0:00

Mike, I'm in the midst of some house renovation,

0:02

right? And I'm a complete

0:04

novice at DIY. So I've been listing the help

0:06

of some of those apps like

0:08

TaskRabbit. And I've

0:10

gone to TaskRabbit and it asked me this, this

0:13

question, right? So I want to, I'm going to pose it to you.

0:15

All right.

0:16

the app prompts me with, I need

0:18

help with

0:21

Oh, that's a long, that's a long

0:23

list, Ben. That is

0:26

a big list. Uh,

0:28

Give me one.

0:29

I think what I will say is that,

0:31

um, right before we started recording,

0:33

we did some prep and then we took a little break and I

0:36

started my laundry and I think we're

0:38

going to be discussing laundry in a few different

0:40

contexts today. So, uh,

0:42

I will say I will need help with folding

0:44

my laundry after this podcast is done.

0:47

Nice.

0:48

How about you? How about you? What do you need help with?

0:50

Well, likewise, apart from,

0:53

you know, understanding how to

0:55

paint and to tile and all the rest of it,

0:57

I need help with understanding when there's a block,

0:59

not a block. and

1:02

when does Elon get to decide? That's

1:04

what I need help with. and

1:13

welcome to Control Alt Speech. Your weekly

1:15

roundup of the major stories about online

1:17

speech, content moderation, and

1:20

internet regulation. This week's episode

1:22

is brought to you with financial support from the Future

1:24

of Online Trust and Safety Fund and sponsored

1:26

by Modulate, the pro social voice

1:28

technology company, making online spaces

1:31

safer and more inclusive. We've

1:33

got a really exciting chat with Mike Papas,

1:35

Modulate's CEO later on today.

1:37

He's going to be talking about the interesting

1:40

question of whether to build or buy trust and safety

1:42

tools, which is a growing topic of conversation

1:44

in the space. but right now

1:46

I'm here with Mike Masnick, the

1:49

other Mike, the original Mike. And

1:52

old mic.

1:54

I'm not making any, uh, jokes about your age.

1:57

I know how that goes down. Um, how are

1:59

you doing? How's, how's your week been?

2:01

Oh, it's been good. Uh, it's been, um,

2:03

super busy, lots of stuff going on

2:05

as always. and,

2:08

uh, yeah, it's just, just, you

2:10

know, everything's been, kind of great. Crazy

2:13

and nonstop.

2:15

Yeah, we were kind of saying before we

2:17

started recording that we could have recorded two podcasts

2:20

about the number of stories we had this week. There's

2:22

an awful lot, even more than usual,

2:24

I'd say, and lots of really quality reporting and

2:26

interesting stories to talk

2:28

Yeah. Yeah. So we were saying at some

2:30

point, you know, we're going to have to add the

2:32

midweek special extra episode.

2:35

But for now, I do think we

2:37

have plenty of good stuff cover in

2:39

the meantime. But oh,

2:42

and we should mention, by the way, I think

2:44

it's important to note, next week, we

2:46

are going to be off. Yeah. We

2:48

have been able to do a podcast every single

2:50

week, even the week we had planned to

2:52

take off in July, uh, for

2:54

July 4th, because the Supreme Court forced

2:57

us or forced me into a motel

2:59

room to record a quick

3:01

special episode. But, uh, next

3:03

week we'll be off, but we'll be back, the week after.

3:06

Yeah, right. And this is confirmed. We won't be doing

3:08

a midweek podcast. I'm worried that listeners will think

3:10

they've got to catch up even, even more content

3:12

moderation news, which isn't the case, but there's

3:15

a lot going on. Um, We've also

3:18

had some really, nice reviews again

3:20

this week, Mike, and we've had a bunch of people get in touch, in

3:22

response to my request for info

3:24

about where they listen to control or speech

3:27

and what they're doing while they're listening,

3:30

laundry comes up.

3:31

it keeps coming up, right? It keeps coming up. And

3:34

I think it's probably because podcasts.

3:36

Generally are an accompaniment to

3:38

boring tasks like laundry. I don't think it's just controllable

3:41

speech, but there's a few other ones as well.

