0:00

This message comes from NPR sponsor

0:02

Rosetta Stone, an expert in language learning

0:04

for 30 years. Right

0:06

now, NPR listeners can get Rosetta

0:08

Stone's lifetime membership to 25 different

0:11

languages for 50% off. Learn

0:14

more at rosettastone.com. You're

0:19

listening to Shortwave from

0:21

NPR. Hey,

0:24

short wavers. I'm your host today, science

0:27

correspondent Jeff Brumfield. Earlier this year, I

0:29

went to Israel to cover the ongoing

0:31

war that began after the Hamas attack

0:33

of October 7th. And I

0:36

found myself on a grassy hill overlooking

0:38

the city of Hebron. Hebron

0:40

is in the occupied West Bank.

0:43

It's home to hundreds of thousands

0:45

of Palestinians, but also living there

0:47

are some militant Israeli settlers. It's

0:49

a tense place, full of soldiers

0:52

and checkpoints and high-powered

0:54

security cameras. Even

0:56

here, you can see there's cameras. There's cameras

0:59

sticking out from the rooftops, sort of peeping

1:01

out of the corners of houses. And

1:03

yeah, I mean, it does feel like you're

1:05

surveilled pretty much everywhere you go here. These

1:08

cameras are doing more than just

1:11

watching people. They're identifying them, thanks

1:13

to facial recognition. Issa

1:15

Amro is a Palestinian activist and

1:17

long-time resident of Hebron. He says

1:20

the cameras know him. They have

1:22

our own data. It's connected to

1:24

the camera with facial recognition. This is what

1:26

they say, facial recognition. I think it's more

1:28

than that. It's body,

1:30

it's eyes, it's your shapes.

1:33

So it's more than that, okay? And

1:36

the cameras tell Israeli soldiers patrolling

1:38

the city everything about him, before he

1:40

even shows them his ID. They've

1:42

got his life story. I am a

1:45

human rights defender. I

1:48

was in jail many times. They tell me about

1:51

that I'm divorced. How

1:53

many times I passed the checkpoint. They

1:55

know that. Where I've been, you

1:57

know, in certain hours. There.

2:00

So just tell me all of that. For

2:02

a few years now, Hebron has

2:04

been a laboratory for Israeli security

2:07

forces and private companies to test

2:09

out their latest facial recognition software.

2:12

But then came the Hamas attack and the

2:14

Israeli invasion of Gaza. Hundreds

2:16

of thousands of Palestinians began fleeing on

2:18

foot from the north of Gaza to

2:20

the south. Under the watch of

2:22

Israeli soldiers, desperate Gaza

2:25

civilians, whole families fleeing

2:27

their homes. And

2:29

as they fled, many reported passing

2:31

through checkpoints with cameras. Some

2:34

of them contacted New York Times tech

2:37

reporter Sherr Frankl. I had previously

2:39

been a Middle East reporter for 10 years, and

2:41

so they knew me from my time there. And

2:43

they knew that I now cover technology. And

2:46

so they essentially just reached out and said, we

2:48

think something's going on here. We think this might

2:50

be a camera system. Maybe it's like the facial

2:52

recognition program Israel uses in the West Bank.

2:55

Can you start to look into this? She decided she

2:57

is going to try and find out what was going

2:59

on. I went about sort of getting in touch

3:01

with various contacts in Israel's defense ministry, Israeli

3:04

soldiers who were serving in the Gaza Strip, who had

3:06

left, who could potentially tell me what was going on,

3:08

and just trying to report out whether they had

3:11

in fact launched a facial recognition program

3:13

in Gaza. And she discovered that the

3:15

Israeli military was trying to set up

3:18

a new facial recognition system with the

3:20

help of private Israeli companies and, believe

3:22

it or not, Google Photos. So

3:28

today on the show, Shira and I

3:30

are going to talk about how Israel's

3:33

government is using facial recognition and AI

3:35

to track Palestinians and why

3:37

people in every single country on Earth should

3:39

be paying close attention to what they're up

3:41

to. You're listening to

3:43

Shortwave, the science podcast from NPR.

3:56

Come from Doubleday, Publishers of Lessons

3:59

in Chemistry. Be inspired, Read

4:01

at Lessons in Chemistry. The number

4:03

one global best seller with more

4:05

than six million copies sold. Mean

4:07

Elizabeth Side as sixties era scientists

4:10

who brings her smarts and unapologetic

4:12

world of you to a Tv

4:14

cooking show that has the power

4:16

to change lives. Lessons in Chemistry

4:19

is available wherever books are sold

4:21

from Doubleday. Support for

4:23

NPR and the following message come

4:25

from the American Cancer Society. Dr.

