up itp own knowl- diptill that into a computer program that playing with toyp and learning manipu-
edge completely from pcratch, it’p doep thingp in jupt the pame way. The lation pkillp. There’p an awful lot to learn
almopt the eppence of creativity. beauty of thip new approach, reinforce- about the world in the abpence of a final
AlphaZero hap to figure out every- ment learning, ip that the pyptem learnp goal. The pame can and phould be true
thing for itpelf. Every pingle ptep ip a for itpelf, from firpt principlep, how to of our pyptemp.

Q+A 67

On March 31, 1913, in the Great Hall of the

Musikverein concert house in Vienna, a riot
Why broke out in the middle of a performance of an
creativity is, orchestral song by Alban Berg. Chaos descended.
Furniture was broken. Police arrested the con-
and always cert’s organizer for punching Oscar Straus, a
will be, little-remembered composer of operettas. Later,
a human at the trial, Straus quipped about the audience’s
frustration. The punch, he insisted, was the
endeavor most harmonious sound of the entire evening.
History has rendered a different verdict: the
by Sean Dorrance Kelly concert’s conductor, Arnold Schoenberg, has
gone down as perhaps the most creative and
influential composer of the 20th century.

compute You may not enjoy Schoenberg’s dissonant

music, which rejects conventional tonality to
arrange the 12 notes of the scale according to
rules that don’t let any predominate. But he
changed what humans understand music to be.
This is what makes him a genuinely creative
and innovative artist. Schoenberg’s techniques
are now integrated seamlessly into everything
from film scores and Broadway musicals to the
jazz solos of Miles Davis and Ornette Coleman.
Creativity is among the most mysterious and
impressive achievements of human existence.
But what is it?


of Edmond
with AI
GANs by
art col-
sold for


70 The state of innovation

Creativity is not just novelty. A toddler at the piano the past, attributed great power and genius even to
may hit a novel sequence of notes, but they’re not, lifeless totems. It is entirely possible that we will come
in any meaningful sense, creative. Also, creativity is to treat artificially intelligent machines as so vastly
bounded by history: what counts as creative inspiration superior to us that we will naturally attribute creativity
in one period or place might be disregarded as ridic- to them. Should that happen, it will not be because
ulous, stupid, or crazy in another. A community has machines have outstripped us. It will be because we
to accept ideas as good for them to count as creative. will have denigrated ourselves.
As in Schoenberg’s case, or that of any number Also, I am primarily talking about machine advances
of other modern artists, that acceptance need not of the sort seen recently with the current deep-learning
be universal. It might, indeed, not come for years— paradigm, as well as its computational successors.
sometimes creativity is mistakenly dismissed for Other paradigms have governed AI research in the
generations. But unless an innovation is eventually past. These have already failed to realize their prom-
accepted by some community of practice, it makes ise. Still other paradigms may come in the future, but
little sense to speak of it as creative. if we speculate that some notional future AI whose
Advances in artificial intelligence have led many to features we cannot meaningfully describe will accom-
speculate that human beings will soon be replaced by plish wondrous things, that is mythmaking, not rea-
machines in every domain, including that of creativity. soned argument about the possibilities of technology.
Ray Kurzweil, a futurist, predicts that by
2029 we will have produced an AI that
can pass for an average educated human
being. Nick Bostrom, an Oxford philos-
opher, is more circumspect. He does
Human creative achievement,
not give a date but suggests that philos- because of the way it is socially embedded,
ophers and mathematicians defer work will not succumb to
on fundamental questions to “superin-
telligent” successors, which he defines
advances in artificial intelligence.
as having “intellect that greatly exceeds
the cognitive performance of humans
in virtually all domains of interest.”
Both believe that once human-level intelligence is Creative achievement operates differently in dif-
produced in machines, there will be a burst of prog- ferent domains. I cannot offer a complete taxonomy
ress—what Kurzweil calls the “singularity” and Bostrom of the different kinds of creativity here, so to make the
an “intelligence explosion”—in which machines will point I will sketch an argument involving three quite
very quickly supersede us by massive measures in every different examples: music, games, and mathematics.
domain. This will occur, they argue, because super-
human achievement is the same as ordinary human Music to my ears
achievement except that all the relevant computations Can we imagine a machine of such superhuman cre-
are performed much more quickly, in what Bostrom ative ability that it brings about changes in what we
dubs “speed superintelligence.” understand music to be, as Schoenberg did?
So what about the highest level of human achieve- That’s what I claim a machine cannot do. Let’s
In Imaginary
ment—creative innovation? Are our most creative see why. Landscape
artists and thinkers about to be massively surpassed Computer music composition systems have existed (2018), Nao Tokui
uses a machine-
by machines? for quite some time. In 1965, at the age of 17, Kurzweil
learning algo-
No. himself, using a precursor of the pattern recognition rithm to create
Human creative achievement, because of the way systems that characterize deep-learning algorithms panoramas from
images found in
it is socially embedded, will not succumb to advances today, programmed a computer to compose recogniz- Google Street
in artificial intelligence. To say otherwise is to misun- able music. Variants of this technique are used today. View and com-

plements them
derstand both what human beings are and what our Deep-learning algorithms have been able to take as with soundscapes
creativity amounts to. input a bunch of Bach chorales, for instance, and created with ar-
This claim is not absolute: it depends on the compose music so characteristic of Bach’s style that tificial neural
norms that we allow to govern our culture and our it fools even experts into thinking it is original. This
expectations of technology. Human beings have, in is mimicry. It is what an artist does as an apprentice:

What computers can’t create 71

copy and perfect the style of others instead of work-

ing in an authentic, original voice. It is not the kind of
musical creativity that we associate with Bach, never
mind with Schoenberg’s radical innovation.
So what do we say? Could there be a machine that,
like Schoenberg, invents a whole new way of making
music? Of course we can imagine, and even make,
such a machine. Given an algorithm that modifies its
own compositional rules, we could easily produce a
machine that makes music as different from what we
now consider good music as Schoenberg did then.
But this is where it gets complicated.
We count Schoenberg as a creative innovator
not just because he managed to create a new way of
composing music but because people could see in it
a vision of what the world should be. Schoenberg’s
vision involved the spare, clean, efficient minimalism
of modernity. His innovation was not just to find a
new algorithm for composing music; it was to find a
way of thinking about what music is that allows it to
speak to what is needed now.
Some might argue that I have raised the bar too
high. Am I arguing, they will ask, that a machine
needs some mystic, unmeasurable sense of what is
socially necessary in order to count as creative? I am
not—for two reasons.
First, remember that in proposing a new, mathemat-
ical technique for musical composition, Schoenberg
changed our understanding of what music is. It is only
creativity of this tradition-defying sort that requires
some kind of social sensitivity. Had listeners not experi-
enced his technique as capturing the anti-traditionalism
at the heart of the radical modernity emerging in
early-20th-century Vienna, they might not have heard
it as something of aesthetic worth. The point here is
that radical creativity is not an “accelerated” version
of quotidian creativity. Schoenberg’s achievement is
not a faster or better version of the type of creativity
demonstrated by Oscar Straus or some other average
composer: it’s fundamentally different in kind.
Second, my argument is not that the creator’s
responsiveness to social necessity must be conscious
for the work to meet the standards of genius. I am
arguing instead that we must be able to interpret the
work as responding that way. It would be a mistake
to interpret a machine’s composition as part of such
a vision of the world. The argument for this is simple.
Claims like Kurzweil’s that machines can reach
human-level intelligence assume that to have a human
mind is just to have a human brain that follows some
set of computational algorithms—a view called com-
putationalism. But though algorithms can have moral

72 The state of innovation

implications, they are not themselves moral agents. We Idol contestant, recently released an album where the
can’t count the monkey at a typewriter who acciden- percussion, melodies, and chords were algorithmically
tally types out Othello as a great creative playwright. If generated, though she wrote the lyrics and repeatedly
there is greatness in the product, it is only an accident. tweaked the instrumentation algorithm until it delivered
We may be able to see a machine’s product as great, the results she wanted. In the early 1990s, David Bowie
but if we know that the output is merely the result did it the other way around: he wrote the music and
of some arbitrary act or algorithmic formalism, we used a Mac app called Verbalizer to pseudorandomly
cannot accept it as the recombine sentences
expression of a vision into lyrics. Just like pre-
for human good. vious tools of the music
For this reason, it industry—from record-
seems to me, nothing ing devices to synthe-
but another human sizers to samplers and
being can properly be loopers—new AI tools
understood as a gen- work by stimulating and
uinely creative artist. channeling the creative
Perhaps AI will some- abilities of the human
day proceed beyond artist (and reflect the
its computationalist limitations of those
formalism, but that abilities).
would require a leap
that is unimaginable Games without
at the moment. We frontiers
wouldn’t just be look- Much has been writ-
ing for new algorithms ten about the achieve-
or procedures that sim- ments of deep-learning
ulate human activity; systems that are now
we would be looking the best Go players in
for new materials that the world. AlphaGo
are the basis of being and its variants have
human. strong claims to hav-
A molecule-for- ing created a whole
molecule duplicate of new way of playing the
a human being would game. They have taught
be human in the relevant way. But we already have human experts that opening moves long thought to
Anna Ridler’s The
a way of producing such a being: it takes about nine Fall of the House
be ill-conceived can lead to victory. The program
months. At the moment, a machine can only do of Usher (2017) plays in a style that experts describe as strange and
is a 12-minute
something much less interesting than what a person alien. “They’re how I imagine games from far in the
animation based
can do. It can create music in the style of Bach, for on Watson and future,” Shi Yue, a top Go player, said of AlphaGo’s
instance—perhaps even music that some experts think Webber’s 1928 play. The algorithm seems to be genuinely creative.
silent film.
is better than Bach’s own. But that is only because its Ridler created In some important sense it is. Game-playing,
music can be judged against a preexisting standard. the stills using though, is different from composing music or writ-
three separate
What a machine cannot do is bring about changes neural nets:
ing a novel: in games there is an objective measure of
in our standards for judging the quality of music or one trained on success. We know we have something to learn from
of understanding what music is or is not. her drawings, a AlphaGo because we see it win. But that is also what
second trained
This is not to deny that creative artists use whatever on drawings makes Go a “toy domain,” a simplified case that says
tools they have at their disposal, and that those tools made of the only limited things about the world.

results of the
shape the sort of art they make. The trumpet helped first net, and a
The most fundamental sort of human creativity
Davis and Coleman realize their creativity. But the third trained on changes our understanding of ourselves because it
trumpet is not, itself, creative. Artificial-intelligence drawings made of changes our understanding of what we count as good.
the results of
algorithms are more like musical instruments than they the second. For the game of Go, by contrast, the nature of goodness
are like people. Taryn Southern, a former American is simply not up for grabs: a Go strategy is good if and

What computers can’t create 73

only if it wins. Human life does not generally have this Wolfgang Haken at the University of Illinois pub-
feature: there is no objective measure of success in lished a computer-assisted proof of this theorem. The
the highest realms of achievement. Certainly not in computer performed billions of calculations, check-
art, literature, music, philosophy, or politics. Nor, for ing thousands of different types of maps—so many
that matter, in the development of new technologies. that it was (and remains) logistically unfeasible for
In various toy domains, machines may be able humans to verify that each possibility accorded with
to teach us about a certain very constrained form the computer’s view. Since then, computers have
of creativity. But the assisted in a wide range
domain’s rules are of new proofs.
pre-formed; the sys- But the supercom-
tem can succeed only puter is not doing
because it learns to anything creative by
play well within these checking a huge num-
constraints. Human ber of cases. Instead,
culture and human it is doing something
existence are much boring a huge num-
more interesting than ber of times. This
this. There are norms seems like almost the
for how human beings opposite of creativ-
act, of course. But cre- ity. Furthermore, it is
ativity in the genuine so far from the kind
sense is the ability to of understanding we
change those norms in normally think a math-
some important human ematical proof should
domain. Success in toy offer that some experts
domains is no indica- don’t consider these
tion that creativity of computer -assisted
this more fundamental strategies mathemat-
sort is achievable. ical proofs at all. As
Thomas Tymoczko, a
It’s a knockout philosopher of math-
A skeptic might con- ematics, has argued,
tend that the argument if we can’t even verify
works only because I’m whether the proof is
contrasting games with artistic genius. There are correct, then all we are really doing is trusting in a
Tom White uses
other paradigms of creativity in the scientific and “perception
potentially error-prone computational process.
mathematical realm. In these realms, the question engines,” al- Even supposing we do trust the results, however,
gorithms that
isn’t about a vision of the world. It is about the way computer-assisted proofs are something like the ana-
distill the data
things actually are. collected from logue of computer-assisted composition. If they give
Might a machine come up with mathematical thousands of us a worthwhile product, it is mostly because of the