3:43

Um, we've got some nice responses from folks

3:45

saying they, they went on, long walks

3:47

while listening to the podcast. And, my

3:50

favorite one was a listener who said that they were

3:52

deboning a turkey carcass in preparation

3:54

for Thanksgiving. Um, they were making

3:56

a soup. So I thought

3:58

it was a kind of nice, nice selection of responses.

4:01

Yeah. Can you imagine listening to us while

4:03

de boning a turkey?

4:05

do you think that the person imagines us

4:08

as, as the Turkey,

4:11

I don't know if, I don't know if I want to go there.

4:14

I also got some really nice

4:16

responses from folks about the

4:18

pedestrian flags, so I'd never seen these

4:20

pedestrian flags and we had a few listeners. Thanks.

4:23

Thanks, John, for sending in photos of some pedestrian

4:25

flags from Seattle. They actually exist.

4:27

You weren't pulling my leg. Okay.

4:29

I was being entirely honest with you. I

4:31

have seen them in multiple places and,

4:33

yeah, they exist. And now you've seen pictures

4:36

of them at least, but at some point I,

4:38

I have to imagine that they must exist

4:40

somewhere in the UK also. Um,

4:43

don't know. I will make a pledge

4:45

to, if there is pedestrian flags in the UK,

4:47

I will find them and I will, I

4:49

we'll get we'll get a picture of you standing waving

4:52

a flag. I think that's that's where we need

4:54

to go.

4:54

Okay. Okay. great stuff. So

4:57

yeah, if people want to share where

4:59

and how they listen to control of speech, get in touch with

5:01

us. we've got some messages on blue sky.

5:03

We also got some emails, podcast at control

5:05

alt speech. com. We will get back to

5:07

you. and we thank you for your inputs

5:09

as ever. Right, Mike. So we

5:12

are jumping straight into the many stories

5:14

that we have this week. we both

5:16

like a good news story,

5:19

you know, we both like a kind of interesting,

5:22

relevant, news lead, but we also like,

5:24

stories that are a bit more interactive, a bit more interesting.

5:26

I don't know if you remember the days of Snowfall,

5:28

the New York Times is very interactive

5:30

piece from 2012. your first

5:32

one is, not dissimilar to that, I'd

5:34

say. Is it? it's a great tale and it's

5:37

told really nicely online. Talk us through what that is.

5:39

Yeah, so this is from NBC News and it's by Brandy

5:41

Zdrozny, who, if you don't

5:43

know, hopefully listeners of this podcast

5:45

know Brandy. She's one of the best

5:47

reporters out there on misinformation

5:50

and has done ton of amazing work

5:52

on that over the years. Um, so this is a, a

5:54

brand new article that just came out this week. On

5:57

how Russian propaganda reaches and influences

5:59

the U. S. And it's a very interactive,

6:02

well done piece in that snowfall

6:04

style, the modern version

6:06

of snowfall. and it's just talking about how

6:09

Russian disinformation campaign start and

6:11

basically like how they are creating

6:14

all sorts of, different

6:16

things bits of nonsense and just sort

6:18

of sending it out into the world through

6:21

this process, creating fake videos,

6:23

sometimes AI generated. Sometimes

6:25

it appears maybe hiring actors to

6:27

play roles of people, you know, sort of confessing

6:30

things and just making up

6:32

complete nonsense stories. And

6:34

then it sort of follows through as to where those stories

6:37

go and how they filter onto,

6:39

fake news sites that Russian

6:41

folks have set up. How they find

6:44

various trolls on

6:47

the site formerly known as Twitter.

6:49

and then often it sort of launders its way

6:52

up through the chain until

6:54

eventually someone like a JD Vance

6:56

or a Donald Trump might repeat them.

6:58

And so it actually follows one particular

7:00

story, that was completely made up,

7:03

regarding Ukrainian,

7:05

officials that are close to, Zelensky

7:07

buying very expensive yachts

7:10

and sort of implying that U. S.

