0:01

I'm Dr. Brian Goldman, host of the CBC

0:03

podcast, The Dose. Each week

0:05

we answer vital health questions that will help

0:07

you thrive, like, what does my mental health

0:10

have to do with my gut? How

0:12

can I prevent melanoma? How much sleep

0:14

do I really need? And how can

0:16

I manage my health without a family doctor?

0:19

I chat with the top experts to bring you

0:21

the latest evidence in plain language, all in about

0:23

20 minutes. Join The Dose on

0:25

the CBC Listen app or wherever you get your

0:28

podcasts. This

0:31

is a CBC Podcast. Hi,

0:37

I'm Nora Young. This is Spark. The

0:39

Paris Olympics kick off in a little over

0:42

a month, and while there's always lots of

0:44

new tech at the games, sports tech, broadcasting

0:46

tech, we've got our eyes on security. This

0:49

summer in Paris, officials are planning a suite

0:51

of security tools bolstered by AI, from

0:54

spotting abandoned packages to predicting the

0:57

movement of crowds. As

0:59

AI security rolls out for major

1:01

public events, how do we balance

1:03

safety, security and privacy? And

1:05

how do we guard against mission creep, where surveillance

1:07

for special events becomes the new normal?

1:10

This time on Spark, AI surveillance at

1:12

home and on our borders. In

1:22

2017, France won the bid to host

1:24

the 2024 Summer Olympics coming up next

1:27

month. In planning for

1:29

security at such a major international event,

1:31

the use of new technologies had to

1:33

be balanced with stringent laws governing not

1:36

only France, but the European Union. My

1:39

name is Medi Gassemi. I'm Head

1:42

of Research and Assistant Professor of Media

1:44

and Communication at the

1:46

Lille Institute of Communication, ISTC.

1:50

And he leads a team focusing on

1:52

the impact of digital media, including AI-based

1:54

surveillance. But

10:00

they say that there hasn't been enough public

10:03

discussion about this law and there hasn't been

10:05

enough transparency in terms of how these technologies

10:07

are implemented and what the motivations are and

10:09

so on and so forth. Because

10:13

I guess, I mean, obviously there's the

10:15

great French tradition of rights, individual rights,

10:17

but then on the other hand, people

10:19

legitimately want to be safe. They're hosting

10:21

a huge international event and they don't

10:23

want a major security crisis.

10:25

So how do you balance those two

10:28

things? I mean, I appreciate the concept of mission

10:30

creep, but I also appreciate that this is an

10:32

enormous international event.

10:35

Yeah, absolutely. I think it's

10:37

important to note that the

10:39

Paris Olympics happen against a

10:41

backdrop. There is context behind this.

10:44

So the terrorist attacks that happened in Paris

10:46

in 2015, so that really marked the general

10:48

public and the general perception in terms of

10:50

security, it was the first time that France

10:52

was attacked on this scale in Paris. And

10:55

the second was basically the catastrophe

10:57

that happened during the Champions League

10:59

final. And what came

11:01

out of this was when France had

11:03

won the bid to host the events in Paris,

11:06

everybody was afraid in terms of like,

11:09

now we have the image of France

11:11

scarred and that's about our national security

11:13

in terms of reputational security. So

11:15

we need to make sure that

11:17

the terrorist attacks and what happened

11:19

during the Champions League final never

11:21

happens again, specifically never happens during

11:23

the Olympics. Yeah. Mehdi,

11:26

thanks so much for your insights on this. Thank

11:28

you, Nora. Mehdi Ghasemi is Assistant

11:30

Professor of Media and Communication at

11:32

the ISTC Institute of Communication in

11:34

Lille, France, where his work includes

11:36

researching AI surveillance and algorithmic monitoring.

11:38

We spoke to him from Paris

11:40

on June the 5th. We

11:43

reached out to four AI companies that won

11:45

the bid for the Olympics video surveillance contract,

11:47

but none of them made themselves available for

11:49

an interview. Thanks

11:55

for watching. that

16:01

was considering the privacy students had vis-a-vis

16:03

a teacher at their school. And

16:06

the court started to add a bit

16:08

of nuance to it, but it was

16:10

still in some ways connected to property.

16:13

Because it was still like while they're in the semi-private

16:15

location of their school, so they can

16:18

have certain expectations. So when

16:20

I read that decision, it doesn't clearly say, well,

16:22

if I was out on the sidewalk or if

16:24

I was in a park, I would get the

16:26

same protection as in that space. But

16:29

we might see some shifts in this in the near future.

