MIT Technology Review 11.12 2021

The
computing

issue

Volume 124 Nov/Dec USD $9.99
Number 6 2021 CAD $10.99



“Well, the basic idea is I break into a company’s network,
encrypt their files and hold the keys for ransom.”

Meet Ransomware

www.extrahop.com/advancedthreats

02 From the editor

“Hello, world!”

I’m Mat Honan, the new editor in chief of MITTechnology Review. Mat Honan ROBYN KESSLER
This is the first issue of the magazine I’ve had the pleasure of is editor in
working on. Maybe you have been reading Technology Review chief of
for years, like me. Or maybe this is your first issue. Either way, I’m MIT Technology
excited by the opportunity to make this magazine something you Review
look forward to reading every time it appears. I hope you enjoy
reading this issue as much as we enjoyed putting it together. about building true pathways into the industry for underrepre-
sented groups (page 24). Lakshmi Chandrasekaran’s examination
I want to start by making you a promise about Technology of the triumph of silicon over other seemingly fallow technol-
Review: We’re going to make it worth your time to read and worth ogies (remember spintronics?) shows how those alternatives
your money to subscribe. We’re going to bring you incredible may ultimately prove their worth (page 26). Meanwhile, Clive
stories about things at the edge of impossibility. We’ll expose Thompson brings us the story of ASML, the Dutch company
hidden truths, and hold the industries and people we cover to whose revolutionary process is keeping Moore’s Law alive, at
account. We’re going to help you understand the ways in which least for now (page 44).
science and technology are reshaping the world we all share.
We’re going to make you dream and wonder about the coming But it’s the future where things get weird and exciting. On page
years. We’re going to make you miss your stop because you just 38, you’ll meet Alán Aspuru-Guzik, who is combining artificial
can’t quit reading. intelligence and robotics in an attempt to accelerate materials dis-
covery—with the ultimate aim of solving really thorny problems
As you likely know, each issue has its own theme. This issue like climate change. And then there’s Antonio Regalado’s story on
is on computing—a topic so utterly central to what we cover it brain-computer interfaces (page 28). I frankly had to just sit down
seemed important to tackle it head on. for a little while and think after I finished reading it, imagining
a coming era that brings not only the ability to control machines
When I was young, personal computers were something with our minds, but also shared agency with an artificial neural
entirely new. They were vaguely mysterious—you had to know network. It’s wild stuff.
the language—and utterly fascinating. I spent countless hours
tinkering on the one in my mother’s home office, writing simple There is, of course, much more to explore within these pages.
programs, mapping out dungeons in Zork, and trying to under- I hope you also find something that grabs you by the collar and
stand the universe inside that box. makes you stop and think. And I hope to see you again soon. Let
me know! I always want to hear your feedback. You can reach me
Today computers are, obviously, everywhere—in every pocket on email at [email protected], or yell at me on
and automobile, even on the walls of our homes. And although Twitter, where I am @mat.
computers, and computing, have become far more ubiquitous
and accessible, their roles are often even more mysterious now Until next time,
than they were when I was a child in the 1980s. Virtually all Mat
aspects of modern life are now modulated by systems beyond
our control. This is not merely because the network or the ser-
vice or the algorithm is maintained by some unseen entity. As
Will Douglas Heaven notes on page 78, the very nature of how
computing works has changed with the rise of artificial intelli-
gence. We want to help demystify things a bit.

This issue explores how we arrived where we are, and where
we are going next. Margaret O’Mara’s sweeping introductory
essay (page 8) grounds the trajectory of computing in its greater
historical context. Siobhan Roberts’s exploration of the beguil-
ing P vs. NP question (page 56) traces the long road Sisyphean
researchers have traveled in trying to find a definitive answer.
Chris Turner’s review of A Biography of the Pixel (page 67) starts
by exploring the complex history of “Digital Light” and builds
to an unexpected, utterly delightful treatise on the triumph of
Steamed Hams. (You’re just going to have to read it.)

But history is meant to serve the present. Morgan Ames delves
into the hype around One Laptop per Child (page 74) to help
us find a better way toward ensuring that the most vulnerable
in our society receive true equity of access. Fay Cobb Payton,
Lynette Yarger, and Victor Mbarika explain how we can think



04 Contents

Introduction Report

8 15 22 24 25 26

The history of the Dispatches Immersion trip Pipelines, pathways, Setting an example What ever hap-
future of computing Computing advances Is the metaverse and diversity A computer science pened to DNA
If you want to know are changing the way just the latest statements educator on why computing?
what will come next we predict deadly internet catchphrase, This is how to she’s talking about Silicon is king. But
in computing, look to storms, speed or can it change how actually diversify identity in the don’t give up on
the past. deliveries, make we approach real tech. classroom. these other contend-
By Margaret O’Mara music, and study problems? By Fay Cobb Payton, By Melba Newsome ers for the throne.
the origins of the By John David N. Lynette Yarger, and By Lakshmi
universe. Dionisio Victor Mbarika Chandrasekaran

Features

28 36 38

A computer mouse inside your head How to fight online hate Reimagining materials discovery

Paralyzed people are using brain The dean of UC Berkeley’s School of Faced with the need for better materials
implants to type and move robotic arms. Information on how computing and data to fight climate change, Alán Aspuru-
Is this the next great computer interface science can identify and mitigate bias. Guzik wants to digitize the discovery
for all of us? By Antonio Regalado By Anil Ananthaswamy process. By Simon Lewsen

44 56 64

The world’s most complex machine The problem to end all problems The transformation of computing

Moore’s Law was grinding to a halt— P vs. NP has been called a fad, a punch- We’re fixated on making faster machines.
until a small Dutch firm pushed the ing bag, and an alibi; it is among The executive vice president for research
limits of physics. By Clive Thompson theoretical computer science’s hardest at Columbia University wants us to fixate
problems. By Siobhan Roberts on the data. By Anil Ananthaswamy

Review Fiction The back page

67 71 74 78 80 88

Awash in digital Don’t get left behind Laptops alone AI is reinventing Be a good example Computation
light A new book predicts can’t bridge the what computers are By April Sopkin evolution
A Biography of the a coming age of digital divide The essence
Pixel shows how digi- technological The failures of One of computing Cover illustration
tization has changed abundance. The Laptop per Child is undergoing a by Saiman Chow
our visual lives. reality is a lot more have plenty to teach fundamental shift.
By Chris Turner complicated. us about real reform. By Will Douglas
By David Rotman By Morgan Ames Heaven

SAVE THE DATE

MIT Technology Review’s
signature AI conference
returns in person
March 2022.

JOIN US IN CAMBRIDGE, MA
OR ONLINE

March 29-30, 2022

SUBSCRIBERS SAVE 10%
WITH CODE PRINTND21 AT

EmTechDigital.com/Register

06 Masthead

Editorial Corporate Consumer marketing MIT Technology Review Insights
and international
Editor in chief Chief executive officer and publisher Senior vice president,
Mat Honan Elizabeth Bramson-Boudreau marketing and consumer revenue Vice president, Insights and international
Doreen Adger Nicola Crepaldi
Executive editor Assistant to the CEO
Michael Reilly Katie McLean Director of digital marketing Director of custom content, US
Emily Baillieu Laurel Ruma
Editor at large Manager of information technology
David Rotman Colby Wheeler Director of event marketing Senior project manager
Brenda Noiseux Martha Leibs
News editor Office manager
Niall Firth Linda Cardinal Email marketing manager Content manager
Tuong-Chau Cai Jason Sparapani
Managing editor Technology
Timothy Maher Growth marketing manager Senior manager of licensing
Chief technology officer Em Okrepkie Ted Hu
Commissioning editors Drake Martinet
Bobbie Johnson Assistant consumer marketing manager Director of custom content, international
Konstantin Kakaes Head of product Caroline da Cunha Francesca Fanshawe
Amy Nordrum Mariya Sitnova
Rachel Courtland Circulation and print production manager Director of business development, Asia
Senior software engineer Tim Borton Marcus Ulvne
Senior editor, MIT News Molly Frey
Alice Dragoon Advertising sales Board of directors
UI product designer
Senior editor, biomedicine Rachel Stein Senior vice president, sales and Martin A. Schmidt, Chair
Antonio Regalado brand partnerships Peter J. Caruso II, Esq.
Associate product manager Andrew Hendler Whitney Espich
Senior editor, climate and energy Allison Chase [email protected] Jerome I. Friedman
James Temple 646-520-6981 David Schmittlein
Software engineer Glen Shor
Senior editor, digital culture Jack Burns Executive director, integrated marketing Alan Spoon
Abby Ohlheiser Caitlin Bergmann
Head of data and analytics [email protected] Customer service and
Senior editor, cybersecurity Danielle Golds subscription inquiries
Patrick Howell O’Neill Executive director, brand partnerships
Product management intern Marii Sebahar National
Senior editors, AI Yufei Zheng [email protected] 877-479-6505
Karen Hao 415-416-9140
Will Douglas Heaven Events International
Executive director, brand partnerships 847-559-7313
Senior editor, computing Senior vice president, Kristin Ingram
Siobhan Roberts events and strategic partnerships [email protected] Email
Amy Lammers 415-509-1910 [email protected]
Senior editor, podcasts and
live journalism Director of event content Senior director, brand partnerships Web
Jennifer Strong and experiences Debbie Hanley www.technologyreview.com/
Brian Bryson [email protected] customerservice
Podcast producer 214-282-2727
Anthony Green Head of international and custom events MIT Records (alums only)
Marcy Rizzo Senior director, brand partnerships 617-253-8270
Editor, Pandemic Technology Project Ian Keller
Lindsay Muscato Senior event content producer [email protected] Reprints
Kristen Kavanaugh 203-858-3396 [email protected]
Senior reporters 877-652-5295
Tanya Basu (humans and technology) Event content producer Senior director, brand partnerships
Eileen Guo (technology policy and ethics) Erin Underwood Miles Weiner Licensing and permissions
[email protected] [email protected]
Reporters Associate director of events 617-475-8078
Charlotte Jee (news) Nicole Silva MIT Technology Review
Tate Ryan-Mosley (data and audio) Digital sales strategy manager
Event operations manager Casey Sullivan One Main Street
Copy chief Elana Wilner [email protected] 13th Floor
Linda Lowenthal 617-475-8066 Cambridge, MA 02142
Event partnership coordinator 617-475-8000
Social media editor Madeleine Frasca Media kit
Benji Rosen www.technologyreview.com/media The mission of MIT Technology Review
Events associate is to make technology a greater force for
Engagement editor Bo Richardson good by bringing about better-informed,
Abby Ivory-Ganja more conscious technology decisions
Finance through authoritative, influential, and
Proofreader trustworthy journalism.
Barbara Wallraff Vice president, finance
Enejda Xheblati Technology Review, Inc., is an
Editorial fellows independent nonprofit 501(c)(3)
Casey Crownhart General ledger manager corporation wholly owned by MIT;
Tatyana Woodall Olivia Male the views expressed in our publications
and at our events are not always shared by
Design Accountant the Institute.
Anduela Tabaku
Chief creative officer
Eric Mongeon

Art director
Emily Luong

Marketing and events designer
Kyle Thomas Hemingway

Photo editor
Stephanie Arnett

Artificial
intelligence,
demystified

The Algorithm Exclusively for
newsletter MIT Technology Review
subscribers.

technologyreview.com/algorithm

08 Computing

INTRODUCTION If the future of computing is anything
like its past, then its trajectory will
If you want to know what It matters who is inventing depend on things that have little to
do with computing itself.
will come next in computing, the breakthroughs and why. Technology does not appear
from nowhere. It is rooted in time,
look to the past. By Margaret O’Mara place, and opportunity. No lab is
an island; machines’ capabilities
The history and constraints are determined
of the not only by the laws of physics
future of and chemistry but by who supports those
computing technologies, who builds them, and where
they grow.
1961: IBM engineers give visitors from Ames Research Popular characterizations of computing NASA AMES/INTERNET ARCHIVE
Center a look at the future. have long emphasized the quirkiness and
brilliance of those in the field, portray-
ing a rule-breaking realm operating off on
its own. Silicon Valley’s champions and
boosters have perpetuated the mythos of
an innovative land of garage startups and
capitalist cowboys. The reality is different.
Computing’s history is modern history—and
especially American history—in miniature.
The United States’ extraordinary push
to develop nuclear and other weapons
during World War II unleashed a torrent
of public spending on science and tech-
nology. The efforts thus funded trained a
generation of technologists and fostered
multiple computing projects, including
ENIAC—the first all-digital computer,
completed in 1946. Many of those funding
streams eventually became permanent,
financing basic and applied research at a
scale unimaginable before the war.
The strategic priorities of the Cold
War drove rapid development of transis-
torized technologies on both sides of the
Iron Curtain. In a grim race for nuclear
supremacy amid an optimistic age of sci-
entific aspiration, government became
computing’s biggest research sponsor and
largest single customer. Colleges and uni-
versities churned out engineers and scien-
tists. Electronic data processing defined
the American age of the Organization Man,
a nation built and sorted on punch cards.
The space race, especially after the
Soviets beat the US into space with the
launch of the Sputnik orbiter in late 1957,
jump-started a silicon semiconductor
industry in a sleepy agricultural region