7:12

money that is going into Ukraine

7:14

is being laundered Through,

7:17

various people to, purchase

7:19

expensive yachts for Zelensky, and

7:22

that then sort of, you know, filtered

7:24

its way through and again laundered

7:27

its way up the chain to

7:29

the point that at one point J. D. Vance sort of mentioned

7:32

it on some podcast that he was on

7:34

that we shouldn't, you know, like, why, why are we

7:36

sending more money to Ukraine for them to buy expensive

7:38

yachts, which was a thing that did not happen. Um,

7:41

but because it sort of, started from Russian

7:44

disinformation and, made its way

7:46

along through, nonsense,

7:48

peddling American trolls on

7:50

Twitter, eventually JD Vance picks

7:52

it up because that's kind of the world that he lives

7:54

in. And then suddenly there's like this legitimate

7:56

question of, Oh wait, are Ukrainian

7:59

officials buying yachts with American funding? it's

8:02

a really well done story and it's,

8:04

really like, again, it's a very interactive

8:07

presentation. The presentation is really fantastic,

8:10

and they sort of show like how they're doing

8:12

this with a whole bunch of stories and most of which aren't,

8:14

catching on at all, but a few of them

8:16

are, are actually breaking through. Yeah.

8:19

this is what's so wild about this piece

8:21

is the scale at which, this group

8:24

is, doing this work, and the systematic

8:26

nature of it right it's, it's huge

8:29

in terms of scale and like you say, Not many

8:31

stories make it through, but when they do, they really hit.

8:34

the story that's, mentioned as well about,

8:36

a kind of Ukrainian man who's a

8:38

kind of actor, who apparently

8:41

confesses to trying to assassinate Tucker Carlson

8:43

for 4, 000. Wow. Tucker

8:45

Carlson's in Moscow, absolutely wild

8:48

and comes with a whole array of like fake

8:50

images of, burner phones and

8:53

pictures of Tucker Carlson in Moscow and

8:55

all packaged up in a really, you

8:57

know, if you don't know any different, believable

8:59

way. what do we know about this kind

9:01

of group that's behind this and,

9:03

and how, kind of concerning is it that

9:06

so many stories are getting through into the mainstream,

9:08

do you think?

9:09

Yeah. So, you know, there have been a few

9:11

different sort of groups related to

9:14

doing all these things people know about, like

9:16

the Internet Research Agency, which was the

9:18

one that sort of became famous eight years ago.

9:21

This one in particular is from this group called

9:23

Storm 1516.

9:25

And they're creating a bunch of these things.

9:27

And there's somehow, you know, connected

9:29

to, Russian officials, but

9:32

it seems like there's not that much information

9:34

directly on the group, but they're

9:36

able to produce these really impressive things. And

9:39

then, the example you gave of,

9:41

the Tucker Carlson story, like that got

9:43

picked up by a bunch of American

9:45

sources. And. Interestingly, there was

9:47

a story about a month ago, which,

9:50

I don't think we actually covered it on the podcast

9:52

about, you know, U. S. officials charged,

9:55

uh, people associated with RT,

9:57

Russia, Today, the sort of, you know, Russian

10:00

propaganda news outlet. and

10:02

they had been funding these American

10:05

YouTube influencers like

10:07

Tim Poole and Benny Johnson and folks

10:09

like that, to ridiculous amounts.