16:32

But it seems to me when we're talking about

16:34

what happens in public space, that is one thing

16:36

when you're talking about human beings sifting through

16:38

hours of CCTV footage or

16:40

sitting in a parked car outside

16:42

of my home, and another when you can

16:45

kind of automatically gather images and have

16:47

AI analyze them. So how does

16:50

the ease and scale of image and

16:52

data gathering change things? It

16:54

changes them dramatically. I think courts and other

16:56

parties that might be involved in thinking through

16:59

privacy, protection, and law need to

17:01

be thinking about what should privacy law

17:03

look like today. And it relates

17:06

to something that a number of academics,

17:08

I'm thinking in particular a couple of

17:11

American academics, Woodrow Hartzog and Evan Selinger,

17:13

have referred to as privacy by obscurity,

17:16

which is this idea that for a long time, or

17:20

at least in many different capacities, we

17:22

could expect privacy just by virtue of the fact

17:24

that it is a

17:26

resource drain to have

17:28

a police officer or a number of police officers

17:30

follow somebody for a long period of time. So

17:33

yes, sure, somebody could sit across the street from

17:35

my house and they could watch my house for

17:37

10 days. But

17:40

if you're paying them and you're paying

17:42

them overtime, that's an incredibly resource exhaustive

17:44

task. And so that's going to

17:47

be limited to cases where, you know,

17:49

there's a really strong reason for them

17:51

to be doing that. Now, with all

17:53

kinds of different technological developments, it's

17:56

easy to do that. And there's actually a case from

17:58

the Ontario Court of Appeal. is

20:00

technology does not evolve in a vacuum.

20:02

And I think the developers of

20:04

different systems are cognizant to some degree of

20:06

the legal restrictions on what they can do

20:09

with those technologies. But we're

20:11

really seeing these types of technologies

20:13

that can identify you accurately

20:15

or inaccurately, different types of concerns with

20:17

each of those, and that

20:20

can be used by law enforcement

20:22

agencies or other government agencies, which

20:24

can be highly concerning because of

20:26

the high stakes nature of

20:28

decisions that could be made there, like border uses,

20:30

whether or not you can enter the country or

20:33

stay in the country, law enforcement, whether you might

20:35

be arrested or not. Companies

20:37

can use this. We've seen examples

20:39

of this in Canada, like in

20:41

shopping malls trying to identify, quote,

20:43

unquote, potential shoplifters. There's high risk

20:45

for profiling in that. But

20:48

also individuals. So we're also seeing a

20:50

growth in the development of

20:52

facial recognition systems that are sort of

20:54

marketed to, or at least

20:57

partly marketed to, people to use on other

20:59

people. And the legal regime

21:01

around that is seriously

21:03

lacking. And that can

21:05

be almost as high stakes, or in some cases,

21:07

just as high stakes, if you have somebody who

21:10

is interested for nefarious reasons, like stalking

21:12

or other number of reasons, in identifying

21:15

you and is capable of just downloading

21:17

an app and identifying you, riding the

21:19

bus. They can associate you with all

21:21

of your online activity, like

21:24

your Twitter and your Facebook, et cetera.

21:26

That's a very dangerous situation as well.

21:29

["The You

21:41

Are Listening To Spark"] Any photo

21:43

that exists out there on the internet has probably been scraped

21:45

up. I don't

21:47

think that more cameras equals more

21:50

safety. In fact, I

21:52

think the inverse is true, that

21:54

for particular communities, more cameras often

21:56

mean less safety. It's

21:58

everywhere. So definitely it's... surveillance culture.

22:00

Yeah. I'm Nora Young. Today

22:13

on Spark we're talking about the proliferation of

22:15

facial recognition and other forms of AI surveillance.

22:18

Right now my guest is Kristen

22:20

Thomason who specializes in Canadian law,

22:22

particularly when it comes to public

22:25

space, privacy, and AI and robotics.

22:27

One of the areas she's been researching is

22:29

what happens when images and data are gathered

22:31

not by the state but by

22:33

private individuals. A

22:36

lot of new technologies, surveillance

22:39

and information collection technologies, what I'm

22:41

really noticing as a trend is

22:43

they're being marketed to individuals to

22:45

use on other individuals. A really

22:47

good example of that, but there's

22:49

many, is Amazon Ring. So

22:52

these doorbell cameras that can record everything that

22:54

happens in front of your house. And in

22:56

a lot of instances, depending on the shape

22:58

of the property and everything, they're probably recording

23:00

the sidewalk as well. And

23:02

Amazon has had things like apps that make

23:05

it really easy for people to then upload

23:07

the footage that they film off of their

23:09

Ring camera, share it with their neighbors, engage

23:12

in sort of like an

23:14

informal vigilante type profiling, share

23:17

that footage with law enforcement agencies,

23:19

either voluntarily or when asked by

23:21

law enforcement. And then

23:23

of course, Amazon has, you know,

23:26

this is all being uploaded through Amazon. So

23:28

Amazon has some access to this information. Amazon

23:31

for a period of time was also

23:33

developing a facial recognition tool called Recognition

23:36

to potentially be used by law enforcement

23:38

or individuals to sift through the information

23:40

collected off these cameras. So

23:42

it's actually a very sophisticated,

23:44

very invasive surveillance network or

23:46

infrastructure that is being developed

23:49

in cities, but through

23:51

private individuals who are far less

23:53

regulated and with a lot less

23:55

legal oversight than, you know, the state

23:57

or a company would have if they did the same thing.