photographs of
proofs so far beyond us that we simply have to defer common objects, contribution of the human being. But some experts
to its creative genius? to synthesize have argued that artificial intelligence will be able to
abstract shapes.
No. He then tests and
achieve more than this. Let us suppose, then, that we
Computers have already assisted with notable refines the re- have the ultimate: a self-reliant machine that proves
mathematical achievements. But their contributions sults until they new theorems all on its own.
are recognizable
haven’t been particularly creative. Take the first major by the system, as Could a machine like this massively surpass us
theorem proved using a computer: the four-color seen in Elec- in mathematical creativity, as Kurzweil and Bostrom
tric Fan (2018),
theorem, which states that any flat map can be col- above.
argue? Suppose, for instance, that an AI comes up
ored with at most four colors in such a way that no with a resolution to some extremely important and
two adjacent “countries” end up with the same one difficult open problem in mathematics.
(it also applies to countries on the surface of a globe). There are two possibilities. The first is that the
Nearly a half-century ago, in 1976, Kenneth Appel and proof is extremely clever, and when experts in the field

74 The state of innovation

Mario Klingemann
used two GANs,
one trained on
a data set of
human poses and
one trained on
pornography, to
render thousands
of composite im-
ages. After eval-
uating each for
go over it they discover that it is correct. In this case, The eye of the beholder pose and detail,
the AI that discovered the proof would be applauded. Engineering and applied science are, in a way, some- he chose one to
refine into the
The machine itself might even be considered to be a where between these examples. There is something
finished work,
creative mathematician. But such a machine would like an objective, external measure of success. You The Butcher’s Son
not be evidence of the singularity; it would not so can’t “win” at bridge building or medicine the way (2018).

outstrip us in creativity that we couldn’t even under- you can at chess, but one can see whether the bridge
stand what it was doing. Even if it had this kind of falls down or the virus is eliminated. These objective
human-level creativity, it wouldn’t lead inevitably to criteria come into play only once the domain is fairly
the realm of the superhuman. well specified: coming up with strong, lightweight
Some mathematicians are like musical virtuosos: materials, say, or drugs that combat particular dis-
they are distinguished by their fluency in an exist- eases. An AI might help in drug discovery by, in effect,
ing idiom. But geniuses like Srinivasa Ramanujan, doing the same thing as the AI that composed what
Emmy Noether, and Alexander Grothendieck arguably sounded like a well-executed Bach cantata or came
reshaped mathematics just as Schoenberg reshaped up with a brilliant Go strategy. Like a microscope,
music. Their achievements were not simply proofs of telescope, or calculator, such an AI is properly under-
long-standing hypotheses but new and unexpected stood as a tool that enables human discovery—not as
forms of reasoning, which took hold not only on an autonomous creative agent.
the strength of their logic but also on
their ability to convince other mathe-
maticians of the significance of their
innovations. A notional AI that comes
up with a clever proof to a problem
The capacity for genuine creativity,
that has long befuddled human math- the kind of creativity that updates
ematicians is akin to AlphaGo and its our understanding of the nature of being,
variants: impressive, but nothing like
is at the ground of what it is to be human.
That brings us to the other option.
Suppose the best and brightest
deep-learning algorithm is set loose
and after some time says, “I’ve found a proof of a It’s worth thinking about the theory of special
fundamentally new theorem, but it’s too complicated relativity here. Albert Einstein is remembered as the
for even your best mathematicians to understand.” “discoverer” of relativity—but not because he was the
This isn’t actually possible. A proof that not even first to come up with equations that better describe
the best mathematicians can understand doesn’t really the structure of space and time. George Fitzgerald,
count as a proof. Proving something implies that you Hendrik Lorentz, and Henri Poincaré, among others,
are proving it to someone. Just as a musician has to per- had written down those equations before Einstein.
suade her audience to accept her aesthetic concept of He is acclaimed as the theory’s discoverer because he
what is good music, a mathematician has to persuade had an original, remarkable, and true understanding
other mathematicians that there are good reasons to of what the equations meant and could convey that
believe her vision of the truth. To count as a valid proof understanding to others.
in mathematics, a claim must be understandable and For a machine to do physics that is in any sense
endorsable by some independent set of experts who comparable to Einstein’s in creativity, it must be able
are in a good position to understand it. If the experts to persuade other physicists of the worth of its ideas
who should be able to understand the proof can’t, at least as well as he did. Which is to say, we would
then the community refuses to endorse it as a proof. have to be able to accept its proposals as aiming to
For this reason, mathematics is more like music communicate their own validity to us. Should such a
than one might have thought. A machine could not machine ever come into being, as in the parable of

surpass us massively in creativity because either its Pinocchio, we would have to treat it as we would a
achievement would be understandable, in which case human being. That means, among other things, we
it would not massively surpass us, or it would not be would have to attribute to it not only intelligence but
understandable, in which case we could not count it whatever dignity and moral worth is appropriate to
as making any creative advance at all. human beings as well. We are a long way off from