10:11

maybe we did mention it

10:13

yeah, I think we'll be there

10:13

we did. We briefly mentioned it. and

10:16

those are, you Notably in this

10:18

story, those are the same, same

10:20

people who picked up on this story about

10:22

Tucker Carlson supposedly being

10:24

targeted for assassination. and they

10:27

all ran with it. And so there's this whole

10:29

ecosystem that has been created to sort of launder

10:32

this Russian propaganda. And,

10:34

there's a little bit. that I hesitate when

10:36

talking about this story, because we heard

10:39

there were similar arguments made,

10:41

over the last eight years, really nine years,

10:44

about like Russian propaganda and

10:46

who it influences. And there's evidence

10:48

that it doesn't. It doesn't actually

10:51

do much to influence people

10:53

in some ways, but the thing that it, seems

10:55

clearly to do, and as the campaigns

10:57

are becoming more sophisticated, I think the real sort of

10:59

takeaway from this is that the campaigns are being more sophisticated,

11:03

but it really shows how much

11:05

of the disinformation world is

11:07

really about confirmation

11:09

bias. It is so much

11:11

about not necessarily convincing

11:14

people. But taking people

11:16

who already think this must be happening

11:18

or these bad things must be going on

11:20

and giving them something to feed off of

11:22

and just say, like, yes, well,

11:25

this story confirms my priors

11:27

and therefore it feels true and therefore it

11:29

must be true. And I think.

11:32

people have made this point before, but I think

11:34

it's one that's worth reminding people and thinking

11:36

about it. Last week, we were talking about like risks versus

11:38

harms, but I think, you know, sort of a related

11:40

issue is the confirmation bias

11:43

at the heart of so many of these discussions

11:45

about, online speech and dis and

11:47

misinformation is that it's

11:49

really all about confirmation bias and people

11:51

looking for content to support what they want

11:53

to believe. And what we're

11:56

creating. In this ecosystem

11:58

right now are tools to support

12:01

anyone's bias, and

12:03

to give them content that allows

12:05

them to accept it. And we're all guilty of

12:07

it. Like lots of people would say like, oh, well,

12:09

you know, I'm, I'm better than that. But it's not

12:11

true. I mean, I think everybody at some point

12:13

falls for some nonsense because they want it to be

12:15

true and it fits with their prior beliefs

12:17

on something. But here,

12:20

what we're seeing when the campaigns get this big

12:22

and this advances, it gets easier for

12:24

people who want to believe these

12:26

things to take them and to,

12:29

expand them further. And because

12:31

they're so sophisticated and because they're so advanced

12:33

and they're creating these videos with all the supporting

12:35

documentation, even though it's all 100

12:37

percent fake and 100 percent made up easier

12:41

for these things to sort of catch on become

12:43

stories in some form or another. Okay. And

12:45

then the one other interesting thing that I think

12:47

is coming out of it is that none of

12:50

these stories, even the ones that have broken through. And a lot

12:52

of these stories, they note like just didn't catch on at all,

12:54

but even the few that did catch on didn't

12:56

go huge. Like they weren't like these big

12:59

disinformation campaigns, but they just sort of like

13:01

creep in. Like even I mentioned JD Vance mentioning

13:03

the Ukraine, Ukrainian yachts or whatever.

13:06

He just mentions that in passing in a podcast.

13:08

It doesn't become a major story, But it's just

13:10

one of these things where he's making these arguments about the

13:12

funding for Ukraine right now and

13:14

saying like, Oh, do we want that funding to go towards,

13:17

more Ukrainian yachts? that's not

13:19

the major point of the story, but it just sort of filters

13:21

into the general narrative.

13:23

Is there, apart from people being more immune

13:25

to narratives like this and being aware

13:27

of how stories get laundered

13:30

through this system, which is really clear

13:32

from this, piece, are there other ways that

13:34

we could kind of like slow, the

13:36

spread information through this chain, do you think,

13:38

are there ways of, of spotting it, you know,

13:40

dots that we can trace maybe in the system,

13:43

because it's really tough for anybody to

13:45

be, so on top of all

13:47

of the different narratives out there that they're not. captured

13:50

by one of the many many narratives

13:52

that have been perpetuated by these Russian organizations.

13:55

How can we slow this down a bit more or stymie

13:57

some of the spread?

13:58

I mean, I think there's a few different things. So one, and

14:01

this is the thing that I always go back to, and it always feels like

14:03

sort of a cheap shot, but Media literacy

14:05

is incredibly important. And again, I just said

14:07

like confirmation bias, everybody falls prey

14:10

to it at some point. I've done it. I'm

14:12

sure you've done it. Like there are stories that you just

14:14

want to believe. And so you fall for it, but

14:16

having general media literacy

14:18

and sort of recognizing like, is this

14:20

coming from a trustworthy source? What are the details?