Introduction 09

of Northern California, eventually shift- the internet’s vast expanse. Wall Street The technology industry—produced
ing tech’s center of entrepreneurial grav- clamored for tech stocks, then didn’t; for- and made wealthy by these immense
ity from East to West. Lanky engineers tunes were made and lost in months. After advances in computing—has failed to
in white shirts and narrow ties turned the bust, new giants emerged. Computers imagine alternative futures both bold and
giant machines into miniature electronic became smaller: a smartphone in your practicable enough to address humanity’s
ones, sending Americans to the moon. (Of pocket, a voice assistant in your kitchen. gravest health and climatic challenges.
course, there were also women playing They grew larger, into the vast data banks Silicon Valley leaders promise space colo-
key, though often unrecognized, roles.) and sprawling server farms of the cloud. nies while building grand corporate head-
quarters below sea level. They proclaim
In 1965, semiconductor pioneer Gordon Fed with oceans of data, largely unfet- that the future lies in the metaverse, in
Moore, who with colleagues had broken tered by regulation, computing got smarter. the blockchain, in cryptocurrencies whose
ranks with his boss William Shockley Autonomous vehicles trawled city streets, energy demands exceed those of entire
of Shockley Semiconductor to launch a humanoid robots leaped across laborato- nation-states.
new company, predicted that the num- ries, algorithms tailored social media feeds
ber of transistors on an integrated cir- and matched gig workers to customers. The future of computing feels more
cuit would double every year while costs Fueled by the explosion of data and compu- tenuous, harder to map in a sea of infor-
would stay about the same. Moore’s Law tation power, artificial intelligence became mation and disruption. That is not to say
was proved right. As computing power the new new thing. Silicon Valley was no that predictions are futile, or that those who
became greater and cheaper, digital innards longer a place in California but shorthand build and use technology have no control
replaced mechanical ones in nearly every- for a global industry, although tech wealth over where computing goes next. To the
thing from cars to coffeemakers. and power were consolidated ever more contrary: history abounds with examples
tightly in five US-based companies with of individual and collective action that
A new generation of computing inno- a combined market capitalization greater altered social and political outcomes. But
vators arrived in the Valley, beneficiaries than the GDP of Japan. there are limits to the power of technol-
of America’s great postwar prosperity but ogy to overcome earthbound realities of
now protesting its wars and chafing against It was a trajectory of progress and wealth politics, markets, and culture.
its culture. Their hair grew long; their shirts creation that some believed inevitable and
stayed untucked. Mainframes were seen enviable. Then, starting two years ago, To understand computing’s future, look
as tools of the Establishment, and achieve- resurgent nationalism and an economy- beyond the machine.
ment on earth overshadowed shooting upending pandemic scrambled supply
for the stars. Small was beautiful. Smiling chains, curtailed the movement of peo- The hoodie problem
young men crouched before home-brewed ple and capital, and reshuffled the global
desktop terminals and built motherboards order. Smartphones recorded death on the 1 First, look to who will get to build
in garages. A beatific newly minted mil- streets and insurrection at the US Capitol.
lionaire named Steve Jobs explained how AI-enabled drones surveyed the enemy the future of computing.
a personal computer was like a bicycle for from above and waged war on those below.
the mind. Despite their counterculture Tech moguls sat grimly before congressio- The tech industry long celebrated itself
vibe, they were also ruthlessly competitive nal committees, their talking points ring- as a meritocracy, where anyone could get
businesspeople. Government investment ing hollow to freshly skeptical lawmakers. ahead on the strength of technical know-
ebbed and private wealth grew. how and innovative spark. This assertion
Our relationship with computing had has been belied in recent years by the per-
The ARPANET became the commercial suddenly changed. sistence of sharp racial and gender imbal-
internet. What had been a walled garden ances, particularly in the field’s topmost
accessible only to government-funded The past seven decades have produced ranks. Men still vastly outnumber women
researchers became an extraordinary new stunning breakthroughs in science and in the C-suites and in key engineering
platform for communication and business, engineering. The pace and scale of change roles at tech companies. Venture capital
as the screech of dial-up modems con- would have amazed our mid-20th-century investors and venture-backed entrepre-
nected millions of home computers to the forebears. Yet techno-optimistic assur- neurs remain mostly white and male. The
World Wide Web. Making this strange and ances about the positive social power number of Black and Latino technologists
exciting world accessible were very young of a networked computer on every desk of any gender remains shamefully tiny.
companies with odd names: Netscape, have proved tragically naïve. The infor-
eBay, Amazon.com, Yahoo. mation age of late has been more effec- Much of today’s computing innovation
tive at fomenting discord than advancing was born in Silicon Valley. And looking
By the turn of the millennium, a presi- enlightenment, exacerbating social ineq- backward, it becomes easier to understand
dent had declared that the era of big gov- uities and economic inequalities rather where tech’s meritocratic notions come
ernment was over and the future lay in than transcending them. from, as well as why its diversity problem
has been difficult to solve.

10 Computing

INTRODUCTION

Silicon Valley was once indeed a place without enough attention to the human the business of war, but their objections
where people without family money or con- biases contained in that data. It has pro- have done little to slow the growing stream
nections could make a career and possibly duced social media platforms that have of multibillion-dollar contracts for cloud
a fortune. Those lanky engineers of the fueled political disruption and violence computing and cyberweaponry. It is almost
Valley’s space-age 1950s and 1960s were at home and abroad. It has left rich areas as if Silicon Valley is returning to its roots.
often heartland boys from middle-class of research and potentially vast market
backgrounds, riding the extraordinary opportunities neglected. Defense work is one dimension of the
escalator of upward mobility that America increasingly visible and freshly contentious
delivered to white men like them in the Computing’s lack of diversity has always entanglement between the tech industry
prosperous quarter-century after the end been a problem, but only in the past few and the US government. Another is the
of World War II. years has it become a topic of public con- growing call for new technology regulation
versation and a target for corporate reform. and antitrust enforcement, with potentially
Many went to college on the GI Bill That’s a positive sign. The immense wealth significant consequences for how techno-
and won merit scholarships to places like generated within Silicon Valley has also logical research will be funded and whose
Stanford and MIT, or paid minimal tuition created a new generation of investors, interests it will serve.
at state universities like the University of including women and minorities who are
California, Berkeley. They had their pick deliberately putting their money in com- The extraordinary consolidation of
of engineering jobs as defense contracts panies run by people who look like them. wealth and power in the technology sec-
fueled the growth of the electronics indus- tor and the role the industry has played
try. Most had stay-at-home wives whose But change is painfully slow. The market in spreading disinformation and sparking
unpaid labor freed husbands to focus their will not take care of imbalances on its own. political ruptures have led to a dramatic
energy on building new products, com- change in the way lawmakers approach
panies, markets. Public investments in For the future of computing to include the industry. The US has had little appetite
suburban infrastructure made their cost more diverse people and ideas, there needs for reining in the tech business since the
of living reasonable, the commutes easy, to be a new escalator of upward mobility: Department of Justice took on Microsoft
the local schools excellent. Both law and inclusive investments in research, human 20 years ago. Yet after decades of bipar-
market discrimination kept these suburbs capital, and communities that give a new tisan chumminess and laissez-faire tol-
nearly entirely white. generation the same assist the first gen- erance, antitrust and privacy legislation
eration of space-age engineers enjoyed. is now moving through Congress. The
In the last half-century, political change The builders cannot do it alone. Biden administration has appointed some
and market restructuring slowed this esca- of the industry’s most influential tech
lator of upward mobility to a crawl, right Brainpower monopolies critics to key regulatory roles and has
at the time that women and minorities pushed for significant increases in regu-
finally had opportunities to climb on. By 2 Then, look at who the industry’s cus- latory enforcement.
the early 2000s, the homogeneity among
those who built and financed tech prod- tomers are and how it is regulated. The five giants—Amazon, Apple,
ucts entrenched certain assumptions: Facebook, Google, and Microsoft—now
that women were not suited for science, The military investment that undergirded spend as much or more lobbying in
that tech talent always came dressed in a computing’s first all-digital decades still Washington, DC, as banks, pharmaceu-
hoodie and had attended an elite school— casts a long shadow. Major tech hubs of tical companies, and oil conglomerates,
whether or not someone graduated. It today—the Bay Area, Boston, Seattle, Los aiming to influence the shape of antici-
limited thinking about what problems to Angeles—all began as centers of Cold War pated regulation. Tech leaders warn that
solve, what technologies to build, and what research and military spending. As the breaking up large companies will open a
products to ship. industry further commercialized in the path for Chinese firms to dominate global
1970s and 1980s, defense activity faded from markets, and that regulatory intervention
Having so much technology built by a public view, but it hardly disappeared. For will squelch the innovation that made
narrow demographic—highly educated, academic computer science, the Pentagon Silicon Valley great in the first place.
West Coast based, and disproportion- became an even more significant benefactor
ately white, male, and young—becomes starting with Reagan-era programs like the Viewed through a longer lens, the polit-
especially problematic as the industry Strategic Defense Initiative, the computer- ical pushback against Big Tech’s power is
and its products grow and globalize. It has enabled system of missile defense memo- not surprising. Although sparked by the
fueled considerable investment in driver- rably nicknamed “Star Wars.” 2016 American presidential election, the
less cars without enough attention to the Brexit referendum, and the role social
roads and cities these cars will navigate. In the past decade, after a brief lull media disinformation campaigns may have
It has propelled an embrace of big data in the early 2000s, the ties between the played in both, the political mood echoes
technology industry and the Pentagon one seen over a century ago.
have tightened once more. Some in
Silicon Valley protest its engagement in

Introduction 11

We might be looking at a tech future Tech firms may decry government ground to Asian competitors in the chip
where companies remain large but regu- intervention as antithetical to their abil- market for years. The economy-choking
lated, comparable to the technology and ity to innovate. But follow the money, consequences of this became painfully
communications giants of the middle part and the regulation, and it is clear that clear when covid-related shutdowns
of the 20th century. This model did not the public sector has played a critical slowed chip imports to a trickle, throttling
squelch technological innovation. Today, it role in fueling new computing discover- production of the many consumer goods
could actually aid its growth and promote ies—and building new markets around that rely on semiconductors to function.
the sharing of new technologies. them—from the start.
As when Japan posed a competitive
Take the case of AT&T, a regulated Location, location, location threat 40 years ago, the American agitation
monopoly for seven decades before its over China runs the risk of slipping into
ultimate breakup in the early 1980s. In 3 Last, think about where the corrosive stereotypes and lightly veiled
exchange for allowing it to provide univer- business of computing happens. xenophobia. But it is also true that com-
sal telephone service, the US government puting technology reflects the state and
required AT&T to stay out of other com- The question of where “the next Silicon society that makes it, whether it be the
munication businesses, first by selling its Valley” might grow has consumed politi- American military-industrial complex of
telegraph subsidiary and later by steering cians and business strategists around the the late 20th century, the hippie-influenced
clear of computing. world for far longer than you might imag- West Coast culture of the 1970s, or the
ine. French president Charles de Gaulle communist-capitalist China of today.
Like any for-profit enterprise, AT&T toured the Valley in 1960 to try to unlock
had a hard time sticking to the rules, espe- its secrets. Many world leaders have fol- What’s next
cially after the computing field took off in lowed in the decades since.
the 1940s. One of these violations resulted Historians like me dislike making predic-
in a 1956 consent decree under which the Silicon Somethings have sprung up tions. We know how difficult it is to map
US required the telephone giant to license across many continents, their gleaming the future, especially when it comes to
the inventions produced in its industrial research parks and California-style sub- technology, and how often past forecast-
research arm, Bell Laboratories, to other divisions designed to lure a globe-trotting ers have gotten things wrong.
companies. One of those products was the workforce and cultivate a new set of tech
transistor. Had AT&T not been forced to entrepreneurs. Many have fallen short of Intensely forward-thinking and impa-
share this and related technological break- their startup dreams, and all have fallen tient with incrementalism, many modern
throughs with other laboratories and firms, short of the standard set by the original, technologists—especially those at the
the trajectory of computing would have which has retained an extraordinary ability helm of large for-profit enterprises—are
been dramatically different. to generate one blockbuster company after the opposite. They disdain politics, and
another, through boom and bust. resist getting dragged down by the realities
Right now, industrial research and devel- of past and present as they imagine what
opment activities are extraordinarily con- While tech startups have begun to lies over the horizon. They dream of a new
centrated once again. Regulators mostly appear in a wider variety of places, about age of quantum computers and artificial
looked the other way over the past two three in 10 venture capital firms and close general intelligence, where machines do
decades as tech firms pursued growth at to 60% of available investment dollars most of the work and much of the thinking.
all costs, and as large companies acquired remain concentrated in the Bay Area. After
smaller competitors. Top researchers left more than half a century, it remains the They could use a healthy dose of his-
academia for high-paying jobs at the tech center of computing innovation. torical thinking.
giants as well, consolidating a huge amount
of the field’s brainpower in a few companies. It does, however, have significant com- Whatever computing innovations will
petition. China has been making the kinds appear in the future, what matters most is
More so than at any other time in Silicon of investments in higher education and how our culture, businesses, and society
Valley’s ferociously entrepreneurial history, advanced research that the US government choose to use them. And those of us who
it is remarkably difficult for new entrants made in the early Cold War, and its tech- analyze the past also should take some
and their technologies to sustain meaning- nology and internet sectors have produced inspiration and direction from the tech-
ful market share without being subsumed enormous companies with global reach. nologists who have imagined what is not
or squelched by a larger, well-capitalized, yet possible. Together, looking forward
market-dominant firm. More of comput- The specter of Chinese competition and backward, we may yet be able to get
ing’s big ideas are coming from a handful has driven bipartisan support for renewed where we need to go.
of industrial research labs and, not surpris- American tech investment, including a
ingly, reflecting the business priorities of potentially massive infusion of public sub- Margaret O’Mara is a historian at the
a select few large tech companies. sidies into the US semiconductor industry. University of Washington and author
American companies have been losing of The Code: Silicon Valley and the
Remaking of America.

RANSOMWARE

WITH THE RISE Cyberattacks are becoming a
matter of national security for
of ransomware and government bodies across
cyberattacks in the the world. E L V I S C H A N
headlines week after week, of the FBI shares a bold
fighting cybercrime should strategy and its results.
be a priority for every
business leader.

REGISTER FOR

CYBERSECURE

AND SAVE

Meet the ransomware
slayer, F A B I A N W O S A R
of Emsisoft, in a one-
of-a-kind, protected
conversation

JOIN US LIVE ONLINE

November 16 & 17, 2021

Over 80% of data breaches
leverage W E A K O R
STOLEN PASSWORDS.
Explore how options like
biometrics could eliminate
the need for passwords.

Gain practical guidance on
finding and fixing existing
security vulnerabilities from
CAROLINE WONG,
CSO at Cobalt.

Get a firsthand account
of the landmark SolarWinds
cyberattack from SolarWinds
CISO & VP of Security
TIMOTHY BROWN.

EVERYWHERE

SUBSCRIBERS SAVE 10%

Use code PRINTND21 at CyberSecureMIT.com/RegisterNow

Let
the story
come to
you.

Tech news delivered straight to your inbox.

W E E K D AY S FRIDAYS (SUBSCRIBER EXCLUSIVE)

The Download The Algorithm

Navigate the world of tech news Artificial intelligence, demystified

W E E K LY S AT U R D AY S

Coronavirus Tech Report Weekend Reads

How covid-19 is changing our world Technology in perspective

W E D N E S D AY S M O N T H LY

The Airlock fwd: Economy

Your gateway to the future Your guide to growth and prosperity
of space technology in the age of technology

technologyreview.com/inbox

“Early in my career, a lot of times things RE 15
would strike without warning, because the
technology wasn’t that good ... You don’t see PVOI E WT
that as much anymore.”
Get upBtoooksps,eed on
the kaerytst,haemndesc,ubltiguriedeas,

anidn mpearjsoprepcltaivyeers
discussed in this issue.