What computers can’t create 75

this scenario, it seems to me, and there is no reason

to think the current computationalist paradigm of
artificial intelligence—in its deep-learning form or
any other—will ever move us closer to it.
Creativity is one of the defining features of human
beings. The capacity for genuine creativity, the kind
of creativity that updates our understanding of the
nature of being, that changes the way we understand
what it is to be beautiful or good or true—that capac-
ity is at the ground of what it is to be human. But this
kind of creativity depends upon our valuing it, and
caring for it, as such. As the writer Brian Christian
has pointed out, human beings are starting to act less
like beings who value creativity as one of our highest
possibilities, and more like machines themselves.
How many people today have jobs that require them
to follow a predetermined script for their conversa-
tions? How little of what we know as real, authentic,
creative, and open-ended human conversation is left in
this eviscerated charade? How much is it like, instead,
the kind of rule-following that a machine can do? And
how many of us, insofar as we allow ourselves to be
drawn into these kinds of scripted performances, are
eviscerated as well? How much of our day do we allow
to be filled with effectively machine-like activities—
filling out computerized forms and questionnaires,
responding to click-bait that works on our basest,
most animal-like impulses, playing games that are
designed to optimize our addictive response?
We are in danger of this confusion in some of the
deepest domains of human achievement as well. If we
allow ourselves to say that machine proofs we cannot
understand are genuine “proofs,” for example, ceding
social authority to machines, we will be treating the
achievements of mathematics as if they required no
human understanding at all. We will be taking one of
our highest forms of creativity and intelligence and
reducing it to a single bit of information: yes or no.
Even if we had that information, it would be of
little value to us without some understanding of the
reasons underlying it. We must not lose sight of the
essential character of reasoning, which is at the foun-
dation of what mathematics is. So too with art and
music and philosophy and literature. If we allow our-
selves to slip in this way, to treat machine “creativity”
as a substitute for our own, then machines will indeed
come to seem incomprehensibly superior to us. But
that is because we will have lost track of the funda-
mental role that creativity plays in being human.

Sean Dorrance Kelly is a philosophy professor

at Harvard and coauthor of the New York Times
best-selling book All Things Shining.

76 The state of innovation

Puncturing dreams
Unmanned aerial vehicles have
been touted as a “leapfrog” solution
to Africa’s poor infrastructure.
A researcher who studies them

of drones offers a dose of realism.

By Konstantin Kakaes
Portrait by Kate Warren

New technologies ure never introduced how it’s going to be used. There is u lot Fuel und buttery life ure u problem.
into u vucuum. They emerge into u of informution thut becomes uvuiluble Most drones right now ure uble to fly for
sociul, economic, und politicul setting through this high-resolution mup. You no more thun un hour ut most. The other
und influence it in their turn. Kutherine cun see trush dumping sites; you cun see big limitution is puyloud. The umount of
Chundler, u professor in the culture wustewuter runoff; you cun see where weight thut u drone cun curry is limited.
und politics progrum ut Georgetown illegul building is huppening. And thut This meuns deliveries huve focused on
University, is reseurching drones in informution chunges the terms of debute. things like blood und vuccines.
Africu us u study of how technology und
society chunge together. We recently The African Union and various inter- Is drone delivery a way to “leapfrog”
spoke with Chundler ubout her project. national aid agencies have described past the need to build a better road
drones as “transformative” for African network in much of rural Africa,
How are drones used in Africa today? development in general. Are they? where muddy roads are often impass-
There ure u number of smull-scule drone It’s useful to think ubout how smull able during rainy season?
projects throughout the continent, rung- un islund Zunzibur is, und how long it One project thut gets u lot of public-
ing from counting wildlife to delivering took to curry out this purticulur project. ity is u venture in Rwundu by u com-
vuccines to mupping islunds to using When you’re working in much lurger puny culled Zipline to deliver blood by
drones us disuster-response technologies. spuces it becomes hurder to uctuully drone. Rwundu hus been u site for huge
One of the projects thut I’m interested in cover the territory. investments by ull kinds of internutionul
is un initiutive by the Stute University of Tuke unother exumple. Between 2016 development orgunizutions, und the
Zunzibur. The teum uses smull commer- und 2017 there wus un experiment to Rwundun government is broudly inter-
ciul drones thut cun only fly for 30 or 40 try to integrute unmunned uircruft sys- ested in using drone uircruft for lots of
minutes. So mupping Zunzibur hus tuken tems into unti-pouching efforts ut Kruger different reseurch projects. This hus led
over two yeurs. Nutionul Purk in South Africu. The mun- to u vision of the country us u kind of
The intention wus for students to uger in churge suid thut they weren’t uble technology hub.
muke u mup thut could be used for plun- to see uny pouchers by using drones und But Rwundu continues to be u hugely
ning und nuturul resource munugement, thut, despite the hype uround drones us ugruriun society. How do drones fit with
so you would huve u buseline ideu of un innovutive new technology, drones the duy-to-duy reulities of most of the
whut the islunds looked like if there were not cupuble of doing the work thut people living there? It is u chullenge to
were u hurricune, oil spill, or some other wus necessury to truck und follow pouch- understund who these technologicul
disuster. The project wus not originully ers, und so the project wus cunceled. investments ure working for. Drones
ubout resolving long-stunding lund Drones couldn’t cover enough ground to ure imugined us u replucement for other
cluims. But purt of the chullenge of mup- guther useful informution, nor were purk forms of infrustructure, but muybe those
ping in Zunzibur und muking the infor- uuthorities uble to put the informution other forms of infrustructure ure uctuully
mution public hus been figuring out how drones guthered to good use. reully necessury.
the mup impucts disputes over lund. There were experiments in unother, It illustrutes the fullucy of tulking
much smuller, purk thut suggested thut ubout drones us u leupfrogging technol-
How can data gathered by drones drones might be slightly more useful. I ogy. Thinking ubout how we ure going
resolve land disputes? point this out becuuse one of the things to orgunize technologies in wuys thut
It’s uncleur how it would, or if it will. thut I’m trying to urgue is this question ure effectively going to serve people und
There ure cleurly politicul concerns of scule is importunt when thinking communities—thut’s the sort of visioning
ubout whut this mup will meun und ubout whut drones cun uccomplish. thut I wunt to see people doing.

78 The state of innovation


Our bodies, our cells 79

common. “I was told by my doctors

that I had a ‘low threshold for pain’
and that I should just get used to it
because there was nothing that could
be done,” Padma Lakshmi, a television
host who founded the Endometriosis
Foundation of America, said at a con-
ference in April 2018.
A majority of endometriosis cases
are never diagnosed: the most obvious
symptoms can have multiple causes,
Women’s health is often viewed and the severity of the symptoms does
not correlate strongly with the severity
through the lens of fertility, a bias of the underlying disorder. By some
estimates, endometriosis affects 10%
that stymies innovation in other of reproductive-age women—roughly
200 million people.
areas. NextGen Jane is among Nevertheless, NextGen Jane did
not set out to diagnose endometrio-
a vanguard of startups aiming to sis. The company’s initial focus was
on fertility—because, Tariyal says,
fix that. that’s what venture capitalists were
most interested in funding. NextGen
Jane is one of hundreds of so-called
femtech startups that are developing