14:22

Who is reporting it? really matters and

14:24

learning like when to be skeptical,

14:27

especially if things that confirm your priors

14:29

is a really important skill. It's not one

14:31

that anyone will ever be perfect at, but. Training

14:33

people on that, I think is important. The

14:35

second thing is just sort of recognizing media

14:37

ecosystems. And this is something that, is

14:40

really, really important in

14:42

the sort of reality

14:45

based world. I

14:47

Sounds fun.

14:48

I, I'm trying to choose words

14:50

diplomatically here, but like in the reality

14:52

based world, mistakes are made.

14:54

Reporters make mistakes, news

14:56

organizations make mistakes, we recognize

14:58

that, but when those mistakes

15:00

are discovered, they tend to

15:03

recognize that, admit that, talk

15:05

about it, make corrections, and do

15:08

things like that. In the

15:10

sort of pure nonsense peddling

15:12

world, which now exists,

15:15

what they will do is not do that,

15:18

right? I mean, they will spin stories, they will

15:20

take fake stories, they will take real story. I mean,

15:22

a lot of these stories will come with a grain of

15:24

truth behind them, because that allows them

15:26

to sort of obfuscate present these things

15:28

in a way that, presents a darker picture

15:30

than the reality says. I mean, we've seen that

15:32

with things like the Twitter files

15:34

and the Hunter Biden laptop, which,

15:37

you know, has been totally misrepresented over

15:39

time. But those sort of start with

15:41

a grain of truth. But the ecosystems

15:43

that are pushing those messages, you'll

15:46

notice that they don't issue corrections.

15:48

They don't hold each other to

15:50

account. They don't, admit

15:53

when they have these kinds of failings, they

15:55

sort of try and cover it up. They don't mention

15:57

it again, or they, they sort of

15:59

dismisses like, Oh, well, you know, we,

16:01

we had just heard that, you know, we were just reporting

16:03

what somebody else said rather than

16:05

admitting that there was a mistake made. So.

16:08

I think there's an element of like watching

16:10

the media ecosystem that you are

16:13

engaged with and seeing how they

16:15

handle when mistakes are made or

16:17

when false things are reported or misleading

16:19

things are reported. And that,

16:21

that is a key indicator for me.

16:23

Yeah. I mean, there's a couple of great quotes in this piece

16:25

from the kind of The right wing grifters

16:27

who not only have made money by being paid

16:29

directly by Russia through,

16:32

through the fake company that you mentioned, but

16:34

also making money, obviously, through the monetization

16:37

of videos about rumors like this, they're

16:39

benefiting twice, but there's some great quotes, you know,

16:42

where they say we actually reported it as being a pox

16:45

and we noted at the time that it wasn't possible

16:47

to verify the claim. It's like, that would have been

16:49

a, if anything, a. Scintilla

16:51

of, um, of a second in which they mentioned

16:53

it and then they would have moved on and

16:56

they would not have gone back to it.

16:57

Yeah, I mean, I think of the quotes someone said at

16:59

one point was like, you know, they reported on it and

17:01

then at the end they were just like, you know, we have no

17:04

real way of verifying this, but, you know,

17:06

we thought it was important. It's like, that is

17:08

not, that is not really

17:10

a good way to handle this

17:11

no, and there's, there's a bigger conversation which we

17:13

should come back to. I'm really interested in, from

17:15

the work I do in my day job around, what are

17:17

the kind of signals that content creators

17:19

and, influencers, inverted

17:22

commas, what kind of, ethics and

17:24

transparency should they have and how should

17:26

the platforms rank them

17:28

or distribute them according to those kind of

17:30

journalistic principles is something that I think a lot, a lot

17:32

of platforms are thinking about. Great.

17:35

share that story in the show notes. It's definitely something to

17:37

go and have a read of. It's really interesting, to

17:39

kind of scroll through and see the different elements. But

17:41

we'll move now onto, our second

17:43

story about laundry and about laundering.