24:00

Yeah, I know there have been cases in the

24:02

states where police departments have been sort of working

24:04

with individuals to

24:06

access that, I think

24:08

it was Ring in particular, that video camera footage.

24:10

Are you aware of any of that happening in

24:12

Canada? I'm aware

24:14

of, well, there was an attempt to

24:17

create a law enforcement ring

24:19

partnership in Windsor, Ontario. This

24:22

was pre-pandemic. It didn't develop

24:24

at that time, and I haven't seen anything

24:26

since. We do have this federal regulation

24:29

around commercial collection of information, of

24:32

commercial information. Their

24:35

laws are different here than they are in the

24:37

United States in important ways. I'm not sure if

24:39

that factored into the ultimate failure of that partnership,

24:41

but there has been an effort there. Then

24:44

I'm also aware that in Vancouver, there have been

24:46

discussions with Vancouver police

24:48

trying to set up

24:51

sort of strategies. I'm not sure

24:53

the exact term they use, but

24:55

strategies or networks to be

24:57

able to have access to people's doorbell cameras. I

25:01

understand why people buy doorbell cameras. I understand

25:03

why they have them. There's a range of

25:05

reasons, and it offers some

25:07

conveniences. I think we don't

25:09

always see, especially in some of the

25:11

advertising for these types of systems, is

25:14

the way that they're then networked beyond that. Yeah, it

25:16

seems to me there's sort of ... There's

25:18

the collection of the footage, and then there's this question

25:20

of what happens to the footage after the fact. For

25:23

example, even if we say that a business has

25:25

the right to capture and analyze images on

25:27

their property, can they then

25:29

turn around and sell that footage

25:32

to a security company? Or can I

25:34

sell my video doorbell footage or my

25:36

drone footage or whatever? Can

25:39

I? There's a really important

25:41

test, Chris. Speaking

25:43

generally, it really brings us back to

25:46

another ... I wouldn't say weakness

25:49

per se is a challenge in

25:51

our legal structure around privacy, a

25:54

challenge that is getting increasingly problematic because

25:56

of the kinds of technologies that we're

25:58

seeing, which is is that a

26:01

lot of privacy legal protection

26:03

is premised around an idea of consent,

26:06

which makes sense, because sometimes we want to

26:08

consent to being filmed, or we want to

26:10

consent to sharing our information because we get

26:12

something out of it, or we see

26:15

some benefit to it. Maybe we see

26:17

some collective benefit to sharing. I'm

26:20

thinking of even medical studies and things

26:22

that are really premised on consent and

26:24

we're benefiting other people by participating. Many

26:29

of our privacy laws are premised around this

26:31

idea of consent. And so that

26:34

idea of could a company collect information and

26:37

then sell it off to another, you'd have

26:40

to read the terms of service

26:42

really closely to see if when

26:44

you signed up to share your

26:46

information, you consented to

26:49

a subsequent use. And

26:51

some of our federal commercial privacy legislation is

26:53

in flex, so there's a bill that's being

26:56

under consideration now to update and

26:58

modify some of our privacy laws.

27:00

So some of these things, if

27:03

you want the most up to date and you're listening to

27:05

this, somebody's listening to this podcast later, they might

27:07

want to just check in on what the current

27:10

legal framework is. But this

27:12

idea that individual consent can

27:14

override a privacy expectation or

27:17

interest is becoming increasingly fraught,

27:20

partly because none of us has time to

27:22

read every terms of service. Probably.

27:25

I would venture to guess many of us don't

27:27

read any of them. I speak anecdotally at least.

27:30

So we're consenting to a lot of things that we're not necessarily

27:32

aware of. And even if we did read them, it's not always

27:34

clear. You might agree that they

27:37

can use your information in other ways and

27:39

maybe don't realize what those ways include.