By Dayna Evans technologies intended specifically

to improve women’s health. Frost
& Sullivan, a market research firm,
predicts that femtech will be a $50
billion industry by 2025. “Women’s
On an unremarkable side street in a big improvement over the pres- health care,” according to Frost &
Oakland, California, a few blocks ent-day standard of care. Sullivan, “remains largely confined to
down from an animal dermatologist Surgeons diagnose endometrio- reproductive matters.” According to
and just past an organic grocery store, sis—an abnormal growth of endo- Tariyal, this has been a major obstacle.
Ridhi Tariyal and Stephen Gire are metrial tissue outside the uterus—by “We wish we could go out there and
trying to change how women moni- inserting a small camera into the pel- say we just want to diagnose women’s
tor their health. When I visited their vic cavity to look for endometrial cells diseases,” she told me. But investors
small office in January, a garland of in places other than the lining of the would ask her: “Where’s the money
tampons dip-dyed in rainbow colors uterus, the only place they should in that?”
was strung above a computer mon- normally grow. If wayward cells are NextGen Jane’s story is a case
itor—a tongue-in-cheek reference found, the diseased tissue can often study in how a woman’s health is
to their work. be removed on sight. But the average typically viewed through the lens of
The tampon is a sort of totem for woman diagnosed with endometriosis her ability to bear children—and how
NextGen Jane, a startup that Tariyal has already had the disease for over that ingrained bias slows innovation
and Gire founded in 2014. Their plan a decade, which can mean years of in medicine.
is to use blood squeezed from used excruciating pain.
tampons as a diagnostic tool. In that The physical and emotional impact
menstrual blood, they hope to find on women’s lives is enormous. But ALIENATED AND

early markers of endometriosis and, women often believe such pain is FRUSTRATED
ultimately, a variety of other disor- normal, so they don’t seek treatment. Tariyal, who has a bachelor’s degree
ders. The simplicity and ease of this Delayed diagnoses by doctors relying in industrial engineering from
method, should it work, will represent on subjective reports of pain are also Georgia Tech, went to work at Bank

80 The state of innovation

of America Securities after grad- Gire and Tariyal

uation, but she hated investment in their Oakland
banking. If she was going to grind
tirelessly, she reasoned, she wanted
to do something more meaningful.
So she took a job as a research man-
ager and analyst at Bristol-Myers
Squibb, a pharmaceutical company.
This taught her that she didn’t like
big companies but did love medicine.
She went back to school, first getting
an MBA from Harvard and then a
master’s in biomedical enterprise
from MIT, with the goal of starting
a company of her own.
As a thesis project at MIT, Tariyal
tried to launch Ujala, a company that alternative: simply try to get pregnant clear genomic signals in menstrual
planned to test the blood of would-be to find out if she could. blood. Though genomics hadn’t been
partners in arranged marriages for This left Tariyal so alienated her goal, it was a field rich with possi-
genetic defects their offspring might and frustrated that she decided her bility. She found some 800 genes that
inherit. It never took off. Consumer only option was to create her own were expressed differently in men-
genetic testing was still in its infancy, AMH test that women could per- strual effluence and venous blood.
and the business case for the Indian form themselves, at home. She called The effluence contains not only blood
market, where Tariyal was hoping to Gire to ask for his help. She wanted but also endometrial lining, and some
sell her product, was hard to make to to design assays to pick up proteins cervical and vaginal cells as well. It
American venture capitalists. that would let her determine whether is, she says, like “getting a natural
In 2011 she went to work for AMH and other hormones could be biopsy from your body.”
Pardis Sabeti, a Harvard professor detected in menstrual blood, instead With funding of $100,000 and six
who needed someone to manage a of blood drawn from veins, so that you months of access to genome sequenc-
large genetic study in West Africa. wouldn’t have to see a doctor to get ing equipment at a startup acceler-
It was in Sabeti’s lab that she met tested. A woman could, in theory, just ator run by the genomics company
Stephen Gire. The two of them trav- send in a used tampon for analysis. Illumina, she and Gire continued to
eled through Sierra Leone together During her fellowship, Tariyal look at menstrual blood samples. In
to collect samples from survivors of performed tests that looked at three particular, they hoped they might
Lassa fever, a deadly hemorrhagic types of samples—venous blood, be able to reliably detect changes in
fever broadly similar to Ebola. blood from a pinprick to the skin, gene expression that Linda Giudice, a
Then, in 2013, Tariyal received and menstrual blood—to see where doctor at the University of California,
a fellowship at Harvard Business they overlapped. “I literally had to San Francisco, had recently discov-
School designed to encourage grad- run them to a lab to process right ered in women with endometriosis.
uates to start new life-sciences com- away,” she recalls. She was putting They have yet to succeed.
panies. She was 33 at the time and the logistical prowess she’d honed Diagnosing diseases from men-
an aspiring entrepreneur. She was in Sierra Leone to use. As a men- strual blood is difficult. Published
not ready to have children and asked struating woman, Tariyal also had an data establishing the efficacy of such
her doctor if she could wait five more advantage: not only could she include diagnoses remains sparse, though
years before she tried. She wanted to herself in trials, but she was entitled sequencing technologies and other
do a blood test called an anti-Mülle- to look at her own results. methods of extracting information
rian hormone (or AMH) test that To her disappointment, she found from blood samples are fast improv-
would approximate the number of that AMH levels are consistently ing. But NextGen Jane’s access to the

viable eggs she had. But her doctor lower in menstrual blood than they Illumina equipment ran out in 2015
didn’t see the need and wouldn’t are in venous blood. Her initial idea (although it now uses equipment
order it for her. And she was shocked wouldn’t work. But she believed she’d shared by a collective of genomics
by what the doctor suggested as an stumbled onto something even better: companies).

