The Wall St.Journal 24Feb2020

THE WALL STREET JOURNAL. Monday, February 24, 2020 |R5

JOURNAL REPORT |BIG ISSUES

ten. But some of the error rates be-

tween racial groups are so low that

slight differences seem huge. Con-

sider the results of a recent govern-

ment study that found in one algo-

rithm an error rate of 0.01% for

white males and 0.79% for Ameri-

can Indian females—the rate for

one group is 79 times larger, but

still less than 1%.

More broadly, this critique as-

sumes that because some systems

have bias, all of them must. As the

study above found, some of the

best-performing algorithms had

lower error rates on black females

than on white males. Obviously,

government agencies should set

standards so that they procure

only the best-performing systems

that have undergone rigorous pub-

lic testing.

Facial-recognition detractors

also point to hypothetical harms to

disadvantaged communities and ig-

nore tangible benefits, notably, bet-

ter policing to address crimes that

might otherwise go unsolved in

those communities. One important

use of facial recognition is to find

victims of human trafficking, a

crime that disproportionately af-

fects people of color, immigrants

and the poor.

Who’s watching?

Another target of criticism is over-

sight, or the supposed lack of it.

But any operator of a facial-recog-

nition system must comply with

privacy laws. Moreover, every gov-

ernment agency that uses facial

recognition is subject to account-

ability measures.

Could police misuse facial recog-

nition? Of course. This calls for

more oversight and transparency,

not fewer resources to limit their

effectiveness. Law-enforcement

agencies should, for instance, be

required to disclose when they use

the technology, how they obtain

images and how long they keep

them. Lawmakers can place reason-

able limits on the use of facial rec-

ognition—such as requiring a war-

rant to use facial recognition to

track someone.

Rather than setting down road-

blocks to further innovation and

use, we need guardrails to ensure

the public and private sectors use

the technology safely and respon-

sibly.

Mr. Castrois vice president of

the Information Technology and

Innovation Foundation, a think

tank for science and technology

policy, and director of the

foundation’s Center for Data

Innovation. He can be reached at

[email protected].

‘Law enforcement

is mostly using

facial recognition

for routine

activities such as

to secure ports

of entry.’

The Technology Is Both Biased

And Prone to Errors

Despite the Cries of Alarm,

The Technology Is Mostly Beneficial

JIM LO SCALZO/EPA-EFE/REX/SHUTTERSTOCK

L

AWMAKERS who care

about constitutional rights

and widening inequality

must halt the use of facial

recognition in sensitive

domains by government and com-

mercial actors.

These systems are biased in how

they work and how they are de-

ployed, and they are often created

and implemented in obscurity,

without public review or full ac-

countability under the law.

Most systems are developed by

private companies that license them

to governments or businesses. Com-

panies hide behind claims of trade

secrecy, entering into contracts that

bypass public-disclosure provisions

and allow facial recognition to be

deployed in secret. Researchers, law-

makers and the public rarely have

access to examine these systems or

answer critical questions.

When researchers do get access

to facial-recognition technology,

they find it doesn’t work as adver-

tised. A growing body of evidence

shows that facial recognition is fre-

quently biased and error-prone,

with a recent U.S. government au-

dit confirming that some systems

were 100 times more likely to be in-

accurate for black and Asian people

than for white people.

While some proponents of facial

recognition claim these inaccura-

cies aren’t important, since there

are also systems that government

tests show to be “accurate,” this

does not stand up to scrutiny. Gov-

ernment audits, like the one cited

here, are performed in controlled

settings in a laboratory. Real-world

uses are often much messier, rely-

ing on blurry or bad images, with

recent reports from the George-

town Center on Privacy and Tech-

nology showing routine cases of

use that falls outside of safe stan-

dard operating procedure.

The power structure

But inaccuracy is far from facial

recognition’s only problem. Facial

recognition is generally applied by

those who already have power—like

employers, landlords and the po-

lice—to surveil and in some cases

oppress those who don’t. And be-

cause there are no restrictions on

data sharing between private users

and government, regulations that

focus solely on government use

don’t account for the full threat of

this technology.

We already see alleged abuse of

power in practice, in cases like the

Baltimore Police Department and

Freddie Gray protesters. The police

reportedly scanned photos to target

people with outstanding warrants,

according to material posted on the

ACLU of Northern California web-

‘These systems

allow government

and businesses to

intrude into our

lives without

detection.’



A facial-recognition

system in use at Dulles

Airport outside

Washington.

site—chilling free speech and un-

dermining the right to due pro-

cess. (The Baltimore Police

Department and mayor’s office

did not respond to requests for

comment.)

While proponents of facial rec-

ognition focus on beneficial uses,

often pointing to the convenience

of unlocking our phones with our

faces or invoking hypothetical

public-safety improvements, they

rarely ask what we’re trading in

exchange for such convenience,

nor do they acknowledge the lack

of evidence supporting claims

that facial recognition increases

public safety.