WEATHER-WISE storms might hit. And by the end
of 2021, the US National Weather
Supercomputers have made it easier for meteorologists to Casey Service (NWS) will receive two
predict deadly storms. But climate change makes it harder Crownhart is brand-new supercomputers. It’s an
to get out of harm’s way. a journalism upgrade they hope will continue the
fellow at MIT steady march toward more accurate
hen Hurricane Michael Predicting the chaos that is the forecasts, which will become even
W made landfall on the center of a hurricane, and under- Technology more essential as climate change con-
standing how storms strengthen, Review. tinues to fuel more intense storms.
Gulf Coast of Florida is still a challenge for forecasters.
DANIEL ZENDER in October 2018, it was a cate- But armed with better models The agency will use the new
gory 5 storm, with wind speeds and more experience, they accu- machines in operational forecast-
over 150 miles per hour. The US rately predicted that Hurricane ing—the system that forecasters use
National Hurricane Center had Ida, which hit New Orleans in to make predictions like the ones on
initially predicted they would September this year, would rap- the nightly news. Once the agency
reach less than half that. idly intensify, although the storm has fully vetted them, probably in
strengthened even more than July 2022, the new supercomputers
Michael went through a pro- they had expected. should help meteorologists better
cess called rapid intensification, predict everything from the chance
where a hurricane develops Supercomputers have been of rain in Denver to the odds that a
massively higher wind speeds part of these improvements in hurricane will hit Miami.
in a short time. And the experts predicting where, when, and how
didn’t see it coming. Each supercomputer (one in
Virginia and one in Arizona, so
there’s always a backup) is about
the size of 10 refrigerators and has
a capacity of 12.1 petaflops. “Flops”
stands for “floating point operations
per second,” so 12.1 petaflops means
the supercomputers can make just
over 12 quadrillion calculations every
second. It’s a huge upgrade—nearly
triple the size of the old system—
and will cost roughly $300 million to
$500 million over the next decade.

Computing capacity upgrades are
a big piece of recent improvements
in forecasts of hurricane path and
intensity, says Michael Brennan,
head of the Hurricane Specialist Unit
at the National Hurricane Center.

16 Computing

Forecasts like the ones Brennan’s 15 NATIONAL WEATHER SERVICE SUPERCOMPUTING CAPACITY,
team releases are made by humans IN PETAFLOPS
who sort through different models
and decide how to synthesize the 12
information.
9
Projections of hurricane paths
have gotten steadily more accurate 6
over the past 30 years as large-scale
weather models, and the comput- 3
ers running them, have improved.
Average errors in hurricane path 2009 2013 2015 2016 2018 2020
predictions dropped from about 100
miles in 2005 to about 65 miles in Each of the NWS geographically precise. Weather The changes, and the resulting NATIONAL WEATHER SERVICE
2020. The difference might seem supercomputers, models work by splitting the globe forecast accuracy, have made life
small when storms can be hundreds up into a bunch of pieces and try- easier for officials like Jim Stefkovich,
of miles wide, but when it comes to at ing to calculate what will happen in a meteorologist for the Alabama
predicting where the worst effects each of them. A higher-resolution Emergency Management Agency
from a hurricane will hit, “every lit- 12.1 model will break up the globe into who helps the state government pre-
tle wiggle matters,” Brennan says. PETA- smaller fragments, which means pare for weather risks. “Early in my
FLOPS, there are more of them to consider. career, a lot of times things would
Understanding and predicting strike without warning, because the
hurricanes’ intensity has been more is about Finally, researchers can put technology wasn’t that good, and the
challenging than predicting their together what’s called an ensemble warnings weren’t always that accu-
paths, since the strength of a hurri- 150,000 model, which they run as many as rate,” Stefkovich says. “You don’t see
cane is driven by more local factors, TIMES 20 or 30 times. Each of those runs that as much anymore.”
like wind speed and temperature at is performed under slightly differ-
the center of a storm. Still, inten- more powerful ent conditions to see how the pre- But even today, signals from
sity predictions have also started to than an ordinary dictions differ. The results are then weather forecasters aren’t always
improve in the past decade. Errors tallied up and considered together. clear to the public, which may not be
in the intensity forecast within 48 laptop. tuned in for every storm, or may not
hours decreased by 20% to 30% You’ve probably seen ensemble understand the difference between a
between 2005 and 2020. models in hurricane predictions. storm watch (hazardous conditions
Consider a storm that starts in the are expected) and a storm warning
Powering predictions Atlantic Ocean. If an ensemble fore- (they’ve already been observed). If
When building models to predict cast contains a wide variety of results, people don’t know how to react,
something as complicated as the with the storm heading to Texas in a better forecast won’t really help
weather, “it’s easy to suck up addi- some and skirting up the East Coast them stay safe, Stefkovich says.
tional computer resources,” says in others, it has many plausible paths.
Brian Gross, director of the NWS But if the ensemble members all This happened during Ida, he
Environmental Modeling Center. show the storm hitting Florida’s Gulf says. The forecasts for the storm’s
Coast, forecasters can be more cer- path and its strength were accu-
Models can benefit from com- tain about where it will land. rate a few days before. But partially
puting power in multiple ways, and because of poor communication,
each model can quickly slurp up Previous supercomputer dozens of people lost their lives,
huge amounts of capacity. A model upgrades have led to improvements and millions lost power or saw their
can get more complex by digesting in multiple areas for some models, homes or cars damaged.
more information or by using more like the all-purpose Global Ensemble
complicated physics to better repre- Forecast System (GEFS). In 2018, the Climate change means the risk of
sent the world. In a weather model, last time the system was upgraded, extreme weather will likely get worse.
this might mean more details about NWS increased its resolution from Forecasters hope that better predic-
processes in the ocean when consid- 34 kilometers to 25 kilometers and tions, communicated in the right
ering the frequency of hurricanes. the number of models in the ensem- way, can help people make better
ble from 21 to 31. decisions when the storms come. Q
More computing power might
also allow a model to get more

Report 17

EN ROUTE 

A food delivery service in
China is using Bluetooth to
track orders more accurately.

DANIEL ZENDER Fu, a driver in Beijing for Indoor localization systems based In early 2018, For now, the system simply acts
M R the food delivery service on Wi-Fi and radio-frequency iden- Alibaba placed as a check on the driver’s own logs,
tification do work, but Bluetooth is and it still requires drivers and mer-
Eleme, makes about a by far the cheapest, most reliable more than chants to have the app open on their
dozen deliveries per shift. But he option. Its accuracy is roughly 10 phones to guarantee a connection.
could make more—and spill less—if meters, good enough to detect peo- 12,000 If drivers try to report their arrival
he didn’t have to constantly get his ple walking into a shop or restaurant.  before their phone receives a signal
phone out to update his status. “I Bluetooth from a merchant’s phone, the app
have to log in every few minutes on In early 2018, Alibaba placed beacons in sends a “too early” prompt.
the app to avoid being penalized if more than 12,000 Bluetooth beacons shops across
the delivery is late because of the in shops across Shanghai. Beacons Shanghai. It’s not perfect—merchants
restaurant,” he says. emit signals that are picked up by can game the system by disabling
drivers’ phones in the form of “ID Yuan Ren is Bluetooth, so couriers’ apps don’t
In China, fierce competition and tuples.” The app uploads each tuple a science and log the time they spent waiting to
the promise of instant delivery drive to the platform’s servers, where it’s pick up an order.And virtual beacons
delivery apps to seek a technological matched with merchant IDs, and technology are less reliable than physical ones.
edge. Now Eleme, one of the main the system logs where and when journalist
players, has rolled out a vast indoor the signal was sent.  In theory, having more accurate
detection system to track drivers and based in data on drivers’ locations means the
ensure that customers receive their Similar networks are widely London. system can better assign upcoming
food on time.Wireless advances and used for tracking goods, people, jobs and ensure that drivers can com-
the explosion of connected devices— and services. One of the largest is plete their deliveries in time.Alibaba
including smartphones—have made in London’s Gatwick Airport, where says automation makes delivery driv-
this system possible. around 2,000 Bluetooth beacons are ers’work easier, but it may also inten-
installed. But Eleme’s is one of the sify pressures on the job.
Eleme, which has 83 million first to be built out on a city scale. 
monthly active users, is owned by Alibaba hopes drivers’ phones
the tech giant Alibaba, which also To take its system to more cities could one day become both beacons
owns Taobao, one of the world’s in China, Alibaba exploited the fact and receivers so their handsets could
biggest e-commerce platforms. that mobile phones can also act as locate each other without relying on
Since launching the new system in Bluetooth beacons.Apple introduced merchants’virtual beacons.The secu-
hundreds of Chinese cities starting this function for iOS devices in 2013, rity and privacy implications of oper-
in 2018, Eleme says, it has saved and similar features are now widely ating so many beacons are unclear.
merchants $8 million in refunds to available on other smartphones.
customers for problems with their Pannuto says the scale with which
deliveries, including lateness. Using this technology, more than Alibaba has expanded the system
3 million merchants and a million is impressive. He isn’t convinced
To build it, Eleme had to find drivers signed on to a pilot pro- it will be replicated elsewhere, but
a cost-effective system that works gram to use their phones as beacons in China, where delivery remains
indoors. GPS is accurate to five or receivers, delivering 3.9 billion cheap and demand high, companies
meters outside, but walls, furniture, orders to 186 million customers in are eager to find any way to outshine
and even people disrupt its signals. 364 Chinese cities.  the competition. Q
“It’s also really bad at elevation,”
says Pat Pannuto, a computer sci-
ence professor at the University of
California, San Diego. That’s a prob-
lem because most urban retailers in
China are in multistory buildings.

18 Computing

COMPUTING Tatyana His algorithms—which run on “We are making a map of the
Woodall is special processors called GPUs— full sky,” says Hsin-Fang Chiang, a
THE STARS a journalism combine advances in artificial intel- member of the Rubin’s data manage-
fellow at MIT ligence and distributed computing. ment team. And in the process, they
Astronomers are using AI, Technology Using either separate computers are building “a huge data set that’s
supercomputing, and the cloud or networks that act as a single going to be useful for many differ-
to organize a universe of data. Review. system, Huerta can identify gravi- ent kinds of science in astronomy.”
tationally dense places like black
space scientists collect The 10-year holes, which produce waves when Although Chiang’s PhD is in
A S more and more data, project will they merge. astronomy, her initial research
had nothing to do with the survey.
observatories around deliver a Huerta’s collection of AI models Years later, she got a chance to be
the world are finding new ways is open source, which means any- involved thanks to the sheer size
to apply supercomputing, cloud 500- one can use them. “Not everybody of the project. She’s proud that her
computing, and deep learning to PETABYTE has access to a supercomputer,” he work could improve the way scien-
make sense of it all. Here are some says. “This is going to lower the tists collaborate.
examples of how these technologies set of data and barriers for researchers to adopt
are changing the way astronomers images to the and to use AI.” The 10-year project will deliver a
study space. 500-petabyte set of data and images
cloud. How has the night sky changed? to the cloud, to help astronomers
What happens when As much as astronomy has answer questions about the struc-
black holes collide? expanded, the field has been slow ture and evolution of the universe.
As a postdoctoral student in the US, to integrate cloud computing. The
astrophysicist Eliu Huerta started to Vera C. Rubin Observatory, cur- “For each position in the sky,
think about how technology might rently under construction in Chile, we’ll have more than 800 images
help more breakthroughs happen in will become the first astronomi- there,” says Chiang. “You could even
his field. Then researchers detected cal institution of its size to adopt a see what happened in the past. So
gravitational waves for the first cloud-based data facility. especially for supernovas or things
time in 2015 with LIGO (the Laser that change a lot, then that’s very
Interferometer Gravitational-Wave When the observatory starts interesting.”
Observatory). up in 2024, the data its telescope
captures will become available as The Rubin Observatory will
Scientists have since charted part of the Legacy Survey of Space process and store 20 terabytes
these observations and scrambled and Time (LSST) project, which of data every night as it maps the
to learn all they can about these elu- will create a catalogue thousands Milky Way and places beyond.
sive forces. They’ve detected dozens of times larger than any previous Astronomers affiliated with the
more gravitational-wave signals, and survey of the night sky. Past surveys project will be able to access and
advances in computing are helping were almost always downloaded analyze that data from anywhere
them to keep up. and stored locally, which made it via a web browser. Eventually, the
hard for astronomers to access each images the telescope takes every
As a postdoc, Huerta searched other’s work. night will be converted into an
for gravitational waves by tediously online database of stars, galaxies,
trying to match data collected by and other celestial bodies.
detectors to a catalogue of poten-
tial waveforms. He wanted to find By comparing 4,000 simulations,
a better way. scientists could rewind time
and ask why some places in the universe
Earlier this year Huerta, who are rife with cosmic activity while others
is now a computational scientist
at Argonne National Laboratory are barren.
near Chicago, created an AI ensem-
ble that’s capable of processing a
month’s worth of LIGO data in just
seven minutes.

Report 19

IMPOSSIBLE
INSTRUMENTS

Researchers wanted to
build hyper-realistic digital
instruments. Musicians had

other plans.

What did the early universe Masato Shirasaki, an assis- The project hen Gadi Sassoon met
tant professor at the National requires a huge W Michele Ducceschi back-
look like? Astronomical Observatory of Japan, amount of data
Advances in computing could help says that question would be almost storage (about stage at a rock concert in
astronomers turn back the cosmic impossible to answer without these Milan in 2016, the idea of making
clock. Earlier this year, Japanese simulations. The project requires a 10 TERA- music with mile-long trumpets
astronomers used ATERUI II, a huge amount of data storage (about BYTES, blown by dragon fire, or guitars
supercomputer that specializes in 10 terabytes, equivalent to 22,000 strummed by needle-thin alien fin-
astronomy simulations, to recon- episodes of Game of Thrones). equivalent to gers, wasn’t yet on his mind. At the
struct what the universe may have time, Sassoon was simply blown
looked like as early as the Big Bang. Shirasaki’s team developed a 22,000 away by the everyday sounds of the
model of how the universe is thought classical instruments that Ducceschi
ATERUI II is helping the to have evolved and applied it to episodes and his colleagues were re-creating.
researchers investigate cosmic each of the simulations to see which of Game of
inflation—the theory that the result may be closest to how it looks Thrones) “When I first heard it, I couldn’t
early universe expanded expo- today. This method made it easier believe the realism. I could not
DANIEL ZENDER nentially from one moment to the to explore the physics of cosmic believe that these sounds were
next. Astronomers agree that this inflation. made by a computer,” says Sassoon,
expansion would have left extreme a musician and composer based in
variations in the density of matter In the next few years, Shirasaki’s Italy. “This was completely ground-
that would have affected both the methods could help shorten the breaking, next-level stuff.”
distribution of galaxies and the way observation time needed for future
they developed. efforts like SPHEREx, a two-year What Sassoon had heard were the
mission slated for 2024 involving a early results of a curious project at the
By comparing 4,000 simulations spacecraft that will orbit Earth and University of Edinburgh in Scotland,
of the early universe—all with differ- gaze at nearly 300 million galaxies where Ducceschi was a researcher at
ent density fluctuations—against the across the sky. With these leaps the time.The Next Generation Sound
real thing, scientists could rewind in computing, our understanding Synthesis, or NESS, team had pulled
time and ask why some places in the of the universe is expanding, bit together mathematicians, physicists,
universe are rife with cosmic activity by bit. Q and computer scientists to produce
while others are barren. the most lifelike digital music ever
created, by running hyper-realistic
simulations of trumpets, guitars, vio-
lins, and more on a supercomputer.

Sassoon, who works with both
orchestral and digital music, “try-
ing to smash the two together,” was
hooked. He became a resident com-
poser with NESS, traveling back and
forth between Milan and Edinburgh
for the next few years.