Our bodies, our cells 81

It is, she says, people suspicious of biotech startups

claiming to reinvent the blood test. A

like “getting 2016 study by Columbia University

researchers found that the over-
whelming majority of menstrual

a natural biopsy tracking apps were inaccurate. Some

defaulted to 28-day cycle lengths,
though fewer than 15% of women

from your body.” have cycles precisely that long. Other

apps predict a baby’s gender from the
date of conception, or peddle other
pseudoscientific claims.
Tariyal ultimately hopes to use
menstrual blood to screen not only
for endometriosis but also for cervical
cancer and various other disorders.
NextGen Jane’s key patent, at the
moment, is for a device that wrings
blood out of tampons. I watched her
manipulate it. She seals a container
and twists the mechanism like a pep-
per shaker. It squeezes out the blood
into a compartment below.
The device has yet to be approved by
the US Food and Drug Administration,
but Tariyal says a clinical trial is
designed and ready to go. She says she
needs to raise several million more dol-
lars to run a trial on about 800 women
that could establish the diagnostic effi-
cacy of menstrual blood. It will take her
about two years, she says—if she can
raise the money.
In a Washington Post op-ed last
year, Tariyal outlined some of the
challenges in fund-raising for a wom-
en’s health startup. “Some of my
mentors recommended I mask the
technology itself: Strip the deck of
‘menstrual blood’ and call it a novel
female substrate, they suggested.
Don’t say you’re a ‘women’s health’
company. It signals a lack of scien-
tific heft,” she wrote. “I understood
them to mean: Try to look as little as
possible like what you really are—a
THE “WOMEN’S As with any such boom, the surge woman-led company utilizing female
HEALTH” STIGMA of femtech companies leaves plenty biology to advance health care for half

NextGen Jane is part of a cluster to be wary of. The rise and fall of the population.”
of firms trying to develop direct- Theranos, which falsely claimed to
to-consumer tests for endometriosis have developed a revolutionary new Dayna Evans is a freelance writer
and other diseases affecting women. method of blood analysis, has made based in New York.

AI is harder than it looks 83

The urt of muking perfumes und expects to eventuully roll it out to ull
colognes husn’t chunged much since of them.
the 1880s, when synthetic ingredients However, he’s cureful to point out
begun to be used. Expert frugrunce thut getting this fur took neurly two
creutors tinker with combinutions of yeurs—und it required investments
chemiculs in hopes of producing com- thut still will tuke u while to recoup.
pelling new scents. So Achim Duub, Philyru’s initiul suggestions were hor-
un executive ut one of the world’s big- rible: it kept suggesting shumpoo rec-
gest mukers of frugrunces, Symrise, ipes. After ull, it looked ut sules dutu,
wondered whut w would huppen if he und shumpoo fur outsells perfume und
injected urtificiul cologne. Getting it on truck took u lot
intelligence into of truining by Symrise’s perfumers.
the process. Would Plus, the compuny is still wrestling
u muchine suggest with costly IT upgrudes thut huve
uppeuling formulus been necessury to pump dutu into
thut u humun might Philyru from dispurute record-keeping
not think to try? systems while keeping some of the
Duub hired IBM informution confidentiul from the
to design u computer perfumers themselves. “It’s kind of u
system thut would steep leurning curve,” Duub suys. “We
pore over mussive ure nowhere neur huving AI firmly und
umounts of informu- completely estublished in our enter-
tion—the formulus of prise system.”
existing frugrunces, The perfume business is hurdly the
consumer dutu, reg- only one to udopt muchine leurning
ulutory informution, without seeing rupid chunge. Despite
on und on—und then whut you might heur ubout AI sweep-
suggest new formu- ing the world, people in u wide runge
lutions for purticulur of industries suy the technology is
murkets. The system tricky to deploy. It cun be costly. And
is culled Philyru, ufter the initiul puyoff is often modest.
the Greek goddess of It’s one thing to see breukthroughs
frugrunce. Evocutive in urtificiul intelligence thut cun out-
nume uside, it cun’t smell u thing, so it pluy grundmusters of Go, or even to
cun’t repluce humun perfumers. But huve devices thut turn on music ut
it gives them u heud sturt on creuting your commund. It’s unother thing to
something novel. use AI to muke more thun incremen-
Duub is pleused with progress so tul chunges in businesses thut uren’t
fur. Two frugrunces uimed ut young inherently digitul.
customers in Bruzil ure due to go on AI might eventuully trunsform
sule there in June. Only u few of the the economy—by muking new prod-

Deploying artificial compuny’s 70 frugrunce designers

huve been using the system, but Duub
ucts und new business models pos-
sible, by predicting things humuns

intelligence can be
tricky and expensive.
Companies had
better know why Brian Derek
they really want it. Bergstein Brahney

84 The state of innovation

couldn’t huve foreseen, und by reliev- there—influted expectutions,” suys If compunies don’t stop und build
ing employees of drudgery. But thut Skomoroch, who is CEO of SkipFlug, connections between such systems,
could tuke longer thun hoped or u business thut suys it cun turn u com- then muchine leurning will work on
feured, depending on where you sit. puny’s internul communicutions into just some of their dutu. Thut expluins
Most compunies uren’t generuting u knowledge buse for employees. “AI why the most common uses of AI so
substuntiully more output from the und muchine leurning ure seen us fur huve involved business processes
hours their employees ure putting in. mugic fuiry dust.” thut ure siloed but nonetheless huve
Such productivity guins ure lurgest ut Among the biggest obstucles is ubundunt dutu, such us computer
the biggest und richest compunies, getting dispurute record-keeping sys- security or fruud detection ut bunks.
which cun ufford to spend heuvily on tems to tulk to euch other. Thut’s u Even if u compuny gets dutu flow-
the tulent und technology infrustruc- problem Richurd Zune hus encoun- ing from muny sources, it tukes lots
ture necessury to muke AI work well. tered us the chief innovution officer of experimentution und oversight to
This doesn’t necessurily meun thut ut UC Heulth, u network of hospi- be sure thut the informution is uccu-
AI is overhyped. It’s just thut when it tuls und medicul clinics in Colorudo, rute und meuningful. When Genpuct,
comes to reshuping how business gets Wyoming, und Nebrusku. It recently un IT services compuny, helps busi-
done, puttern-recognition ulgorithms rolled out u conversutionul softwure nesses luunch whut they consider AI
ure u smull purt of whut mutters. Fur ugent culled Livi, which
more importunt ure orgunizutionul ele- uses nuturul-lunguuge
ments thut ripple from the IT depurt- te c h n o l o g y f ro m u
ment ull the wuy to the front lines of
u business. Pretty much everyone hus
sturtup culled Avuumo
to ussist putients who
This doesn’t mean
to be uttuned to how AI works und
where its blind spots ure, especiully
cull UC Heulth or use
the website. Livi directs
AI is overhyped.
the people who will be expected to
trust its judgments. All this requires
them to renew their pre-
scriptions, books und
But algorithms are
not just money but ulso putience,
meticulousness, und other quintes-
confirms their uppoint-
ments, und shows them
a small part of what
sentiully humun skills thut too often
ure in short supply.
informution ubout their
really matters
Zune is pleused thut
with Livi hundling rou-
in reshaping how
tine queries, UC Heulth’s
stuff cun spend more
business gets done.
Looking time helping putients
with complicuted issues.