4:27

Alpa Patel leads a team that

4:29

researches cancer risk factors, and she

4:32

shares how her team makes an

4:34

impact. We always do what

4:36

we like to think of as actionable

4:38

science. So the work that

4:40

we do makes its way to things

4:42

like nutrition and physical activity guidelines for

4:45

cancer.org, where millions of people come each

4:47

year to learn about

4:49

how they can better prevent cancer.

4:52

To learn more, go to

4:54

cancer.org. So.

4:57

First things first Year Attack Reporter: I'm

4:59

a science reporter. We should probably talk

5:01

briefly about how modern facial recognition works

5:03

cause it's quite closely related to the

5:05

sort of Ai revolution were living in

5:08

right now, right? Boom! Guess very much

5:10

part of this new wave of technology

5:12

that coming out and a I backed

5:14

which is really changing the capabilities but

5:17

we new technology was was able to

5:19

do until now. Right in these facial

5:21

recognition programs are are neural networks right?

5:23

They are similar to sort of chap

5:25

Gp tier cousins. Of those models.

5:28

Exactly. I mean the way I like to

5:30

think of it because I'm actually seeing how

5:32

to. Very technical percent of their creating maps

5:34

of your face. These programs are taking your

5:37

an edge and their mapping every element of

5:39

your face and then storing that as would

5:41

have been as ones and zeroes and then

5:44

when another images upload it scans or image

5:46

and it's has or their matches. Here

5:48

too, we think these he faces actually

5:50

belonged to the same person And just

5:52

like the Say I models, they're shown

5:55

huge databases and they learn to identify

5:57

different faces by training and thus databases.

5:59

A. Exactly. And like any AI system,

6:01

they're only as good as the training that goes

6:03

into them. So the more data you give them,

6:06

the more examples you give it of these

6:08

two people are in fact the same, the more

6:10

they learn to recognize in fact what makes

6:12

up a human face and how they can with

6:14

some degree of confidence say we believe these

6:16

two faces belong to the same individual. So

6:19

the tech itself has gotten very good,

6:21

assuming the algorithm is properly trained, but

6:23

you know, it's just as good as

6:25

the system you're plugged into. So when

6:28

you're at the airport, you know, there are

6:30

facial recognition technologies being used more and more

6:32

now. I've come across them recently. And

6:34

that's a situation, it's a very controlled

6:36

environment. You have a photo

6:39

ID, you are staring straight at a camera,

6:41

you know you're being logged by a facial

6:43

recognition system. But the Israelis

6:45

have been using it in

6:47

different ways in the West Bank and elsewhere, right?

6:50

Sure. I mean, I think the way to think

6:52

about it is in the airport, you're making a

6:54

choice to be scanned by facial recognition software, you're

6:56

deciding to allow that system to scan your face.

6:58

And you know, a lot of people prefer

7:01

it, it makes the makes their time at the airport go

7:03

faster than enough to pull out a passport every time

7:05

and send the reports at least they use these

7:07

systems, you can just have your face scanned. So

7:09

it's seen as time saving and people basically

7:11

see it as a service. The

7:14

difference between that and governments, including the Israeli

7:16

government that use facial recognition is that in

7:19

the West Bank, for instance, Palestinians who

7:21

need to move between two areas because

7:24

they work in one or because they

7:26

just need to visit a hospital or

7:28

get, you know, literally cross a road,

7:30

they're not choosing to be scanned by that facial recognition

7:32

software, there's no opt-in opt-out system. Yeah, I mean,

7:35

you know, one of the things I heard

7:37

at the while I was in Hebron is that

7:40

they don't always even know when they're

7:42

being scanned, right? Like the they have

7:44

high powered cameras that aren't at the

7:46

checkpoints that at least people like Yisrael

7:48

think can pick up their face just

7:51

walking through through the city. Right.

7:53

And again, so there's the checkpoints themselves, which

7:55

the cameras are very visible, they know they're

7:57

being scanned. And then there are a new number

8:00

of cameras that are just scattered through cities and

8:02

and we'll get to the soon I think that

8:04

in Gaza there are drones that fly above head

8:06

there are entire systems that are in place that

8:08

that nobody in Gaza has an opt into they

8:11

they are just living their Lives in Gaza and

8:13

their faces are being scanned. So turning to Gaza.