It was a steep learning curve. “I
would say the first year was spent
just learning. They were very patient

20 Computing

with me,” says Sassoon. But it paid Will Douglas One downside is that fewer people
off. At the end of 2020, Sassoon Heaven is a will learn to play physical instruments.
released Multiverse, an album cre- senior editor
ated using sounds he came up with for AI at MIT On the other hand, computers could
during many long nights hacking Technology start to sound more like real musicians—
away in the university lab.
Review. or something di erent altogether.
Computers have been making
music for as long as there have been the boundaries of what was possi- fretboard with needle-tip fingers, he
computers. “It predates graphics,” ble: trumpets that required multiple could make the strings vibrate with-
says Stefan Bilbao, lead researcher hands to play, drum kits with 300 out losing energy. “You get these
on the NESS project. “So it was interconnected parts. harmonics that just fizzle forever,”
really the first type of artistic activity he says.
to happen with a computer.” At first the NESS team was taken
aback, says Sassoon. They had spent The software developed by NESS
But to well-tuned ears like years making the most realistic vir- continues to improve. Their algo-
Sassoon’s, there has always been a tual instruments ever, and these rithms have sped up with the help
gulf between sounds generated by a musicians weren’t even using them of the university’s parallel com-
computer and those made by acoustic properly. The results often sounded puting center, which operates the
instruments in physical space. One terrible, says Bilbao. UK’s supercomputer Archer. And
way to bridge that gap is to re-create Ducceschi, Bilbao, and others have
the physics, simulating the vibrations Sassoon had as much fun as any- spun off a startup called Physical
produced by real materials. one, coding up a mile-long trumpet Audio, which sells plug-ins that can
into which he forced massive vol- run on laptops.
The NESS team didn’t sample umes of air heated to 1,000 °K—a.k.a.
any actual instruments. Instead they “dragon fire.” He used this instru- Sassoon thinks this new gener-
developed software that simulated ment on Multiverse, but Sassoon ation of digital sound will change
the precise physical properties of soon became more interested in the future of music. One downside
virtual instruments, tracking things more subtle impossibilities. is that fewer people will learn to play
like the changing air pressure in a physical instruments, he says. On the
trumpet as the air moves through By tweaking variables in the sim- other hand, computers could start to
tubes of different diameters and ulation, he was able to change the sound more like real musicians—or
lengths, the precise movement of physical rules governing energy loss, something different altogether. “And
plucked guitar strings, or the fric- creating conditions that don’t exist that’s empowering,” he says.“It opens
tion of a bow on a violin. They even in our universe. Playing a guitar in up new kinds of creativity.” Q
simulated the air pressure inside this alien world, barely touching the
the virtual room in which the virtual
instruments were played, down to DANIEL ZENDER
the square centimeter.

Tackling the problem this way
let them capture nuances that other
approaches miss. For example, they
could re-create the sound of brass
instruments played with their valves
held down only part of the way,which
is a technique jazz musicians use to
get a particular sound. “You get a
huge variety of weird stuff coming
out that would be pretty much impos-
sible to nail otherwise,” says Bilbao.

Sassoon was one of 10 musi-
cians who were invited to try out
what the NESS team was building.
It didn’t take long for them to start
tinkering with the code to stretch

Listen
to us

In Machines Business Curious Extortion
We Trust Lab Coincidence Economy

Technologies with the Unpacking the new Coming Soon: Coming Soon:
ability to learn have technologies coming Disseminating fact from An exploration of
unlocked a world of out of the lab and fiction while retracing common types of
possibilities that moving into the the origins of covid-19. cyber-crime, those
we’re only beginning marketplace. affected and those that
to understand. have secretly profited.

Download wherever you listen.

22 Computing

VIRTUAL WORLDS

IMMERSION Technology is starting to change that.
TRIP High-density screens,virtual-reality goggles
and glasses, surround sound, and spatial
The metaverse might sound like an internet catchphrase, but the audio are putting more genuinely immer-
concept has potential to change how we think about pain management, sive experiences within reach. Cameras are
gaining 3D capabilities, and single micro-
grieving, and systemic prejudice. phones are giving way to microphone arrays
that capture sound with better depth and
By John David N. Dionisio position. Augmented reality, which over-
lays virtual objects onto a video feed of
The first person to write about the “metaverse” was Neal John David N. the real world, provides a bridge between
Stephenson in his 1992 novel Snow Crash, but the concept Dionisio is purely virtual and analog or real experi-
of alternative electronic realms, including the “cyberspace” of a professor ences. There is progress toward adding a
William Gibson’s 1984 novel Neuromancer, was already well of computer sense of touch, too, in the form of multi-
established. touch screens, haptic technologies, control
science in the gloves, and other wearables. Wraparound
In contrast to what we typically think of as the internet, a Seaver College environments like Industrial Light and
metaverse is a 3D immersive environment shared by multi- of Science and Magic’s Stagecraft are within reach only
ple users, in which you can interact with others via avatars. A Engineering at to certain industries for now but may see
metaverse can, with the support of the right technology, feel like Loyola Marymount wider use as technology follows the typical
real life, with all the usual elements of work, play, trade, friend- curve of adoption and commoditization.
ship, love—a world of its own. University.
The tech giants weigh in
Perhaps the best-known prototype metaverse is the online
virtual world Second Life, whose very name implies an alternate The core ideas of a metaverse can be found
existence. Other games might also be said to be metaverses in most readily in games. But that’s likely
their own right: World of Warcraft, Everquest, Fortnite, Animal to change, as evidenced by the way cer-
Crossing. Each of these offers its own version of an immersive tain tech CEOs are now talking openly
world, although they don’t quite have the ability to completely about how a metaverse might work for
take over your senses. Most users experience these games from them. Facebook’s Mark Zuckerberg and
the outside looking in: screens front and center, with speakers on Microsoft’s Satya Nadella have already
the sides. Actions are mediated by a keyboard, mouse, trackpad, publicly mused about the possibilities.
or game controller instead of players’ hands and feet.
Zuckerberg uses the term “embodied
internet” for his version of the metaverse:
he imagines a system that is already much
like Facebook’s now-familiar communi-
ties, photos, videos, and merchandise,
but instead of looking at that content, in
Zuckerberg’s vision you’d feel as if you
were inside and surrounded by the con-
tent—an experience he presumably aims to
deliver with technologies from Facebook-
owned Oculus VR.

Nadella, meanwhile, has called
Microsoft’s Azure cloud services and other
offerings a “metaverse stack”—he’s used
the phrase “digital twin” in reference to
a system in which users can engage with
data, processes, and each other as richly in
virtual form as in reality, only with greater
speed and flexibility. Microsoft’s Surface
and HoloLens technologies would then play
the role that Oculus would for Facebook.

Report 23

JULIA DUFOSSÉ Let’s say these visions come to frui- furniture, Apple phones and computers)— people viscerally experience the results
tion. Would that be a good thing? Given but those functions are nice-to-haves, not of irreversible global warming.
all the misinformation and loss of privacy essentials. A truly immersive metaverse
produced by the old-fashioned internet, could go much further. Immersion might also help us under-
it’s easy to be skeptical about what mas- stand each other. The National Center for
sive technology companies might do with For one thing, there’s therapeutic Civil and Human Rights in Atlanta has an
a metaverse. Just like the internet and potential for the likes of PTSD, anxiety, exhibit where participants experience being
social media, the metaverse can and will and pain. Programs for burn victims at the target of racist taunts and threats. With
be misused. Deepfake technology can the University of Washington, pediat- audio alone, this is revelatory; if similar
already produce images that are indistin- ric patients at Children’s Hospital Los experiences were made available to more
guishable from photographs. People can Angeles, and women in labor at Cedars- people, in a manner that included visuals
be misled by much less. How much more Sinai indicate that virtual reality helps and haptics, metaverse technologies could
powerful might an immersive environ- mitigate pain in a very real way. These be used to advance the cause of diversity,
ment prove to be? initiatives involve synthetic environ- equity, and inclusion by helping people
ments where patients connect alone; a empathize with marginalized groups and
What’s it really good for? fully realized metaverse, with family and understand the effects of systemic prejudice.
caregivers also “dialing in,” could have
Those important caveats aside, though, additional benefits. The metaverse can and should become
there is reason to believe a metaverse newsworthy for reasons other than being
could actually empower us to do quite Immersive environments can also help some privileged executive’s dream. A
a bit. people experience things that would other- fully realized metaverse can stand not
wise be out of reach. Projects at Rensselaer only as a feat of technological innovation
Metaverses have already been Polytechnic Institute and Penn State, as and engineering but also, with the right
employed to let potential customers expe- an example, have sought to change atti- applications, as a vehicle for good in the
rience real estate and merchandise (Ikea tudes toward climate change by letting real world we all inhabit. Q

24 Computing

DIVERSITY

Fay Cobb Payton WHO GETS economy. Research shows that more
is a University TO SHAPE diverse teams are more innovative and
Faculty Scholar THE FUTURE OF generate more revenue.
COMPUTING?
professor We often hear the path to a technology
at North Tech s diversity and inclusion problems are career described as a pipeline. Most diver-
Carolina State preventing people and organizations from reaching sity efforts in our field have focused on get-
University. ting more people from diverse backgrounds
their full potential. into this pipeline. And yet representation
Lynette Yarger remains stubbornly low. Between 2014 and
is an associate By Fay Cobb Payton, Lynette Yarger, 2020, the proportion of Black and Hispanic
and Victor Mbarika tech professionals at Facebook increased
professor and by less than two percentage points.
assistant dean Last year, in response to Black Lives Matter, many US organi-
at Pennsylvania zations published diversity statements and made bold claims Why? The pipeline metaphor ignores
about fostering social change. As Black scholars in comput- the realities of racism, classism, and sexism
State ing, we saw these statements and pledges as reactionary and faced by those historically excluded from
University. largely ineffective. technology careers. Individuals who leak
out are often deemed deficient. This kind
Victor Mbarika Corporate America pledged $50 billion to address racial of thinking screams: Fix the people and
is Stallings justice but allocated only a fraction of a percent of those not the system.
Distinguished funds to direct grants, the best way to bring about systemic
Scholar at change. Meanwhile, at least 230 higher-education institu- Enter the pathway model, an alter-
East Carolina tions issued statements within two weeks of George Floyd s native to the pipeline metaphor. Pathway
University. murder. Many mentioned solidarity, equality, and greater advocates try to create multiple entry
inclusion, but only one in 10 included concrete action items points that can lead someone to a tech-
to address racial issues. nology career. The idea is that people will COURTESY PHOTOS
flow in from other fields, such as engi-
The track record of these institutions does not engender neering, the arts, mathematics, and even
confidence that they will follow through on whatever prom- the humanities. One way to promote this
ises they did make. There is little accountability, and no way flow is for two-year and four-year schools
to assess whether these commitments have actually improved to make it easy for people to start in one
the lives and livelihoods of Black people. program and finish in a different one.

Diversity and inclusion (especially of Black people) can Even when pathways provide more
improve product development, spur innovation, and spark entry points, getting through remains
creativity and entrepreneurship, all of which drive the nation s challenging, particularly for minorities in
America. One still has to be familiar with
the opportunities for academic success
and career readiness—and aware of the
barriers that can stand in the way. Those
vary between schools, and even between
departments within the same school. And
students must also be able to apply that
knowledge to navigate antiquated pro-
cesses and complex power structures.

The question is, what would be better?
We advocate for an ecosystem approach in
which many organizations work together
to address the lack of representation in
tech. The tech ecosystem should involve
K–12 schools, higher-education institu-
tions, companies, nonprofits, government
agencies, and venture capitalists. Public-
private partnerships could help design
environments that would be inclusive from

Report 25

the time people start their education to the Setting people, then what? If
day they finish their careers. an example you haven t changed the
environment, you ll lose
This might require us to rebuild systems A computer science educator on why she s them as quickly as you
like gateway mathematics courses (classes talking about identity in the classroom. get them.
such as pre-calculus that students must
pass in order to continue their program By Melba Newsome From college to
of study) and registration holds (which industry, every effort has
prevent a student from registering for NickiWashington could be mistaken been focused on the
classes until tuition and fees are fully paid). for a social scientist. Washington deficit of people with
These systems slow student progress and has long argued that computer sci- the most marginalized or
perpetuate disparate outcomes. entists like her should better under- minority identities. They
stand how their own identities affect are not the issue; the
Universities and technology companies their work. She joined Duke University in June 2020 issue is everyone who
could provide professional development and launched a groundbreaking course that analyzes marginalizes them.
opportunities for students from under- how race, gender, and class influence the way tech-
represented groups. But these organi- nologies get developed. Q: What can those in
zations would have to first change their We spoke about computing s challenges around power do?
own cultures to be more inclusive. That diversity, equity, and inclusion (DEI) and how she A: People in positions
means reimagining recruitment practices, found her way in the field. This interview has been of power have to rec
which typically rely on professional net- condensed and lightly edited for clarity. ognize that they re not
works and result in a homogeneous pool the most knowledgeable
COURTESY PHOTO of applicants, and addressing sources Q: The students in stronger dose of social about these things. So,
of algorithmic bias, such as automated Duke s computer sci sciences. Why? first and foremost, listen
r sum screeners that select candidates ence department are A: The problems in tech to Black women! Black
from particular schools and avoid those overwhelmingly white or nology don t begin with women have been tell
with ethnic-sounding names. Asian men. What made technology; they begin ing us on so many lev
you decide to discuss with the environment els what the issues are.
Organizations and fields of study that race, gender, and class where people are learn Second, allow yourself
adopt this approach will foster excel- in your course? ing and working. Some to be as uncomfortable
lence, innovation, and creativity. Georgia A: Being well aware of disciplines, like health as possible and sit with
State University is a good model. The the lack of diversity in care, teach cultural com that discomfort, which
university has eliminated achievement tech, I always worked petence because they means unlearning and
gaps by introducing meta-majors that to get more Black and work with clients and yielding space.
students select when they enroll. A biol- brown students into patients from a range of
ogy major who chooses a meta-major like computing early. After backgrounds and iden Q: How did having a
STEM takes classes together with stu- a while, I realized that it tities. Why are we not mom who was a com
dents who are pursuing careers in other no longer made sense, doing this in comput puter scientist affect
STEM fields, like medicine or math. Today, because they are not ing when the technol your career decisions?
African-American and Hispanic students going to stay if they re ogy we re developing A: When Mom graduated
at Georgia State graduate at the same rate experiencing racism. impacts the same peo college in 1973, she was
as white students. We have to change the ple in ways that are marginalized, ignored,
mindset of a workforce equally as harmful? and had to deal with
Ecosystems depend on both universities that is overwhelmingly problematic managers.
and companies to go beyond diversity state- white, Asian, and male Q: Why haven t diver She also had a small
ments. What we need is sustainable, inten- and get them to rec sity efforts been more group of friends who
tional change. Donating money to a cause ognize that new per successful? graduated from Black
can help, but it must be paired with policies spectives lead to more A: Because they re colleges and started at
that can make technology more equitable. innovation. focused on increasing IBM at the same time. So
numbers and represen when people talk about
Most important, we must hold today s Q: You say com tation. But once you how representation mat
leaders accountable by implementing poli- puter science needs a have more minoritized ters, it really did for me.
cies and procedures that emphasize trans- It was normal for me to
parency, compliance, and enforcement. see programmers, engi
The best way to fix systems that benefit neers, and managers
some and exclude others is to address the who looked like me.
underlying structures, not just the people. Q

26 Computing

SILICON AND ON AND ON ...