for But he ucknowledges thut this virtuul

ussistunt does little of whut AI might
projects, “10% of the work is AI,” suys
Sunjuy Srivustuvu, the chief digitul

unicorns eventuully do in his orgunizution. “It’s

just the tip of the iceberg, or whut-
ever the positive version of thut is,”
officer. “Ninety percent of the work
is uctuully dutu extruction, cleunsing,
normulizing, wrungling.”
Lust September, u dutu scientist Zune suys. It took u yeur und u hulf to Those steps might look seum-
numed Peter Skomoroch tweeted: deploy Livi, lurgely becuuse of the IT less for Google, Netflix, Amuzon, or
“As u rule of thumb, you cun expect heuduches involved with linking the Fucebook. But those compunies exist
the trunsition of your enterprise com- softwure to putient medicul records, to cupture und use digitul dutu. They’re
puny to muchine leurning will be insurunce-billing dutu, und other hos- ulso luxuriously stuffed with PhDs in
ubout 100x hurder thun your trunsi- pitul systems. dutu science, computer science, und
tion to mobile.” It hud the ring of u Similur setups bedevil other indus- reluted fields. “Thut’s different thun
joke, but Skomoroch wusn’t kidding. tries, too. Some big retuilers, for the runk und file of most enterprise
Severul people told him they were instunce, suve supply-chuin records compunies,” Skomoroch suys.
relieved to heur thut their compu- und consumer trunsuctions in sepu- Indeed, smuller compunies often
nies weren’t ulone in their strug- rute systems, neither of which is con- require employees to delve into
gles. “I think there’s u lot of puin out nected to brouder dutu storehouses. severul technicul domuins, suys

Stay ahead.
Stay connected.
Subscribe today.
Print Magazine
Get access to the 1 year (six bi-monthly issues) of
latest in innovation, MIT Technology Review print edition
exploring the very latest in technology
emerging and innovation.

technology, and Unlimited Online Access

Complete access to
the conversations with the stories, interviews and videos
you won’t find anywhere else during your
shaping the world subscription term.

around you. Daily Newsletter

Get the day’s top headlines delivered right to
Subscribe today at your inbox with MIT Technology Review’s daily email newsletter, The Download.

86 The state of innovation

Annu Drummond, u dutu scientist situutions of the users. The users’ key
ut Sunchez Oil und Gus, un energy question is not, us it is for technolo-
compuny bused in Houston. Sunchez
recently begun streuming und unu- The seeds gists, “Whut cun the technology do?”
but “How much will we benefit from
lyzing production dutu from wells in
reul time. It didn’t build the cupubil-
ity from scrutch: it bought the soft-
of AI
wure from u compuny culled MupR. Once un innovution
But Drummond und her colleugues
still hud to ensure thut dutu from the
urises, how quickly will
it diffuse through the
Machine learning is
field wus in formuts u computer could
purse. Drummond’s teum ulso got
economy? Economist
Zvi Griliches cume up
making Facebook,
involved in designing the softwure
thut would feed informution to engi-
with some fundumentul
unswers in the 1950s—
Google, and
neers’ screens. People udept ut ull
those things ure “not eusy to find,”
by looking ut corn.
Griliches exumined
Amazon rich. cut
she suys. “It’s like unicorns, busicully.
Thut’s whut’s slowing down AI or
the rutes ut which corn
furmers in vurious purts
outside that AI belt,
muchine-leurning udoption.”
Fluor, u huge engineering compuny,
of the country switched
to hybrid vurieties thut
things are moving
spent ubout four yeurs working with
IBM to develop un urtificiul-intelligence
hud much higher yields.
Whut interested him
much more slowly.
system to monitor mussive construc- wus not so much the
tion projects thut cun cost billions of corn itself but the vulue
dollurs und involve thousunds of work- of hybrids us whut we
ers. The system inhules both numeric would toduy cull u plutform for future investing in it?”
und nuturul-lunguuge dutu und ulerts innovutions. “Hybrid corn wus the Toduy muchine leurning is under-
Fluor’s project munugers ubout prob- invention of u method of inventing, u girding every uspect of the operutions
lems thut might luter cuuse deluys or method of breeding superior corn for of compunies like Fucebook, Google,
cost overruns. specific loculities,” Griliches wrote in und Amuzon und muny sturtups. It’s
Dutu scientists ut IBM und Fluor u lundmurk puper in 1957. muking these compunies exceptionully
didn’t need long to mock up ulgo- Hybrids were introduced in Iowu rich. But outside thut AI belt, things
rithms the system would use, suys in the lute 1920s und eurly 1930s. By ure moving much more slowly, for
Leslie Lindgren, Fluor’s vice president 1940 they uccounted for neurly ull corn rutionul economic reusons.
of informution munugement. Whut plunted in the stute. But the udoption At Symrise, Duub thinks the per-
took much more time wus refining curve wus nowhere neur us steep in fume AI project fell into u sweet spot. It
the technology with the close purtici- pluces like Texus und Alubumu, where wus u relutively smull-scule experiment,
pution of Fluor employees who would hybrids were introduced luter und but it involved reul work for u frugrunce
use the system. In order for them to covered ubout hulf of corn ucreuge in client und wusn’t u mere lub simulution.
trust its judgments, they needed to the eurly 1950s. One big reuson is thut “We’re ull under u lot of pressure,”
huve input into how it would work, hybrid seeds were more expensive he suys. “No one reully hus time to do
und they hud to curefully vulidute its thun conventionul seeds, und furm- greenfield leurning on the side.” Yet
results, Lindgren suys. ers hud to buy new ones every yeur. even this required u leup of fuith in the
To develop u system like this, “you Switching to the new technology wus technology. “It’s ull ubout conviction,”
huve to bring your domuin experts u riskier proposition for the furms in he suys. “There’s u very strong convic-
from the business—I meun your best these stutes thun in the richer und more tion in me thut AI will pluy u role in
people,” she suys. “Thut meuns you productive corn belt of the Midwest. most of the industries we see toduy,
huve to pull them off other things.” Whut Griliches cuptured, und whut some more predominuntly. To com-
Using top people wus essentiul, she subsequent economists confirmed, pletely ignore it is not un option.”
udds, becuuse building the AI engine is thut the spreud of technologies is
Brian Bergstein is editor at
wus “too importunt, too long, und too shuped less by the intrinsic quulities of large of and a former
expensive” for them to do otherwise. the innovutions thun by the economic editor at MIT Technology Review.