17:46

piece that, came out this week as well, really

17:49

fantastic piece by, Daphne Keller

17:51

from Stanford who stood in for me a couple

17:53

of weeks ago on the podcast. She's written

17:55

for Lawfare about the rise

17:57

of the compliant speech platform. and

18:00

it's a long read. It's definitely worth spending time with.

18:02

There's lots of great elements to it. Essentially,

18:04

Daphne speaks to. around

18:06

a hundred kind of trust and safety experts. And she's

18:08

pulled together all the insights from that

18:10

to basically make the argument that content

18:13

moderation is becoming a compliance function,

18:15

much like, banks have compliance

18:18

functions and factory floors have to

18:20

be compliant to kind of keep workers

18:22

safe, Trust and safety is now moving

18:24

in that direction, and her

18:27

argument is that the regulation that has been

18:29

introduced over the last few years, the Digital Safety

18:31

Act, the Digital Services Act, the

18:33

Online Safety Act, and the Code

18:36

of Practice for online services in Singapore

18:38

have created these conditions where

18:40

platforms are essentially having to be

18:42

able to report on everything they do,

18:45

all the decisions that they make, they're having

18:47

to standardize how they work. They're having

18:49

to make all of the decisions trackable.

18:51

That's a big part of the DSA is that the

18:54

takedown decisions are recorded in a database.

18:57

And that's. Creating a whole

18:59

ecosystem around, auditing

19:01

those decisions and tracking whether those

19:03

processes have been followed correctly, and she

19:05

kind of makes the case that this is essentially compliance

19:08

and she goes on to say that there are a bunch of

19:10

downsides to this, which, are

19:12

ones that we talk about a lot on the podcast, which

19:14

are definitely worth noting and somewhat

19:16

concerning around government having

19:19

overdue influence into speech

19:21

rules on platforms about how,

19:24

there's a kind of overdue focus on, on metrics

19:26

and following the right process, even if the

19:28

outcome isn't the, the best for users

19:30

or for society as a whole. And so,

19:32

you know, it's a really great kind of original

19:35

piece of reporting, I'd say, which we both

19:37

read and we both thought, yeah, this is great. and

19:39

we have to talk about it on the podcast. So here we are.

19:41

yeah, yeah. No, I think

19:44

it's, it is a really valuable

19:46

contribution to the way of thinking about this

19:48

understanding all of these things. And,

19:51

um, I, I think like Daphne

19:53

have some concerns about it. there's

19:55

an argument I can't remember again, if

19:57

I've made this argument directly on this podcast before

20:00

that, I think that there's an important

20:04

role in, having companies

20:07

and top executives at companies thinking

20:09

of trust and safety specifically,

20:12

as a really important,

20:14

uh, component. sort of marketing function

20:17

that is important, centrally

20:19

to the company. I think that a lot

20:21

of companies think of trust and safety

20:24

as a cost center. And we see that

20:26

like there are layoffs all the time. You have to hire a whole bunch of

20:28

people and it's a pain that

20:30

is frustrating to

20:32

many Company executives and

20:34

that comes through when you see them testify

20:37

or when they talk about these things or Mark Zuckerberg's

20:39

recent Comments on this stuff. I think

20:41

he sees trust and safety as

20:43

just a nuisance and a cost center

20:45

for meta And it

20:47

reminds me of the way that

20:50

companies 20, 30 years ago

20:52

viewed customer service. It was

20:54

a cost center. You have to spend all this money

20:56

on people who answer calls from angry

20:58

customers. And you know, you

21:01

want to get them off the phone as quickly as possible

21:03

and not have to deal with their complaints.

21:05

And then some companies began to come around

21:07

to the idea that wait

21:10

a second, like this is one of the major touch

21:12

points we have between our company

21:15

and our customers. This is

21:17

where they are interacting with us. This is where

21:19

they're communicating with us. Customer

21:21

service itself is actually a marketing function.

21:23

It is a way for us to present a, good

21:26

side of what we do to our customers.