8:15

What's going on there? How is it different from the

8:17

West Bank? So Israel has

8:19

not had a military presence themselves in

8:21

Gaza since 2005 Which

8:23

is when Israel announced it was going to

8:25

go through a process called the disengagement It

8:27

was going to withdraw its troops from the

8:30

Gaza Strip So any kind of

8:32

surveillance including facial recognition software that's done

8:34

in Gaza is done from afar

8:36

It's done literally from the physical border that Gaza

8:38

shares with Israel There are a number of cameras

8:40

that are set up along that fence It's

8:43

done via drones and it's

8:45

done via online surveillance The

8:47

Facebook pages the Instagram the YouTube

8:49

all the social media of Palestinians who live in

8:52

Gaza is very closely monitored by Israel and this

8:54

is all being done by this Specific

8:57

part of military intelligence called unit 80 200 or

8:59

at least They

9:02

have a central role and what you found out

9:04

what's going on, right? So unit 80 200 which

9:06

is in fact one of the largest units

9:08

in the Israeli military does their

9:10

digital surveillance their online surveillance So

9:13

unit 80 200 had for a

9:15

long time had a very sort

9:17

of basic facial recognition program running

9:19

in Gaza it wasn't nearly as

9:22

Broad as what they had

9:24

in the West Bank because they didn't have the same system

9:26

of cameras and soldiers on the ground But they

9:28

had been keeping track of people that they believed to

9:31

be members of Hamas They had been keeping track

9:33

of people who they thought were relatives

9:35

of those who were in Hamas or

9:37

other You know extremist ismous groups in

9:40

Gaza and that included their images After

9:42

October 7th, they really you

9:45

know, they stepped up that program is the best way I can think of it

9:47

They immediately went through the videos that had been filmed

9:49

on October 7th and any individual who had

9:52

taken part in coming across the border They

9:54

had tried to identify by name and also

9:56

get multiple images of them sometimes pulling from

9:59

their social media And then

10:01

they had very quickly said, well, look, we are

10:03

definitely sending soldiers in. Let's talk about

10:05

how we're going to launch a broader facial

10:07

recognition program in the Gaza Strip. And

10:09

so they ended up partnering with this

10:12

Israeli company called CoreSight. Right.

10:14

CoreSight is one of the the largest and

10:17

most prominent facial recognition companies based in Israel.

10:19

And this is a company that for years

10:21

has kind of touted its ability to recognize

10:24

faces even from photos that only showed a

10:26

very small part of the face. And

10:28

so this was very attractive for the Israeli

10:30

military because they had a lot of images

10:32

from October 7th in which only, you know,

10:34

a very small part of a person's face

10:36

were shown. And I got to say, when

10:38

I read your article, that did raise my

10:41

eyebrows because like facial

10:43

recognition, as we discussed, it's very good when you're standing

10:45

in front of a camera. But

10:48

at high angles with faces partially

10:50

obscured, maybe with low resolution or

10:52

at night, like, you know,

10:55

all of this really degrades the

10:57

software's ability to identify faces. Right.

11:00

Right. And I mean, I think as anyone who has ever seen PR

11:03

claims of a company knows there's a big gulf between

11:05

what a company says it can do and what it

11:08

can actually do. And CoreSight's claims

11:10

about being able to recognize people's faces

11:12

that were like, you know, almost fully

11:14

covered by masks or bandanas or balaclavas

11:16

or whatever. They were pretty

11:18

extraordinary. And I think the Israeli military really

11:20

saw an opportunity to test that. What did they

11:22

find, I guess, when they tried this out? Well,

11:25

they actually discovered that there was quite a

11:27

high failure rate, an error rate of faces

11:29

being misidentified. This became especially clear

11:32

to them because it wasn't just being

11:34

used to identify members of Hamas

11:36

and other militant groups, but

11:38

it was also being used to identify Israelis who

11:40

had been taken hostage and taken into the Gaza

11:42

Strip. Israel wanted to be able

11:44

to look through video footage and say, OK, this

11:46

individual was taken from their home and can we

11:49

spot them anywhere else in Gaza? One

11:51

thing that became interesting to me as I did

11:53

this reporting and I spoke to quite a few

11:55

members of Unit A200 is that they weren't as

11:57

bothered when they misidentified Palestinians because they were not.

12:00

The kind of hundred attitude of like oh well you

12:02

know will take the wrong person in for questioning and

12:04

that's okay because in question and will figure out that

12:06

the wrong person where. The. Other things I'm really

12:08

wild about: your story is like. Because.

12:11

Impart it seem like foresight system was

12:13

a working that. while I guess Unity

12:15

two hundred was also like using Google

12:17

Photos. key talk a little about the.