WHAT
EVER
HAPPENED

TO

DNA Spintronics Memristors
COMPUTING?
Computer chips are built Classic electronics is based on
Silicon has bested every around strategies to control three components: capacitor,
potential replacement. the flow of electrons—more resistor, and inductor. In 1971,
But let’s not count them all out just yet. specifically, their charge. In the electrical engineer Leon
addition to charge, however, Chua theorized a fourth com-
By Lakshmi Chandrasekaran electrons also have angular ponent he called the memris-
momentum, or spin, which tor, for “memory resistor.” In
W hen the first transistor was created, in can be manipulated with 2008, researchers at Hewlett-
1947, few could have imagined the even- magnetic fields. Spintronics Packard developed the first
tual impact of this device—the switch that emerged in the 1980s, with practical memristor, using
lies at the heart of logic chips. the idea that spin can be used titanium dioxide.
We have silicon to thank for comput- to represent bits: one direc-
ing’s great takeover. Add a minute pinch of impurities to the tion could represent 1 and It was exciting because
element, and silicon forms a material almost ideal for tran- the other 0. memristors can in theory be
sistors in computer chips. used for both memory and
In theory, spintronic tran- logic. The devices “remem-
For more than five decades, engineers have shrunk silicon- sistors can be made small, ber” the last applied voltage,
based transistors over and over again, creating progressively allowing for densely packed so they hold onto information
smaller, faster, and more energy-efficient computers in the chips. But in practice it has even if powered down. They
process. But the long technological winning streak—and the been tough to find the right also differ from ordinary resis-
miniaturization that has enabled it —can’t last forever. “There substances to construct tors in that their resistance
is a need for technology to beat silicon, because we are reach- them. Researchers say that a can change depending on the
ing tremendous limitations on it,” says Nicholas Malaya, a lot of basic materials science amount of voltage applied.
computational scientist at AMD in California. still needs to be worked out. Such modulation can be used
to perform logic operations.
What could this successor technology be? There has been Nevertheless, spintronic If done within a computer’s
no shortage of alternative computing approaches proposed technologies have been com- memory, those operations can
over the last 50 years. Here are five of the more memorable mercialized in a few very cut down on how much data
ones. All had plenty of hype, only to be trounced by silicon. specific areas, says Gregory needs to be shuttled between
But perhaps there is hope for them yet. Fuchs, an applied physicist at memory and processor.
Cornell University in Ithaca,
New York. So far, the biggest Memristors made their
success for spintronics has commercial debut as nonvol-
been nonvolatile memory, the atile storage, called RRAM
sort that prevents data loss or ReRAM, for “resistive ran-
in the case of power failure. dom access memory.” But the
STT-RAM (for “spin transfer field is still moving forward.
torque random access mem- In 2019, researchers devel-
ory”) has been in production oped a 5,832-memristor chip
since 2012 and can be found that can be used for artificial
in cloud storage facilities. intelligence.

Report 27

Carbon nanotubes DNA computing Molecular electronics What comes next?

Carbon isn’t an ideal semi- In 1994, Leonard Adleman, It’s a compelling vision: tran- Silicon still reigns
conductor. But under the a computer scientist at the sistors keep getting smaller supreme, but time is run-
right conditions it can be University of Southern and smaller, so why not jump ning out for everyone’s
made to form nanotubes that California in Los Angeles, ahead and make them out favorite semiconductor.
are excellent ones. Carbon made a computer out of a of individual molecules? The latest International
nanotubes were first crafted soup of DNA. He showed Nanometer-scale switches Roadmap for Devices
into transistors in the early that DNA could self-assem- would make for a supremely and Systems (IRDS) sug-
2000s, and studies showed ble in a test tube to explore cost-effective, densely gests that transistors are
they could be 10 times more all possible paths in the packed chip. The chips might expected to stop shrinking
energy efficient than silicon. famous “traveling salesman” even be able to assemble after 2028 and that inte-
problem. Experts predicted themselves thanks to interac- grated circuits will need to
In fact, of the five alter- DNA computing would beat tions between molecules. be stacked in three dimen-
native transistors discussed silicon-based technology, par- sions to keep making faster
here, carbon nanotubes may ticularly with massively paral- Groups at Hewlett- and more efficient chips
be the farthest along. In 2013, lel computing. Decades later, Packard and elsewhere in the possible.
Stanford researchers built the researchers concluded that early 2000s raced to make
world’s first functional com- DNA computing isn’t fast the chemistry and electronics This might be the time
puter powered entirely by enough to do that. work together. when other computing
carbon nanotube transistors, devices find an opening,
albeit a simple one. But DNA holds some But after decades of work, but only in conjunction
advantages. Researchers the dream of molecular with silicon technology.
But carbon nanotubes have shown that it’s possible electronics is still just that. Researchers are explor-
tend to roll into little balls to encode poetry, GIFs, and Researchers have found that ing hybrid approaches
and clump together like spa- digital movies into the mole- single molecules can be fin- to making chips. In 2017,
ghetti. What’s more, most cules. The potential density is icky, working as transistors researchers who had made
conventional synthesis meth- staggering. All of the world’s under only very narrow con- progress with carbon nano-
ods make semiconducting digital data could be stored ditions. “No one has shown tube transistors integrated
and metallic nanotubes in a in a coffee mug full of DNA, how single-molecule devices them with layers of nonvol-
messy mix. Material scien- biological engineers at MIT can be reliably integrated atile memristors and sili-
tists and engineers have been estimated in a paper earlier into massively parallel micro- con devices—a prototype
researching ways to cor- this year. The catch is cost: electronics,” says Richard for an approach to improv-
rect and work around these one coauthor later said that McCreery, a chemist at the ing speed and energy con-
imperfections. In 2019, MIT DNA synthesis would need University of Alberta. sumption in computing by
researchers used improved to be six orders of magni- moving away from tradi-
techniques to make a 16-bit tude cheaper to compete with The dream of molecu- tional architecture.
microprocessor with more magnetic tape. lar electronics has not com-
than 14,000 carbon nanotube pletely died, but these days Classic silicon-based
transistors. That’s still far Unless researchers can it is largely relegated to the chips will still make some
from a silicon chip with mil- cut the cost of DNA storage, chemistry and physics labs, progress, says AMD’s
lions or billions of transistors, the stuff of life will stay stuck where researchers continue Malaya. But, he adds, “I
but it’s progress nonetheless. in cells. struggling to make endlessly think the future will be
fickle molecules behave. heterogeneous, in which
all the technologies are
used probably in a comple-
mentary way to traditional
computing.”

In other words, the
future will still be silicon.
But it will be other things
as well.

Lakshmi Chandrasekaran
is a freelance science
writer based in Chicago.

28 Computing

Brain machine interfaces 29

A
computer
mouse
PARALYZED PEOPLE ARE USING BRAIN
IMPLANTS TO TYPE AND MOVE ROBOTIC
ARMS. IS THIS THE NEXT GREAT
COMPUTER INTERFACE FOR ALL OF US?

inside In a 12-by-20-foot room at a skilled-nursing facility
in Menlo Park, California, researchers are testing
the next evolution of the computer interface inside
the soft matter of Dennis Degray s motor cortex.
Degray is paralyzed from the neck down. He was
hurt in a freak fall in his yard while taking out the
trash and is, he says, as laid up as a person can
be. He steers his wheelchair by puffing into a tube.

But Degray is a virtuoso at using his brain to
control a computer mouse. For the last five years,
he has been a participant in BrainGate, a series of
clinical trials in which surgeons have inserted sili-
con probes the size of a baby aspirin into the brains
of more than 20 paralyzed people. Using these
brain-computer interfaces, researchers can measure
the firing of dozens of neurons as people think of
moving their arms and hands. And by sending these
signals to a computer, the scientists have enabled
those with the implants to grasp objects with robot
arms and steer planes around in flight simulators.

Degray is the world s fastest brain typist. He
first established the mark four years ago, using
his brain signals to roam over a virtual keyboard
with a point-and-click cursor. Selecting letters on
a screen, he reached a rate of eight correct words

BY ILLUSTRATIONS BY

ANTONIO SELMAN
REGALADO DESIGN

your head

30 Computing

in a minute. Then, right before the covid-19 pan- about everything from their impact that is therapeutic and restorative, I am PBS NEWSHOUR STILL / NEWSHOUR PRODUCTIONS LLC (OPPOSITE)
demic began, he demolished his own record, using on inequality (what if only some peo- into.Anything that is elective, enhance-
a new technique where he imagined he was hand- ple can afford one?) to the conse- ment—I don’t want to work on that.
writing letters on lined paper. With that approach, quences of directly linking people’s But when the technology is so early,
he managed 18 words per minute. brains to social media. But he has you can’t pursue the restorative stuff
made a Faustian bargain in working without being generally aligned with the
One of the people responsible for the studies with Neuralink, which is bringing people who want to take it beyond.We
with Degray is Krishna Shenoy, a Stanford University much-needed resources to commer- are on the early part of the same path.”
neuroscientist and electrical engineer who is among cializing an interface that—at first,
the leaders of the BrainGate project. While other at least—promises huge benefits for Monkey Pong
brain-interface researchers grabbed the limelight paralyzed people.
with more spectacular demonstrations, Shenoy’s Neuralink is a secretive company
group has stayed focused on creating a practical “It’s not comfortable, but welcome that communicates with the public
interface that paralyzed patients can use for every- to science,” says Shenoy. “Anything mostly via theatrical presentations.
day computer interactions. “We had to persevere
in the early days, when people said Ah, it’s cooler
to do a robotic arm—it makes a better movie,” says
Shenoy. But “if you can click, then you can use
Gmail, surf the Web, and play music.”

Shenoy says he is developing the technology for
people with “the worst afflictions and the most need.”
Those include patients who are utterly locked in and
unable to speak, like those in the end stage of ALS.

But if the technology allows people like Degray
to link their brain directly to a computer, why not
extend it to others? In 2016, Elon Musk started a
company called Neuralink that began developing a
neural “sewing machine” to implant a new type of
threaded electrode. Musk said his goal was to establish
a high-throughput connection to human brains so that
society could keep pace with artificial intelligence.

The same month Neuralink went public with
its plans, Facebook announced it would develop
a “noninvasive” brain-reading helmet to translate
thoughts into social media posts. What’s followed
has been a huge influx of investment in brain inter-
faces of all kinds, including EEG readers, magnetic
headbands, and new types of high-density implanted
probes capable of measuring signals from tens of
thousands of neurons at a time.

More than $300 million has been raised by
such companies in the last 12 months, even though
Facebook this year dropped its quest (it determined
a brain-reading helmet won’t be a feasible way to
send texts for years). “The field was un-investable
until Elon entered. That is what sent shock waves
through the venture capital world,” says Shenoy.
“Now there are nearly infinite resources.”

The money comes with a catch, though. Medical
researchers like Shenoy want to help desperate
cases. But entrepreneurs want the next interface
for everybody. Musk has said he is aiming for brain
implants that would be available to any consumer
who wants one—Neuralink even designed a sleek
white surgical chair where he imagines people will
sit for a routine 30-minute implant procedure.

Shenoy, who is a paid consultant to Neuralink,
told me he’s living a scientific paradox. He is
opposed to consumer brain implants; he worries

Brain-machine interfaces 31

The latest, released in April 2021, showed a rhe- Shenoy says he is developing the
sus monkey named Pager playing the video game technology to restore a digital existence
Pong with its mind. The demo led to an excited to people with “the worst afflictions and
response on social media—as well as a lawsuit by the most need.”
animal rights activists—but mind Pong was not
new. A BrainGate subject named Matt Nagle had <--- Dennis Degray uses brain signals to type on a virtual keyboard. He manages 18 words per directly from your brain. Or you can from around 100 electrodes at once.
played the game against a Wired editor in 2005. minute, about half as fast as the average able-bodied person texting from a smartphone. even imagine connecting to someone In general, brain implants use each
else’s mind, seeing and hearing what electrode to listen to one neuron.
The real advance made by the other person is doing. Neuralink’s N1 implant measures from
Neuralink was something not visible 1,024 electrodes that lie along thin
in the video—the implant itself. Chip Musk says all this is part of a metal threads; that means it’s listening
designers at the company have built strategy to offset existential risks he to about a thousand neurons. And it
a soda-cap-size disc, containing pro- thinks future artificial intelligence will has only been tested in monkeys and
cessors and a wireless radio, that con- pose to humankind—like a scenario pigs so far.
nects to electrodes stitched into the in which an AI decides to wipe out
monkey’s cortex. The disc lies flush humanity, Terminator style. His view When it comes to consumer
with the monkey’s skull and is covered is that to prevent such an outcome, implants installed via elective brain
with skin—giving the implant a more humans should become cyborgs and surgery, regulators, public opinion,
practical footprint than the cables that merge with AI. “If you can’t beat em, and even the medical profession
protrude from Degray’s head. join em,” Musk typed into Twitter in may also stand in the way. In 2016, a
July 2020, describing the phrase as Pew Research poll found that 69% of
In a blog post, Neuralink said that the “Neuralink mission statement.” Americans were either very or some-
Pong was just a demonstration—and what worried about the prospect of
also articulated for the first time what Neuralink says its eventual goal brain chips that offered an improved
its implant would be used for, at least is “creating a whole brain interface ability to concentrate or process infor-
in the near term. It read, “Our first capable of more closely connecting mation. According to Pew, this oppo-
goal is to give people with paralysis biological and artificial intelligence.” sition was strongly related to a fear of
their digital freedom back: to commu- Technologically, achieving that goal “loss of human control.”
nicate more easily via text, to follow means developing a high-bandwidth
their curiosity on the web, to express brain-computer connection that can And brain surgeons will still need
their creativity through photogra- tap into thousands or millions of neu- some convincing before they drill
phy and art, and, yes, to play video rons at once. holes in the heads of healthy people.
games.” A Neuralink engineer later Jaimie Henderson, the Stanford neuro-
told IEEE Spectrum that the com- The technology isn’t there yet. The surgeon who put in Degray’s implants
pany had the specific aim of beating system used on Degray measures and co-leads the project with Shenoy,
Degray’s brain-communication record. says he thinks small implants done
with minimal trauma are “fairly low
But Musk’s long-term plans risk,” with the main hazard being a 3%
are equally clear: he thinks human to 5% chance of infection—a risk that
brains need to be directly connected may be worth it to improve a severely
to phones, computers, and applica- disabled person’s life. The question
tions. You could run Google searches will be whether healthy people gain
enough from an implanted computer
mouse to offset the dangers, even if
they’re small.

“It’s unclear to me what benefits
able-bodied people would be able to
get from any current brain-computer
interface system,” Henderson says.
“Our goal has been to try to restore
function for people who have lost it,
as best we can—not to provide some
sort of ‘superhuman’ capability.”