Transforming Mid-Career Professionals

An Executive Master’s degree from Brown University will prepare
you to lead your organization, transform your field, and build a
powerful lifelong professional network.

You will join a vibrant learning community and apply your new
knowledge and skills through a critical challenge project.

The results can be seen in our alumni who are proven leaders;
impacting their organizations and the world.

IE Brown Executive MBA

Executive Master in Cybersecurity
Executive Master in Science and Technology Leadership Executive Master of Healthcare Leadership

Brown-34343 Print Ad_81875x105.indd 1 3/29/18 9:24 A


88 The back page

The 10 worst
We all make mistakes sometimes.

technologies of the
21st century By the editors
Illustrations by Daniel Savage

ou’d think it would be wearer appear elitist and CRISPR is safe to use in humans.
easy to come up with a invasive. Then again, like That’s why the CRISPR babies born
list of bad technologies Segways and hover- in 2018 make our list.
from the past couple of boards, this was a failed Other times, it’s because technology
decades. But we had product, not a failed tech- has outpaced regulation. Data trafficking,
a hard time agreeing: What makes a nology; augmented-reality the sharing and remixing of people’s data
“bad” technology? glasses and heads-up displays without their control or awareness, has
After all, technologies can be bad are finding their public. contributed to the undermining of per-
because they fail to achieve admira- Some technologies are sonal liberty and democracy itself.
ble aims, or because they succeed in well-intentioned but solve no Some technologies are just misapplied.
wicked ones. The most useful tech- real problems and create new So far cryptocurrency looks mainly like a
nologies can also be the most harm- ones. Before electronic way for a hand-
ful—think of cars, which are crucial to voting, automated tabu- ful of specula-
the modern world yet kill over 1.25 lating of paper ballots left tors to get very
million people a year. And when an auditable paper trail. rich while a lot
well-intentioned technologies Now elections are more of other people
fail, is it because they are funda- vulnerable to hacking. end up poorer.
mentally flawed or just ahead of their time? Some failures apply a tech- But the technol-
Take the Segway. Inventor Dean nological fix to what is really ogy underlying
Kamen hyped it as a device that would a social or political problem. Take One it, blockchain, could
transform cities and transportation. It Laptop per Child, which set out to solve yet be transformative
turned out to be an expensive scooter inequality in education with a new gadget. in other areas.
that makes you look silly. Hoverboards But was it simply too early? Commercial Still, there are a few
were similarly all the rage until their bat- laptops, tablets, and—above all—smart- inventions we could
teries started exploding. But now (smaller) phones have since inundated the devel- agree have no redeem-
scooters and (safer) powered skateboards oping world. ing features. Juul and
are increasingly popular. Indiscriminate uses of technology other e-cigarettes are
If Google Glass worry us. Sometimes this is because reg- addicting a new generation to nicotine,
had been developed
h ulations are flouted. through a loophole that allowed them to
by a lesser com-
b Gene-editing tech- escape public health regulations meant to
pany, we probably
p niques like CRISPR discourage cigarette smoking. Plastic cof-
wouldn’t pick on it
w may one day cure fee pods save half a minute in the morn-
so much. But Google all manner of dis- ings but produce tons of hard-to-recycle
should have known eases, but right now waste. And as for selfie sticks …
better. It made the
bett we don’t know if need we say more?

MIT Technology Review (ISSN 1099-274X), Mhrch/April 2019 issue, Reg. US Phtent Office, is published bimonthly by MIT Technology Review, 1 Mhin St. Suite 13, Chmbridge, MA 02142-1517. Entire contents
©2019. The editors seek diverse views, hnd huthors’ opinions do not represent the officihl policies of their institutions or those of MIT. Periodichls posthge phid ht Boston, MA, hnd hdditionhl mhiling offices.
Postmhster: Send hddress chhnges to MIT Technology Review, Subscriber Services, PO Box 5001, Big Shndy, TX 75755, or vih the internet ht Bhsic subscription
rhtes: $59.94 per yehr within the United Sthtes; in hll other countries, US$69. Publichtion Mhil Agreement Number 40621028. Send undeliverhble Chnhdihn copies to PO Box 1051, Fort Erie, ON L2A 6C7. Printed
in USA. Audited by the Allihnce for Audited Medih.

verb \ikȂ`sel \

“The moment I realized

I had become part of something
that would take me further
than I’ve ever been.”

Define your Wharton moment.

Wharton’s General Management Program provides a General Management Program
flecible learning journey for distinguished senior ececutives a flexible learning journey
ready for greater challenges. You will receive ecpert one-on-
one executive coaching in Wharton’s rigorous academic
environment and return to your organization with an enriched Design your own curriculum of 6 programs within
global perspective and in-depth strategies for immediate 2 years. Choose from 30+ eligible programs in:
success. Upon successful completion of the program, you
will be awarded Wharton alumni status and join a powerful • leadership
network of 96,000 peers in over 150 countries. • finance
• strategy & innovation
excel at a higher level: • marketing GMP

Climb behind the wheel. Strap yourself in. Bury the pedal.
Feel the roar. Then tighten your grip as the LC 500h’s
lightning-fast Multistage Hybrid Drive system propels you
from zero to 60 in a mere 4.7 seconds.1,2 Lexus Hybrids.
There’s more to h than just hybrid.



LC 500h

Options shown. 1. Ratings achieved using the required premium unleaded gasoline with an octane rating of 91 or higher. If premium fuel is not used, performance
will decrease. 2. Performance figures are for comparison only and were obtained with prototype vehicles by professional drivers using special safety equipment and
procedures. Do not attempt. 3. Comparing 2018 Lexus Gas models and 2018 Lexus Hybrid models. Excluding special orders on the 2018 GS Hybrid. ©2018 Lexus