27:42

That ought to be caught by the law

27:44

to some degree, but I think there's a possibility

27:46

that you're agreeing to something without entirely

27:49

realizing it. So there's

27:51

just the time commitment aspects of

27:53

consent. But increasingly with these kinds

27:55

of machine learning based or artificial

27:57

intelligence based technologies, there's also

27:59

a collection. versus individual

28:01

issue. So I could individually consent

28:03

to share my information or I

28:06

might not mind you know that me

28:09

going in and out of my house is caught on

28:11

my ring camera and Amazon has access to that. Like

28:13

I might make that choice but

28:16

there is a way in which that

28:18

information collected all together. So like we're

28:20

each sort of like drops in the

28:22

bucket and when you collect it all

28:24

together into the full bucket of water

28:27

you can analyze it and draw out

28:29

perhaps profitable or insightful or

28:31

revealing insights into

28:34

the kinds of activities that people engage in,

28:36

what they might be interested in buying, where

28:38

they might be going, what their political

28:40

leanings might be, what their network of

28:42

friendships and relationships in the real world

28:44

might look like. And that can be

28:46

as invasive if not potentially more invasive

28:49

as the sort of individual loss of

28:51

personal information. Yeah. I

28:54

mean it seems to me especially in the case of

28:56

artificial intelligence which

28:58

is premised on making correlations in enormous amount

29:00

of data. So I may not be

29:03

imagining that assistant is

29:05

going to correlate what I bought at the

29:07

drugstore with what I did here and what

29:09

I did there and come up with some

29:11

problematic uses that I could not have anticipated

29:14

because I'm not anticipating the uses that AI is

29:16

going to be put to. Absolutely

29:18

and to be honest maybe those companies

29:20

aren't anticipating them yet either. Yeah. Like

29:22

there is an incentive right now to

29:24

collect a lot of information with a

29:28

sort of an expectation probably you know

29:30

a pretty well informed expectation that as

29:33

different techniques evolve and develop we can

29:35

draw out more and more rigorous

29:37

and refined insights from that data.

29:40

And so you know the

29:42

reality with consent too is that at the

29:44

time that I consented if you know three

29:46

years later machine learning techniques have

29:48

developed extensively even the

29:50

company that drafted up the consent

29:53

form wouldn't have known yet what it

29:55

is that I'm agreeing to. And the

29:57

idea that going back and seeking consent

29:59

later criminal

52:00

justice issues like predictive policing, even

52:03

the surveillance of sports stadiums. So

52:05

this is again, it's not just

52:07

about the border or migration or something

52:09

happening over there that we can't relate

52:11

to, but rather because it becomes normalized

52:13

in places like the border or a

52:15

refugee camp, it can then proliferate into

52:17

other parts of public life. And

52:21

just finally, we've talked about how technologies

52:23

are being used against people

52:25

on the move, but to what extent

52:27

can migrants use these technologies themselves, for

52:29

example, just to make TikTok

52:31

videos of their experiences, to use them as

52:33

communication tools, et cetera? Yeah. I

52:36

mean, the majority of my work and the book

52:38

looks at kind of the sharp edges of the

52:40

tech, but there's also so many ways of resisting

52:42

the kind of violent border regimes that

52:44

are happening and also to upscale communities

52:47

and empower communities on the move through

52:49

technology. Maybe that'll be a second book.

52:52

But there are amazing ways that people on

52:54

the move have been using technology to share,

52:56

for example, their experiences with their friends and

52:59

communities on TikTok or migrant talk, as it's

53:01

called, using different archival

53:03

methods, for example, for a psychosocial

53:05

support archive, which is a project

53:08

I know about really interesting, using

53:10

chatbots to get information into refugees'

53:12

hands directly. And so definitely

53:14

that's kind of the other side, being

53:17

creative and finding ways of kind of sitting

53:19

in this joyful resistance of what technologies can

53:21

also do and how we can maybe dream

53:23

of a different world that is also being

53:25

led by people on the move who really

53:27

are the ones who are experiencing this and

53:29

should be in the driver's seat when it

53:31

comes to development of technology, too, that can

53:34

actually assist them. Yeah. Petra,

53:36

thanks so much for talking to us. Thank you so

53:38

much, Nora. Petra Molnar is

53:40

a lawyer and anthropologist and the

53:42

author of The Walls Have Eyes,

53:44

Surviving Migration in the Age of

53:46

Artificial Intelligence. You've

54:01

been listening to Spark. The show is

54:03

made by Nichelle Parisi, Samarit Yohannes, Megan Carty

54:06

and me, Nora Young. And

54:08

by Mehdi Ghasemi, Kristen Thomason and

54:10

Petra Molyneux. I'm

54:14

Nora Young. You can check out back issues of

54:16

Spark, find and follow us wherever you get your

54:18

podcasts. Talk to you soon. Okay,

54:33

Corey, let's get our feet wet. Show us how to

54:35

surf the net. What

54:38

makes this cool is the fact that you

54:41

can point that camera at anything. Cameras can

54:43

be on every corner in the world. I mean,

54:46

and if you can request that data from anywhere

54:48

else in the world over the World Wide Web,

54:50

you're laughing. What

54:53

happens when I can have a look anywhere? For

55:04

more CBC podcasts,

55:06

go to cbc.ca/podcasts.