5 milestonesinthehistory Still, Shenoy was one of several implants built by a company called
of brain-machine academic scientists who told me that, Blackrock Neurotech, based in Salt
interfaces like it or not, they do think consumer Lake City. The implant, aptly called the
brain implants are going to be possi- Utah array, is a silicon square with 100
2005—Matt Nagle, who is ble. Enough subjects like Degray have small needles, which is pushed into the
paralyzed, uses a brain- lived with implants for years, with surface of the brain. Blackrock mostly
computer interface to play few ill effects, and they’re achieving sells systems to researchers experi-
Pong against a Wired editor. useful mastery of the brain mouse. menting on animals, but as investors
“Technologically, I see no barrier. I have flocked to implants, observers
2005 would not have said that 10 years ago have sometimes called Blackrock and
and might not have said it five years Neuralink the Lyft and Uber of brain
2010 2008—Early ago,” Shenoy says. “It’s basically elec- interfaces.
brain-interface trodes, chips, and a radio.”
company Cyber- The president of Blackrock, an
kinetics ceases To some, such an interface is electrical engineer named Florian
operations. intriguing because of the sheer amount Solzbacher, thinks the timing is right
Academic exper- of time we now spend on phones, play- to take implants forward for paralyzed
iments continue ing video games, listening to podcasts, people. “People would say Oh my God,
under the name or scrolling through social media. it’s brain surgery, but actuallywe haven’t
“BrainGate.” That’s propelling investments in new seen any problems,” he says. Every time
ways of interfacing with the brain, there is a video of someone controlling a
2016—Elon Musk says Nita Farahany, a law professor robot or eating a Twinkie with a robotic
founds Neuralink at Duke University who is writing a hand, Solzbacher says, he gets calls from
with the aim of book on consumer neurotechnology. paralyzed people wondering when a
developing a commercial product could be available.
high-bandwidth “The question of why seemingly It’s only recently that he’s been willing
brain-machine in- disparate companies are investing is to say it could happen soon: “It’s always
terface connecting that if you could use your brain as the been 15 years away, and now what I can
humans to artificial controller, instead of a mouse or joy- say for the first time is soon you will be
intelligence.  stick, it’s not so crazy to want to invest,” able to take it home.”
says Farahany. “This may be the next
2015 revolution in the computer interface.” That’s due to several factors, includ-
ing the development of a wireless ver-
2017—Facebook says it has Nathan Copeland is another par- sion of the BrainGate hardware. Instead WIKIMEDIA COMMONS; FACEBOOK; STANFORD HEALTH CARE/SCHOOL OF MEDICINE (KEYBOARD)
assigned 60 engineers to develop- alyzed person living with a brain of cables, subjects have a hockey-puck-
ing a thought-reading helmet. implant—he’s part of a study in size wireless transmitter screwed onto
It abandons the project four years Pittsburgh. Last year he became the their brain ports. It’s nothing as com-
later as technically infeasible.  first to plug his head into a tablet com- pact and sleek as Neuralink’s elec-
puter at home, on his own time, not as tronics, but it works. Solzbacher says
2020 part of scientific session (it normally his company plans to seek approval to
takes a small team of medical workers sell its own improved wireless system
2021—Implant recipient Dennis in a clinical setting). Copeland told me to people with ALS or severe paralysis.
Degray’s brain-typing record of 18 at first he was using the device eight
words a minute is reported in Nature. hours a day, playing video games and Solzbacher says Degray’s typing
using drawing programs. He later tired points to the potential of the technol-
of it—his tablet is a medical device that ogy—he can tap out words much faster
uses an older version of Windows, and than anyone using an EEG headband,
its battery doesn’t last long. for example. “That means you are 10
times faster than anything that is out
Still, Copeland told me he believes there,” he says.“Nowyou can start being
paralyzed people are “test pilots” for productive, and you have performance
future consumer brain interfaces. In close to an able-bodied person.”
his own case, he says, he’s mostly
interested in being able to play more However, Solzbacher is being
video games—one of his favorite pas- financed by people who aren’t only
times—at a higher level. interested in helping paralyzed peo-
ple. This year his company raised
Game changer $10 million from investors includ-
ing the German billionaire Christian
Of the 35 or so people who have Angermayer, who invests widely in
received a long-term brain implant psychedelics, longevity treatments, and
to interface with a computer, 29 of mental health. In a tweet, Angermayer
them, including Degray, have electrode

Brain-machine interfaces 33

left no doubt he thinks a general-purpose brain The “mixed reality” experiments done
mouse is the ultimate goal: “It’s fundamentally an in virtual space hint at how able-bodied
input-output device for the brain, and it can benefit people might experience computer worlds
ALL. We can unlock truly astonishing use cases & through brain interfaces.
I believe Blackrock will be the one to take us there.
Ppl will communicate with each other, get work done recent online presentation. “Some of these devices are a “sneak peek” at
+ even create artwork, directly with their minds.” the applications we are working on how consumers might use future non-
have blown my mind.” invasive systems like brain-reading
Solzbacher says for now, none of helmets or headbands, should accu-
Blackrock’s internal plans or projec- The researchers at APL include rate ones be developed.
tions involve consumer brain implants. Michael Wolmetz, manager of the
Still, he acknowledges that could hap- Human and Machine Intelligence When I asked Wolmetz what he
pen: “I expect there is part of society program.Wolmetz says the demonstra- thought people might use such inter-
that may really want it, even though tions are a glimpse of “fundamental” faces for in the future, he said it is
there is nothing wrong with them.” changes ahead in human-computer hard to predict. “It’s like asking what
interaction, especially the concept of the computer is going to be for,” he
I asked Solzbacher whether any “mixed reality.”The experiments in vir- says. “I think that in our lifetimes it
able-bodied person had ever requested tual space hint at how able-bodied peo- will be for everything. But in the next
such a device. He says he’s hasn’t got- ple might experience computer worlds five years, it’s hard to answer.”
ten such a request, yet. through brain interfaces—making the
APL project one of the most explicit Some want not only the computer
Mixed reality explorations of how such technology mouse but the entire interface—includ-
might lead to human enhancement. ing the screen, or whatever replaces a
Robert “Buz” Chmielewski had his screen—in the brain. One of them is
head down in concentration, and “For all of biological history, the Max Hodak, the former president of
because of a screen, he couldn’t see only way we’ve interacted with the Neuralink. He was fired by Musk in
which of two toy-size soccer balls had environment is with senses and motor March—it’s not clearwhy—but quickly
been placed in the robotic hand he function,”Wolmetz says. “We have, for formed a new company, called Science
was controlling. Using his thoughts, the first time, the ability to go outside Corp., with financial backing from the
Chmielewski closed the plastic and that paradigm. It’s the first time any cryptocurrency billionaire Jed McCaleb.
metal hand and squeezed the ball. biological organism has done that.” Hodak says he plans to develop a new
“Pink ball,” he called back. When the type of implant that rests on the retina
researcher swapped it for another, Wolmetz doesn’t know whether and can send information to the visual
stiffer ball, Chmielewski could sense surgically implanted brain interfaces cortex at the back of the brain.
the change. “Black ball,” he said. will ever be widely used, but he says

Chmielewski, 50, got his Utah Key Blackrock BrainGate DARPA
arrays implanted in 2019, 30 years players Neurotech An academic con- The US defense
after a surfing accident in Ocean City, Markets the “Utah sortium has put agency has spent
Maryland, left him in a wheelchair. array,” the implant implants in the $120 million on
During the two years the experiment most often used brains of over 20 brain interfaces in
lasted (it ended in September), he had in brain-computer people. the last five years.
more implants put in than any other experiments.
patient—a total of six, in both hemi-
spheres of his brain. Because of this, Kernel Neuralink Paradromics Synchron
he was able to control two robot arms Has developed a Elon Musk’s The startup The Australian com-
simultaneously. What’s more, three type of wearable company raised is developing pany started a trial
of the probes placed into his sensori- brain scanner a further $200 high-density brain of a brain sensor
motor cortex sent signals back into his that uses infrared million in 2021. electrodes. inserted through
brain, allowing him to receive tactile beams. a blood vessel. It
information from the robots. allows simple com-
puter control.
Chmielewski was part of a project
at Johns Hopkins University’s Applied
Physics Laboratory that’s testing new
forms of perception. He also tried
out the Microsoft HoloLens headset
and used his sense of virtual touch
to arrange blocks in virtual space. “If
you would have told me three years
ago I would be controlling things with
my thoughts, I would have said you’re
crazy,” Chmielewski said during a

34 Computing

Shenoy says his concern is that putting dangers. In 2017, the same year that both Neuralink
computer interfaces into people’s minds and Facebook’s brain-interface plans were unveiled,
will lead to inequality and the same sorts of a group of researchers calling itself the Morningside
information abuses seen on the internet. Group published a manifesto in the journal Nature. It
sounded alarm bells about a “convergence” between
Initially Hodak’s new company engineer approached him and said brain technology and AI advances. <--- Robert “Buz” Chmielewski, 50, had implants in both hemispheres of his brain.
will be looking to help people, like he was an avid gamer. The engineer When they were in place, he was able to control two robot arms simultaneously.
his grandfather, who went blind from wanted to know if it would be possi- The group formed at the urging of Rafael Yuste, a
retinal diseases. But a medical prod- ble to have a third thumb. neuroscientist at Columbia University,
uct is a stalking horse for a bigger who became alarmed over experi-
ambition, which is to create a device “That’s taking things to an extreme. ments in his own lab, in which he
that can produce images in the eyes I don’t want to implant electrodes into could not only read from the visual
of healthy people as well. people so they can be better gamers,” center of a mouse brain but also use
says Donoghue. “I always challenge a laser to make the animal perceive
“It could just be a computer screen all of these ideas because I don’t see things that were not there. “We had
that looks as solid as any ever has, and what it gets you. But I don’t dismiss control over the visual perceptions of
it’s just floating in front of you,” he says. it, either … that is what is driving peo- the mice, and we could run them like
“When your eyes are open, you would ple. It’s the cool factor, that you could puppets,” says Yuste.
see the world of atoms.When you close have this new interface.” Yuste keeps a list of experiments he
your eyes, you see the world of bits.” thinks point to how neurotechnology
Hodak thinks that in a generation, Donoghue doubts that implants could compromise human autonomy.
children will be “baffled when we tell will provide superpowers, or that For instance, there’s the work of Jack
them that there used to be just noth- you’ll be able to download French Gallant, in California, who has used
ing there when we closed our eyes.” for Beginners directly into your head MRI scanners to deduce what images
anytime soon. The brain has evolved people are seeing. Then there’s the
Ethics questions to receive and send information at the scientist who wired one monkey’s
speed that it does—not at the rate of brain to control the arm of different
Before Musk and venture capital- an Ethernet cable. “Have you listened monkey, calling one the “master” and
ists arrived on the scene, DARPA, an to a podcast at 4x speed? It doesn’t the other its “avatar.”
R&D agency of the US Department work very well,” he says. “Our brains The fundamental fear is that
of Defense, was the world’s largest are made to make and intake speech everything bad about the internet—
funder of brain-interface research. at a level that we can use it.” disinformation, malicious hackers,
government control, corporate manip-
Andy Schwartz, a researcher at the But others believe that mind ulation, endless harassment—could
University of Pittsburgh, told me he reading and mind control are rising get much worse if technology were to
is convinced the military’s fascination breach what the Morningside Group
with the technology springs from a calls “the last frontier of privacy” and
1982 Clint Eastwood film, Firefox, know our thoughts. “There is a huge
whose plot involves an effort to steal problem, and it’s the problem of men-
a thought-controlled Soviet MiG tal privacy,” says Yuste.
jet. After the military had one of his
research subjects fly a simulated war
jet, Schwartz says, he stopped collab-
orating with the agency.

John Donoghue, a professor at
Brown University and one of the
founding scientists of BrainGate, is
also concerned about a “circus-like
atmosphere” around brain implants.
He spent time in a wheelchair as a
child, which is one reason he has pur-
sued the goal of restoring movement
to paralyzed people. But when he gave
a talk at Google a few years ago, an

Brain-machine interfaces 35

JOHNS HOPKINS APL (OPPOSITE) In May, Yuste hosted a day-long a noninvasive headset to read brain “I have to change me”
online gathering of ethicists and activity, said during the event.
neurotech entrepreneurs to discuss In certain ways, the field of brain-computer inter-
responsible neural-interface design. Yuste wants far stricter privacy faces is already beginning to realize its loftiest goal
rules than those governing internet and some people’s greatest fear: the merger of
Several participants said they data or what’s on your iPhone. He humans and AI.
believed there was a need to establish would like to see brain data treated like
rules before it becomes possible to col- transplant organs—carefully tracked That’s certainly the case with research volunteers
lect brain information easily.“We don’t and with a ban on any profit-making. like Degray.The buzzing of his neurons is interpreted
want to go through this cycle of big cor- At a minimum, he says, brain data by AI software called a recurrent neural network.
porations harvesting data to profit from should be protected like medical Each day that Degray uses his implant, he starts by
and then, in the end, facing regulations information. He also says the military imagining some simple movements, like drawing
and asking for forgiveness,” Ryan Field, should be forbidden from employing a circle. The neural network listening to his neu-
CTO of Kernel, which is developing brain implants. rons then calibrates the statistical map that relates
each neuron’s activity to the movement. And most
brain-computer interfaces won’t only use software
to interpret brain signals, but also to improve on
them—for example, programs might predict what
word someone is trying to spell on the basis of the
first few letters.

This results in what Blackrock’s Solzbacher calls
“shared agency,” or outputs that are picked partly by
a person and partly by a machine. “That is scientif-
ically interesting but is also an ethics question,” he
says. “Because who is actually making the decisions
when the systems adapt?”

Currently, the closest thing there is to brain-
interface experience design is the experiments being
carried out with Degray in California. Most recently,
the team has been trying to get Degray to try mental
touch-typing. If software can track what movements
he’s thinking of making with his fingers, that could
increase his communication speed even more. The
problem is that before his accident, Degray was never
more than a hunt-and-peck typist. He now has paper
keyboards pasted on the ceiling above his bed so he
can practice thinking about typing.

One thing I wanted to know from Degray is what
it feels like to operate a computer with his brain. He
described what he calls a “meeting of the minds”
with the cartful of machines and software reading
his thoughts. This was particularly true when he was
performing the imagined-handwriting task.

“It’s a very personal interaction. You have to feel
for where the movements are in your own body,” he
says. “You are trying to write the letters, and it is try-
ing to understand you. I wouldn’t call it a relationship,
but it’s close. It’s almost a conversation between the
device and myself. Some days it’s a little bit surly at
first—it’s hard to wake it up. Of course, the machine
is perfectly constant. So I have to change me to get
it to work.”

One day Degray imagined writing 5,000 words.
He worked so hard at it the researchers had to remind
him to breathe. “I just pounded it out,” he says. “Over
the course of doing so many words, you can get
consistent. You quickly lapse into a pattern that the
computer can recognize.”