12:20

So. What they discovered and literally

12:22

killing specifically with the hostages was that

12:24

google photos which is the same google

12:27

photos your I can have an. Our

12:29

phone will upload a bunch of photos and

12:31

then you can identify. In my case I'll

12:33

say okay, this is my child. This is

12:35

my. Cat And then you can then

12:37

search your photos and say please find

12:39

the other photos of my child of

12:41

my cat on Google Photos is really

12:43

really good at this because they're really

12:45

developing the same kind of Ai technology

12:47

to nice faces, not faces the other

12:49

companies are an Israeli Army discovered that

12:51

Google signal as young. This was actually

12:53

so good that it was better than

12:55

this custom built off were for course

12:57

that course I had made for them

12:59

and so they started increasingly using Google.

13:02

Photos taken a fight to Sicily hostages

13:04

on that had been taken on that

13:06

day and they found that even with

13:08

only a very small part of the

13:10

face visible in a photo and video

13:12

on Googles, technology was excellent identifying as

13:14

he's. So. They were just basically

13:16

building photo albums like with a Google

13:19

account. Exactly. They were just uploading photos

13:21

into photo album and and asking Google to find

13:23

that. He says that where the things I will

13:25

note here that this is the free off the

13:27

cell technology that anybody can use And so Google

13:30

didn't know that it's technology was being used in

13:32

this way until I got in touch with them

13:34

and said i'm. I'm hearing from multiple

13:36

people in unit eighty two hundred multiple

13:38

Intelligence Officers at they're using your technology

13:40

and I think they were pretty surprised

13:43

that they were in use by the

13:45

Israeli army. You know, my phone on

13:47

lox with face Id, Google photo, Id?

13:49

airports A mean it's everywhere now, right?

13:51

Sure, I like. I just wonder trying

13:54

to how you're thinking about. Facial.

13:56

recognition both being in silicon valley

13:58

and seeing it everywhere then

14:00

going to Israel and seeing these really extreme

14:02

cases of how it's being used in a

14:04

place like Gaza You

14:07

know much like AI or social media, which

14:09

is something else I cover a lot of

14:11

I sort of just accept that technology is

14:13

here There's no putting the genie back in

14:16

the bottle No government that runs facial recognition

14:18

programs is going to step back that facial

14:20

recognition program and I think what we

14:22

need to think about as individuals is How

14:24

much of our consent is going into this and

14:27

how much surveillance are we willing to accept in our

14:29

everyday lives? so often We

14:32

and mass kind of accept this technology into our

14:35

lives because it makes our lives so much easier

14:37

It really helps us connect to other people or

14:39

it helps us go through airport lines faster Whatever

14:41

it is and then it's only years later that

14:43

we kind of sit up and go. Oh, man

14:45

I gave up so much of my privacy I

14:48

had no idea I was consenting to all that and So

14:51

like really much like any other new type

14:53

of technology I just think it's a really

14:55

important moment for people to say like, okay

14:57

Well, what kind of consent is being given

15:00

and how much do I know about what

15:02

types of privacy? I'm giving up when I when I

15:04

opt into it Well,

15:11

Shira, thank you so much for joining us

15:13

today, this has been a great conversation. Thank

15:15

you so much for having me This

15:20

episode was produced by Rachel Carlson

15:22

and edited by our showrunner Rebecca

15:24

Ramirez It was fact-checked by me

15:27

and Rachel Gilly Moon was the

15:29

audio engineer Beth Donovan

15:31

is our senior director and Colin Campbell

15:33

is our senior vice president of podcasting

15:36

strategy. I'm Jeff Brumfield Thanks

15:38

as always for listening to shortwave

15:40

the science podcast from and beer

15:46

You Support

15:54

for NPR and the following message come

15:57

from indeed when it comes to hiring

15:59

the best way to search for

16:01

a candidate isn't to search at

16:03

all. Don't search. Match with Indeed.

16:05

Get a $75 sponsored

16:08

job credit at indeed.com/shortwave.

16:10

Terms and conditions apply.

16:13

This message comes from NPR sponsor

16:15

Rosetta Stone, an expert in language learning

16:17

for 30 years. Right

16:19

now, NPR listeners can get Rosetta

16:21

Stone's lifetime membership to 25 different

16:24

languages for 50% off. Learn

16:27

more at rosettastone.com. rosettastone.com/NPR.

16:31

Last year, over 20,000 people joined the Body Electric

16:34

study to change their sedentary

16:36

screen filled lives. And

16:39

guess what? We saw amazing effects.

16:41

Now you can try NPR's

16:43

Body Electric Challenge yourself. Listen

16:46

to updated and new episodes wherever

16:48

you get your podcasts.