It is difficult to think of an in-

dustry where we permit compa-

nies to treat the public as experi-

mental subjects, deploying

untested, and faulty technology

that has been proved to violate

civil rights and to amplify bias

and discrimination. The dangers

of bias and error have serious im-

plications even when the systems

are automating “routine” func-

tions like check-ins at airports.

A call to halt

The idea that we can continue de-

ploying this technology, assuming

“the right rules” will prevent

harm, isn’t supported by evidence.

Technical advancement has his-

torically outpaced the law, and

these systems allow government

and businesses to intrude into our

lives without detection, threaten-

ing our constitutional rights and

enabling suspicionless surveil-

lance and social control.

It is time for lawmakers to halt

government and commercial use

of facial recognition in sensitive

domains. Once this is done, re-

search assessing facial-recogni-

tion’s risks can be undertaken.

Ms. Whittakeris co-founder of

the AI Now Institute at New

York University, which studies

the social implications of

artificial intelligence. She can be

reached [email protected].

W

HEN ITcomes to

new technologies,

it is easy to imag-

ine the worst.

In the 1960s, for

example, many people feared tran-

sistors would spell the end of pri-

vacy, with miniature electronics

used to eavesdrop on private con-

versations.

Today, we see alarm about fa-

cial recognition. But just as tran-

sistors didn’t give rise to wide-

spread eavesdropping, neither will

facial recognition lead to perva-

sive surveillance.

The most common facial-recog-

nition applications are benign.

Millions of Americans use facial

recognition to secure their mobile

phones and tag photos. Mean-

while, there are countless exam-

ples of how facial recognition is

reducing financial fraud at banks,

preventing medical errors in hos-

pitals, protecting small businesses

from repeated theft and robbery,

and improving security and con-

venience at schools and airports.

On the case

The biggest area of controversy is

law enforcement. Some people

fear that facial recognition will

supercharge surveillance net-

works. But Fourth Amendment

protections, coupled with deeply

held views about civil liberties,

put limits on what government

can do.

Aside from a few scenarios,

such as quickly locating an ab-

ducted child or a terrorist, closely

tracking people’s movements isn’t

being considered domestically.

Law enforcement is mostly using

facial recognition for routine ac-

tivities such as to secure ports of

entry and identify victims in

crimes. For example, immigration

officers can check whether travel-

ers’ faces match the passports

they are holding.

Charges of bias or error in the

technology are overblown. Critics,

for instance, say that some racial

groups are misidentified more of-

A Closer Look

State and local measures to limit use

of facial-recognition technology date

back to at least 2016 and have

increased in the past year.

Source: Electronic Frontier Foundation

*These state measures prohibit the use of

facial-recognition technology in conjunction with police

body cameras and, in California, with other mobile

devices.

Passed in June 2016

December

BANON

MUNICIPALUSE

May

San Francisco

June

Oakland, Calif.

October

Brookline, Mass.

Northampton,

Mass.

MORATORIUMON

SOMEPOLICEUSE*

October

California

(through end

of 2022)

Berkeley, Calif.

BANON

SOMEPOLICEUSE*

New Hampshire

2017

Oregon

Somerville, Mass.

2019

2020

January

Cambridge, Mass.

Facial recognition is becoming an im-

portant tool in a range of consumer, busi-

ness and law-enforcement uses.

But is it accurate—and fair?

Facial-recognition software promises to speed

up the job of determining, or verifying, people’s

identity by rapidly checking faces against a data-

base. U.S. Customs and Border Protection has

been testing the technology at airports, using it

to verify the identity of people entering the U.S.

as well as departing travelers. Other law-en-

forcement agencies, such as local po-

lice forces, are using the technol-

ogy to identify suspects or

missing people.

Facial recognition is also

showing up in business

and consumer applica-

tions. Some airlines are

using it to speed up

check-in, for instance,

and some electronic de-

vices use it as a security

measure.

But as the systems spread, civil-liberties ad-

vocates and other groups are calling for a mora-

torium on the use of the technology.

Critics point to several incidents of alleged

abuse of the technology by police. Meanwhile, a

far-reaching government analysis of the most

widely used facial-recognition algorithms found

most of them misidentified Asian-Americans

and African-Americans far more often than Cau-

casians. The research, conducted by the National

Institute of Standards and Technology—a labora-

tory affiliated with the Commerce Department—

also found significant differences in accuracy

when an algorithm is used to compare two

photos to determine whether it is the same

person.

Due to concerns about reliability, some

makers of law-enforcement technology

have said they aren’t including facial rec-

ognition in their products. Government

pressure is growing, as well, with some lo-

cal lawmakers restricting use of the tech-

nology and some members of Congress

voicing opposition.

ShouldGovernment

HalttheUseof

Facial-Recognition

Technology?

BY MEREDITH

WHITTAKER

BY DANIEL

CASTRO

No

Yes