Antonio Regalado is the senior editor for
biomedicine at MIT Technology Review.

GUTTER CREDIT HERE

Q&A 37

During her time at Microsoft and in academia, Jennifer Chayes showing that the system classified white faces more accu-
has been fighting to use data science and computing to make rately than it did brown and Black faces. This finding caused
artificial intelligence more fair and less biased. the company to turn down a lucrative contract with a police
department and start working to remove the bias from such
From dropping out of school at the age of 15 to becoming the algorithms. The FATE (Fairness, Accountability, Transparency
doyen of data science at the University of California, Berkeley, and Ethics in AI) group was created at the lab.
Chayes has had quite the career path. She joined UCLA in 1987
as a tenured professor of mathematics. Ten years later, Microsoft Anil Ananthaswamy asked Chayes, now associate provost
lured her to cofound its interdisciplinary Research Theory Group. of the Division of Computing, Data Science, and Society and
dean of the School of Information at Berkeley, how data sci-
It was in her Microsoft lab in New York City that researchers ence is transforming computing and other fields.
discovered bias in the company’s facial recognition software,

Jennifer Chayes

CHRISTIE HEMM KLOK Q: What was it like to transition from aca- Q: Can that cause problems? automatically. We are not allowed to
demia to industry? A: Some consider it a problem in cases in use gender or race. Even if I don’t look
at [these] protected attributes, there are
A: That was quite a shock.The VP of research which you have, [for example], biomed- many things [in the data] that are proxies
at Microsoft, Dan Ling, called me to try ical data. The data very accurately pre- for gender or race. If you have gone to
to convince me to go for an interview. dicts what’s going to work and what’s certain schools, if you grew up in certain
I talked to him for about 40 minutes. not going to work, without an underlying neighborhoods, if you played certain
And I finally said, “Do you really want to biological mechanism. sports and you had certain activities,
know what’s bothering me? Microsoft is they are correlated [with gender or race].
a bunch of adolescent boys, and I don’t Q: Any advantages?
want to spend my life with a bunch of A: What the data has allowed us to do now, Q: Do algorithms pick up on these proxies?
adolescent boys.” A: They exacerbate it. You must explicitly
in many cases, is to run what an econo-
Q: How did he react to that? mist would call a counterfactual, where understand this, and you must explicitly
A: He said, “Oh, no, we are not. Come and you actually see random variation in the prevent it in writing the algorithm.
data that allows you to draw conclusions
meet us.” I met some incredible women without doing the experiments. That’s Q: How can we address such issues?
there when I visited, and I met phenom- incredibly useful. A: There is this whole area of FATE: fair-
enally open-minded people who wanted
to try things to change the world. Do I really want to try out different ness, accountability, transparency, and
educations on different populations? Or ethics in AI, which is the design of these
Q: How has data science changed do I want to see [that] there was random algorithms and understanding what they
computing? variation at some point that will allow me are. But there is so much more that we
to draw a really good causal inference, and need to do.
A: As we’ve gotten more data, computer sci- therefore I can base policy on it?
ence has begun to look outward. I think of Q: And data science helps?
data science as a marriage of computing, Q: Do you see a problem in how data A: This is absolutely data science. There is
statistics, ethics, and a domain emphasis
or a disciplinary emphasis, be it biomedi- is being used, especially by big part of the web called the “manosphere,”
cine and health, climate and sustainability, where a lot of hate is originating. It’s kind
or human welfare and social justice, and companies? of hard to trace. But if you use natural-
so on. It is transforming computing. A: There are myriad problems. It’s not only language processing and other tools,
you can see where it’s coming from. You
Q: Is there a difference in how data scien- being used by tech corporations. It’s being can also try to build interfaces that allow
tists solve problems? used by insurance companies. It’s being advocacy groups and others to find this
used by government platforms, public and to help root it out. This goes beyond
A: With the advent of all of this data,we have health platforms, and educational plat- just being fair. This is turning the tables
the opportunity to learn from the data forms. If you do not explicitly understand on the way in which these platforms have
without having a theory of why some- what biases can be creeping in, both in been usurped to increase bias and hate
thing is happening. Especially in this age the data sets themselves and in the algo- and saying, “We are going to use the
of machine learning and deep learning, it rithms, you will likely exacerbate bias. power of computing and data science
enables us to draw conclusions and make to identify and mitigate hate.”
predictions without an underlying theory. These biases sneak in [when] there
isn’t much data. And it can also get cor-
related with other factors. I personally
worked on interpreting bios and CVs

_ > R E

_

_ 38

_ ===================================================

MIT Technology Review The Computing Issue

IMAGIN

=================================================== I N G
_ Faced with the world’s need for better materials to
_ address climate change, Alán Aspuru-Guzik has an
_ audacious vision of digitizing the discovery process.
_

---------------------------------------------------

MATERI

--------------------------------------------------- A L S

_ By Simon Lewsen

_ ===================================================

_ Photographs Derek Shapton

_

---------------------------------------------------

DISCOV

---------------------------------------------------
_ Opposite:
_ Aspuru-Guzik is a leading evangelist for using
_ computer science to transform chemistry.
_



40 Computing

serial entrepreneur, Aspuru-Guzik just

may have the right mix of computational

expertise and imagination to connect the

multiple tools essential to making it hap-

pen. He has emerged as one of the more

convincing evangelists for the new way of

W hen Alán Aspuru-Guzik, a Mexico City–born, Toronto- doing chemistry.
based chemist, looks at climate-change models, his “Alán can see beyond what people think
eyes gravitate to the error bars, which show the range
of uncertainty surrounding any given prediction. “As is possible,” says Joshua Schrier, a Fordham
University chemist and frequent collab-

scientists,” he says, “we have a duty to contemplate worst-case scenar- orator. He is the kind of innovator, says

ios.” If climate change proceeds as expected, humanity might have a Schrier, who changes the way everybody

couple of decades or so to come up with materials that don’t yet exist: around him practices science.

molecules that enable us to quickly and cheaply capture carbon, and For Ryan Babbush, head of the quan-

batteries—made of something other than lithium, a metal that is costly tum algorithms team at Google, Aspuru-

and difficult to mine—to store the global supply of renewable energy. Guzik’s most prominent character trait is

his creative restlessness. “Alán spends his

time and energy on the newest thing, the

most uncharted territory,” he says. “He

And what if the situation gets worse That’s the idea, at least. Actually execut- doesn’t stick around and focus on incre-

than we expected it to? The need for new ing it is another matter. The structures of mental developments.”

materials will go from urgent to extremely molecules are mind-bogglingly complex, That can be a problem given the time

urgent to dire. Could we quickly come up and chemical synthesis is often more art and hard work it takes to bring a new

with the things we need? than science, defying efforts to automate material to market—an undertaking that

Aspuru-Guzik (one of MIT Technology the process. But advances in AI, robot- requires dogged, narrowly focused research

Review’s 35 Innovators Under 35 in 2010) ics, and computing are bringing new life and endless business patience. But ulti-

has devoted much of his life to versions to the vision. mately, Babbush says, Aspuru-Guzik is

of this question. Materials discovery— Aspuru-Guzik cochaired a 2017 work- interested in reimagining the process of

the science of creating and developing shop in Mexico City where 133 partici- materials discovery, equipping scientists

useful new substances—often moves pants—including Nobel Prize–winning in the community with the computational

at a frustratingly slow pace. The typi- scientists and representatives from 17 and automation tools they need to speed

cal trial-and-error approach, whereby national governments—came together to up their job.

scientists produce new molecules and focus the global research community on Today, Aspuru-Guzik is building a lab

then test each one sequentially for the this goal. The conference was a pivotal in Toronto where AI algorithms design

desired properties, takes an average of moment, helping take the field of accel- novel molecules, and robots quickly make

two decades, making it too expensive erated materials discovery from a niche and test them. The lab is a kind of proto-

and risky for most companies to pursue. area of inquiry to a worldwide priority for type, meant to demonstrate how materi-

Aspuru-Guzik’s objective—which he many of those attendees. After the event, als discovery might work in the future. “I

shares with a growing number of computer- Canada, India, and the EU, among others, want to enable a whole new era, the age

savvy chemists—is to shrink that interval began investing in initiatives to speed up of materials on demand, where every lab

to a matter of months or years, enabling material research. can easily create new compounds,” he

humanity to quickly amass an arsenal of The work itself is ambitious and tech- says. In the future, he hopes, we’ll be bet-

resources for fighting climate change, like nically difficult because it spans so many ter positioned to address the next global

batteries and carbon-capture filters. The disciplines. But as a chemist, software crisis. “The problems of the world require

goal is to revive the moribund materials engineer, AI pioneer, quantum computer molecules,” he adds. “And right now, we

industry by incorporating digital simu- programmer, robotics enthusiast, and suck at making them.”

lations, robotics, data science, artificial

intelligence, and even quantum comput-

ing into the discovery process.

Imagine computer programs that use

precise knowledge of molecules’ elec-

tronic structure to create new designs;

imagine robots that make and test these

molecules. And imagine the software ----------------------------------------------------
and robots working together—testing Opposite:
molecules, tweaking designs, and testing Aspuru-Guzik’s passions (from top left) range from
again—until they produce a material with stickers for street art to lab robotics to Mexican

the properties we’re looking for. lucha libre masks to automated fluid handling.

Materials discovery 41

----------------------------------- his intestines and cauterize the ruptured level, you can begin to make inferences
Battle scars blood vessels, leaving him with a scar that about the resulting material: how it com-
============ runs, like a median line, down the center bines with other materials, how hard or
of his abdomen. soft it is, or how quickly it decomposes.
Aspuru-Guzik speaks exuberantly, digres- At least that’s the idea. But for most mate-
sively, and very quickly. When I first vis- After this early brush with mortality, rials the Schrödinger equation becomes
ited his office at the University of Toronto, he became committed to a life of intellec- too complicated for even today’s largest
he showed me a collection of lucha libre tual adventurousness. If a field of inquiry supercomputer.
(Mexican wrestling) masks—bright blue, intrigued him, he’d pursue it, even if it was
green, and pink balaclavas adorned with esoteric or beyond his expertise. To make the math doable,Aspuru-Guzik
Aztec patterns. “The masks are a human- set about creating versions of the equation
ization tool,” he says. “You bring a Nobel At the time, there was great excitement that require fewer approximations, mak-
Prize winner or an executive from Hitachi over the possibility of using computer- ing them more accurate—a project that
into your office, and after talking for a while, based modeling to design molecules with became the focus of his doctoral studies at
it’s good to stop and say, ‘Pick a mask. Take desired properties, forgoing slow and the University of California, Berkeley. The
a selfie.’” It’s hard not to view the masks as tedious experiments. Scientists talked goal was to streamline the calculations to
a metaphor for his multifaceted life. about a new era of virtual chemistry, only the point where a computer could handle
it didn’t work very well. Computers were them but not so far that the model became
Aspuru-Guzik grew up in a half too slow and molecules too complex. scientifically useless. Using Aspuru-Guzik’s
Catholic, half Jewish family of writers, algorithms, a researcher could model—
musicians, and architects. As a 19-year- While browsing journals in the uni- that is, simulate—a random molecule and
old chemistry student at the National versity library, Aspuru-Guzik came across immediately make predictions about the
Autonomous University of Mexico, he was a paper about the challenges of doing properties of the resulting substance.
returning from an overnight rave in the molecular chemistry inside a computer.
city of Cuernavaca when the car he was In 1926, the physicist Erwin Schrödinger Other scientists had designed similar
riding in veered off the road and crashed. had published an equation to predict the algorithms, but the ones Aspuru-Guzik
Surgeons had to open his belly to repair behavior of subatomic particles, like elec- came up with as a grad student were
trons and protons. If you can mathemat- impressive enough to get him a job at
ically model a molecule at the subatomic Harvard when he finished as a postdoc
at Berkeley. As an assistant professor at
Harvard—and as director of the Aspuru-
Guzik research group, a 40-person team of
computer scientists, biologists, engineers,
physicists, and chemists—he threw him-
self into an initiative called the Harvard
Clean Energy Project. Most solar panels
use silicon to transform sunlight into elec-
tricity. But were there cheap, easy-to-make
organic substances that could do the job?

Over six years, Aspuru-Guzik and his
team ran simulations of 2.3 million dif-
ferent organic molecules to see which
might have photovoltaic properties. He
was hardly the first researcher to practice
virtual chemistry, but he was doing it at
an unprecedented scale. The increased
computing capacity of the era meant that
a single molecule could be simulated in a
matter of minutes; in the 1990s, such sim-
ulations had taken days. Most important,
he had access to seemingly limitless server
space, much of it borrowed from other peo-
ple’s devices. In a system akin to the old
SETI@Home program, people who wanted
to support the project could download a
screen saver that would temporarily lend
their hard drive to Aspuru-Guzik and his
team. “We had one of the biggest super-
computers in the world,” he says, “but it
was distributed all over the planet.”

42 Computing

In the end, Aspuru-Guzik discovered of quantum mechanics, the devices are Aspuru-Guzik’s career was flourishing, but
many organic materials that could, theo- better suited, at least in theory, to simu- suddenly the prospect of remaining in the
retically, be used for photovoltaic cells. The lating molecules. country no longer appealed to him. One
problem was that these winning molecules week after the election, he began email-
were too complicated to be manufactured In practice, though, somebody had to ing colleagues in Australia and Canada,
cheaply. “My mistake,” he says, “was not figure out how to make the simulations looking for a new job.
consulting with organic chemists at the work. In 2014, Aspuru-Guzik and a team
beginning to find out which molecules of researchers released the Variational The University of Toronto offered him
were easily makeable.” Quantum Eigensolver (VQE), a program a prestigious government-funded position
to model molecules, albeit on small, error- meant to lure top-tier researchers to the
With the Clean Energy Project, prone quantum devices that, unlike all- country and a cross-appointment at the
Aspuru-Guzik had basically been doing purpose quantum computers, actually exist Vector Institute for Artificial Intelligence,
combinatorial chemistry—the old trial- today. While the Schrödinger equation is a a nonprofit corporation cofounded by
and-error approach—inside computers kind of abstraction—a mathematical for- machine-learning pioneer Geoffrey Hinton
instead of inside a lab. Then, beginning mula meant to describe subatomic parti- that is quickly making Toronto a global hub
in 2012, researchers in Toronto and else- cles—the VQE uses quantum bits to mimic for AI. The biggest inducement, however,
where made a series of breakthroughs the behavior of the particles in a molecule, was a promise to build a radical new mate-
in deep learning and other methods of much as players in a reenactment might rials lab called the Matter Lab, a project
machine learning. Like many chemists perform the Battle of Gettysburg. Aspuru-Guzik had dreamed of for years.
looking for new materials, Aspuru-Guzik
transitioned to AI, which enabled him to In time, as companies develop more -----------------------------------
discover molecules in a faster, more delib- powerful quantum computers, the VQE
erate way. “The computer simulations are could enable chemists to run strikingly “Fuck it”
like a machine gun shooting randomly in accurate simulations. These models might
the air in the hopes of getting a hit,” he be so precise that scientists won’t need to ==========
says. “AI is a sniper. It chooses a target synthesize and test the materials at all. “If
and takes aim.” we ever reach this point,” Aspuru-Guzik “In the Matter Lab, we only attack a prob-
says, “my work in materials science will lem after asking three questions,” says
First, he had to train a neural network be done.” Aspuru-Guzik. “Does it matter for the
by feeding it a data set describing the world? If not, then fuck it. Has some-
molecular composition and chemical prop- When Donald Trump was elected body else already done it? If the answer
erties of 100,000 organic substances. The president of the United States in 2016, is yes, there’s no point. And is it remotely
AI program could start recognizing pat-
terns—that is, correlations between a given
molecule and the substance it forms. It
could then use this knowledge to invent
candidate molecules to be synthesized
and tested in the lab. With the help of AI,
Aspuru-Guzik discovered new organic
light-emitting diodes, or OLEDs, that
were brighter than typical LEDs. He also
identified new chemicals to be used in
future organic flow batteries, massive
industrial batteries that won’t require
metals like lithium.

Meanwhile, he threw himself into the
nascent field of quantum computing. The
Schrödinger equation is hard to run on
classical computers precisely because
electrons and protons don’t obey the laws
of classical physics. They operate, instead,
according to quantum mechanics: they can
be entangled (behaving in concert with one
another, even if they aren’t connected), and
they can exist in so-called superposition
(occupying multiple opposing states at the
same time). The math required to model
these complex phenomena is dizzyingly
complex, too. But because the qubits in
quantum computers also obey the laws

Materials discovery 43

possible?” Here, the word “remotely” is key. through the plastic hoses to an analytical operatic vocals and macabre stage antics.
Aspuru-Guzik wants to tackle challenges machine the size and shape of a mini- He named the character Bruho (a varia-
that are within the range of feasibility, but fridge, which separates out unwanted tion of “brujo,” Spanish for sorcerer) and
barely so. “If a material is too easy,” he says, by-products. The refined material will flow decided to impose his artwork on the urban
“let other people find it.” into another robot that will test it to learn landscape. He bought a sticker printer
about its properties. Then the robot will and began plastering the Bruho avatar on
Located in a postwar brick building feed the results of the experiment back mailboxes and streetlights. Soon he was
in downtown Toronto, the lab is unlike into the ChemOS program, enabling the part of the city’s bustling street-art scene.
any other at the university. The ceiling AI to update its data and instantly gener-
is adorned with maroon and burgundy ate a new, better slate of candidate mole- Today, Aspuru-Guzik has two goals for
acoustic panels, an homage to the beloved cules, until—after rounds of predictions, the near future. The first is to design a mod-
Mexican architect Luis Barragán. Tucked syntheses, and tests—a winner emerges. ular, affordable version of his closed-loop
away in an inconspicuous corner is a typ- system that can serve as a model to scien-
ical lab bench—a table with flasks, scales, The idea of an automated, closed-loop tists around the world. He wants to build an
and beakers beneath a fume hood—where discovery system has, partly because all-in-one lab box, containing the ChemOS
graduate students can practice chemistry of Aspuru-Guzik’s tireless advocacy, package along with synthesis and character-
in much the same way their grandparents’ become increasingly popular among the ization robots. With this device, a user will
generation did. One gets the sense that new practitioners of chemistry. Peers punch in a description of a given material,
this workstation isn’t often used. in Vancouver, New York, Champaign- and the system will immediately simulate
Urbana, and Glasgow are building simi- and test candidate molecules. If we are to
In the center is a $1.5 million robot—a lar facilities. These labs are intended as usher in a new era of materials on demand,
nitrogen-filled glass-and-metal enclosure all-purpose, automated spaces of molec- Aspuru-Guzik reasons, the technology has
housing a mechanical arm that moves ular creation. That’s why Aspuru-Guzik to proliferate—and it has to be easy to use.
back and forth along a track. The arm can doesn’t speculate too much about what,
select powders and liquids from an array specifically, the Matter Lab will produce His second medium-term goal is to make
of canisters near the sides of the enclosure next. Such decisions will be dictated by his mark, artistically, on the city of Toronto.
and deposit the contents, with exacting curiosity, perhaps, or by the imperatives
accuracy, in one of a number of reactors. of a global crisis. A few days after my visit to the lab, I
“The robot is like a tireless lab assistant joined him and his crew for a night of stick-
ering and postering. Like his materials work,
----------------------------------------------------- this too was collaborative. Our eight-person
Opposite: group included Soap Ghost, an aloof young
The new materials lab in Toronto combines conventional woman with full-sleeve tattoos; Urban Ninja,
chemistry equipment and the latest in automation and AI. a wiry middle-aged man who arrived pull-
ing a cart with a bucket of wheat paste, a
who mixes chemicals 24/7,” says Aspuru- ----------------------------------- homemade liquid adhesive; and Life, a flinty
Guzik. It can make 40 compounds in a insomniac, his hair split down the middle,
mere 12 hours. Making a mark one half dyed blond like Cruella de Vil’s.
“I’ll keep going until sunrise,” he told me,
Two additional features make the ============= gamely. Everyone had bundles of stickers or
Matter Lab’s experimental setup unique. rolls of posters they’d designed themselves.
The first is software that Aspuru-Guzik and In 2020,Aspuru-Guzik experienced a period
his collaborators designed, called ChemOS. of early-pandemic weight gain,which caused In Toronto, this kind of street art—
It includes an AI system that generates his surgical wound to reopen. At the same which doesn’t require spray paint—is pun-
candidate molecules and a program that time, he felt trapped and bored by the 2D ishable by fines (even though the police
interfaces with the robot, directing it to world of Zoom calls and frustrated at not often look the other way), so we moved
synthesize candidates on demand. being able to roam freely about his lab. His quickly and furtively. Ninja took us down
harried work life had left little space for the an alleyway to a bare plywood wall of a
The second distinct feature is the kind of aimless—or seemingly aimless— boarded-up building, and we descended
“closed loop” nature of the production pro- pursuits that, in the past, had fostered cre- on it with our brushes, covering the surface
cess. To explain how this works, Aspuru- ative breakthroughs. He needed a change. with the paste and papering it with imag-
Guzik points to a pair of narrow hoses at es—a bearded Buddha, a ukulele-playing
the back of the robot. “That’s where the A few months later, he began doodling rat, a Bruho figure, robed like a Jedi. The
pee-pee comes out,” he says. Once a reac- on his computer, drawing a lucha libre assemblage didn’t make a whole lot of
tion is finished, the resulting liquid runs mask resembling Screamin’ Jay Hawkins, visual sense, but it had a kind of anarchic
the rock ’n’ roll pioneer known for his beauty to it. Within an impossibly short
time frame, emptiness had given way
to multiplicity, and Aspuru-Guzik was
thrilled. “This wall was blank a minute ago,”
he exclaimed. “Look at it now.”

Simon Lewsen is a Toronto-based
magazine writer.

44 Computing

The upper module
of ASML’s next-
generation EUV
machine was built
from a 17-ton piece
of milled aluminum.

ASML 4455

INSIDE THE MOST COMPLICATED
MACHINE ON THE PLANET

—
Moore’s Law was grinding to a halt—until a small

Dutch firm pushed the limits of physics.
—

by Clive Thompson
Photographs by Christopher Payne

46 Computing

Patrick Whelan peers through the faceplate of his
clean-room bunny suit to see how things are going.

Before him is a gleaming chunk of glass, roughly the size of a contraptions zipping back and forth while being tested. It
toaster oven, that is carved with so many scooped-out sections weighs 30 kilograms, but it moves in a blur.
to reduce its weight that it looks like an alien totem. Whelan’s
team is gluing it to a large, coffee-table-size piece of aluminum. “This is accelerating faster than a fighter jet,” Whelan
Both metal and glass are eerily smooth, having been polished for says, his close-cropped beard and glasses obscured by his
weeks to remove minute imperfections. Over the next 24 hours, gear. “If there’s anything that’s loose, it’ll fly apart.” What’s
as the glue solidifies, workers will neurotically monitor the posi- more, he says, the apparatus has to stop on a spot the size
tion of the glass and metal to make sure they fuse together just so. of a nanometer—“so you have one of the fastest things on
earth settling at pretty much the smallest spot of anything.”
“These will be placed together to microns of precision,”
Whelan tells me, gesturing at the apparatus. This combination of speed and accuracy is key to keeping
up with Moore’s Law—the observation that the number of
A nearby technician worries he’s too close, and yelps: transistors crammed into a microchip doubles roughly every
two years as components become ever smaller, making the
Back up! chips cheaper and more powerful. The more tightly you pack
“I’m not touching! I’m not touching!” Whelan says, laughing. transistors, the faster electrical signals can zip around the chip.
Precision is serious business here. I’m in Wilton, Connecticut, Since the ’60s, chipmakers have shrunk the components by
switching, every decade or so, to a new form of light with a
in a clean room of the Dutch company ASML, which makes the smaller wavelength. But by the late ’90s, manufacturers were
world’s most sophisticated machine for lithography—a crucial stuck at 193-nanometer light—and they were hotly debating
process used to create the transistors, wires, and other essential what to do next. The situation grew more and more dire.
components of microchips. It’s a coveted device, with models Chipmakers had to use increasingly complex designs and
costing as much as $180 million, that is used in making micro- techniques to keep Moore’s Law going, but they managed
chip features as tiny as 13 nanometers at a rapid clip. That level to eke out another two decades of increasing performance.
of precision is crucial if you’re Intel or TSMC and want to man-
ufacture the world’s fastest cutting-edge computer processors. Then, in 2017, ASML unveiled its production-ready EUV
The final machine, assembled at ASML’s headquarters in the machine, which uses light with a wavelength of just 13.5 nanome-
Netherlands, is the size of a small bus and filled with 100,000 ters. With a wavelength that short, chipmakers could pack tran-
tiny, coordinated mechanisms, including a system that gener- sistors more densely than ever before. CPUs can crunch numbers
ates a specific wavelength of high-energy ultraviolet light by
blasting molten drops of tin faster, use less power, or just
with a laser 50,000 times a get smaller. The first gen-
second. It takes four 747s to erations of chips with tiny
ship one to a customer. EUV features are already
at work for huge firms
“It’s a very difficult tech- like Google and Amazon,
nology—in terms of com- improving language transla-
plexity it’s probably in the tion, search-engine results,
Manhattan Project category,” photo recognition, and even
says Sam Sivakumar, Intel’s AI that, like GPT-3, talks and
director of lithography. writes with an eerily human
quality. The EUV revolution
Here in Wilton, the glass- is also reaching everyday
and-metal module that consumers, since ASML’s
Whelan and his team are machines are being used
building is particularly crit- to make chips for products
ical. It will carry the patterns
needed to make a microchip, THIS PAGE: This glass clamp (black
and it’ll whiz back and forth rectangle, upper center) is used
while the machine blasts to hold masks, which contain chip
it with extreme ultraviolet patterns to be transferred to a wafer.
(EUV) light, illuminating
different parts of the chip FACING PAGE: ASML uses this
pattern. The light will then orange robot, built by KUKA
bounce down to a dinner- Robotics, to move heavy pieces of
plate-size wafer of silicon, EUV machines around the clean-
burning the pattern in place. room floor.

Whelan walks over to a
video monitor that shows
one of these glass-metal

ASML 47

including some Apple smartphones and Macs, AMD proces- is simple. You design the components of a chip—its wires
sors, and Samsung’s Note10+ phone. As EUV machines become and semiconductors—and then etch them into a series of
more common, it’ll boost the performance and reduce the power “masks,” much as you make a stencil to put a pattern on a
demands on ever more everyday devices. EUV technology also T-shirt. Then you put each mask over a silicon wafer and shine
enables simpler designs, which lets chipmakers move faster light through it (roughly equivalent to spraying paint over the
and produce more chips per wafer, resulting in cost savings stencil). The light hardens the “resist,” a chemical layer on the
that can be passed on to consumers. surface of the wafer; then other chemicals etch that pattern
into the silicon. In the ’60s, chipmakers used visible light for
The success of EUV lithography was far from guaranteed. The this process, with a wavelength as small as 400 nanometers.
light is so devilishly hard to manipulate that for years experts Then they shifted to ultraviolet light, at 248 nm, and gradu-
predicted ASML would never figure it out. In fact, ASML’s rivals, ally reduced it to 193 nm—what’s often called deep UV. Each
Canon and Nikon, both gave up trying years ago. So ASML switch bought them several years’ extension of Moore’s Law.
now has a corner on the market: if you want to create the most
cutting-edge processors, you need one of its machines. ASML But by the late ’90s, they’d focused deep UV as narrowly as
makes only 55 of them a year, and they sell briskly to the indus- they could manage, and they weren’t sure how to go smaller.
try’s chip giants; currently over 1,000 are installed. It seemed that a new light source was needed. ASML at the
time was a small firm of 300 people that had been successfully
“Moore’s Law is basically falling apart, and without this selling its deep-UV lithography tools. But to stay relevant, they
machine, it’s gone,” says Wayne Lam, a director of research at realized, they’d need to do some serious R&D.
CCS Insight. “You can’t really make any leading-edge proces-
sors without EUV.” Benschop—a tall, angular executive with an exuberant
but wry manner—was hired as the first research employee.
It’s extremely rare for a single firm to possess a monopoly on He started going to big conferences, held twice a year, where
such a key part of microchip production. Even more astonish- deep thinkers from major chip firms and government agen-
ing is the sheer grind of work: it took ASML $9 billion of R&D cies would stroke their chins and argue about what form of
and 17 years of research, a nonstop spree of experimentation, light to use next.
tweaking, and “aha” breakthroughs. EUV is now here—it’s
working. But the effort and time it took to make it happen— “What would be the next kid on the block?” was how
and its late entry on the scene—raises some inevitable ques- Benschop put it when we spoke on Zoom this past summer.
tions. How long will EUV be able to keep Moore’s Law going? The experts pondered several options, all of which had huge
And what will happen next? problems. One idea was to use a spray of ions to draw pat-
terns onto chips; that would work, but nobody could figure
When Jos Benschop joined ASML in 1997, he’d come off a out how to do it rapidly at scale. The same went for shooting
long stint with Phillips and landed smack dab in a chip indus- beams of electrons. Some advocated for using x-rays, which
try worried about its future. Over decades, engineers in chip have a tiny wavelength, but they had challenges of their own.
fabrication had mastered the art of lithography. The concept The final idea was extreme ultraviolet, with a wavelength that

48

ABOVE: This polished optic is RIGHT: A closer look at a
part of an energy sensor that polishing unit. The pieces of
helps control the intensity of light glass shown here are set at
inside lithography machines. angles to achieve the correct
bevel.
BELOW: These polishing
units are used to smooth down OPPOSITE: A few optics
components that go into ASML’s like the one shown above are
EUV machine. A component can mechanically polished.
spend many weeks in a multi-
stage polishing process, with
technicians checking smoothness
down to nanometer precision.