21:28

And therefore, we should invest in it because

21:31

that will come back in the future

21:33

in support. And the argument I keep

21:35

making is that We need to move trust and safety

21:37

to that kind of role where

21:39

it is seen as we are making this platform

21:42

better. And in the long run,

21:44

that means more users, more advertisers,

21:46

more whatever, because we are making

21:48

the platform better. That's what a trust and safety

21:50

role should be. It should be seen as an investment,

21:53

not as a cost sink. And

21:55

my worry is That as

21:57

trust and safety moves more and more into being a compliance

22:00

function, it goes further and further

22:02

away from that. It is again, seen as

22:04

like, okay, these are boxes we

22:06

need to check to make the regulators

22:08

happy, not this is what is

22:10

actually best for our users. This is

22:12

what is going to create a better internet,

22:15

a better service, a better, whatever

22:17

overall. Whereas if

22:19

we could get the companies to recognize

22:21

that this is in their best interest, not

22:24

because the auditors say so,

22:26

not because the regulators say so, but

22:28

because it is actually better for the long term

22:30

health of the wider Internet and

22:32

the services themselves, that would be better. So

22:34

my reaction to this is I think it's

22:36

great contribution and I think it's accurate

22:39

and really important, but I worry

22:41

that This sort of, capitulation

22:43

to the idea of it being a compliance

22:46

function actually takes us further from the

22:48

world that we want.

22:49

Yeah. I mean, I do like the idea of. trust

22:51

and safety being like customer experience and

22:53

it going in that direction, but what's the downside?

22:56

Do you think of it being more like compliance? Like

22:58

if it doesn't go that way.

23:00

Yeah. I mean, some of this comes out. It's kind of

23:02

funny as you read Daphne's piece that,

23:04

like everybody is tiptoeing around the

23:06

idea that like, we're not setting the rules.

23:08

We're not telling you what to leave up or take

23:10

down. But like,

23:13

you know, we have all these things that you need to do

23:15

in order to make it work. and

23:18

so. What happens when you have

23:20

that sort of situation, you have auditors involved,

23:22

what everyone is trying to do is basically

23:24

just cover their ass. So that if

23:27

there's a lawsuit, they can say, well, we

23:29

did everything by the book. You

23:31

know, we did the things that the auditors

23:33

told us to do, that the regulators indicated

23:35

were okay. Or in some cases, where

23:38

there have been previous court cases

23:40

or previous, fines, we

23:42

followed the rules that everybody else laid

23:44

out. And so then it's just a, issue

23:47

of checking boxes as opposed to actually

23:49

doing what's right. It is what is, going

23:51

to decrease the liability on us? What is

23:53

going to decrease the legal risk when you have

23:55

auditors involved? That's what you're talking about.

23:58

Not what is actually going to be the best

24:00

for the internet and for the users

24:02

of the internet and for their speech. And

24:04

there is. The potential. that

24:07

those two things align that what the government

24:09

wants, what the auditors suggest, and what

24:11

is best for the users and for speech align.

24:14

But historically, we haven't really seen that

24:16

play out.

24:17

hmm

24:17

And so that's, that's where my fear is.

24:19

yeah, I mean I Daffy mentions

24:21

banking in her piece and

24:24

I wonder if compliance

24:26

is actually that bad. And if banking hasn't

24:29

shown us that compliance isn't

24:31

that bad. So I slightly disagree with

24:33

kind of where you're coming from. And I just want to explain why. So

24:36

banks pre 2008 were

24:39

not dissimilar to tech platforms in some sense,

24:41

right? They had a lot of users.

24:43

They were processing a lot of personal data. They

24:45

were going through millions of transactions every day.

24:48

they had a lot of kind of wealth and money involved,

24:51

and they were very central to people's lives. Not,

24:53

not hugely dissimilar platforms in some

24:55

sense. 2008 happens,

24:57

the trust in banks kind of completely disappears.

24:59

There's course of regulation, and naturally

25:02

there's pushback from the banks and so they

25:04

will start to say things that we actually

25:07

have heard quite recently, right? So that

25:09

they, you know, Regulation would affect

25:11

the bank's ability to lend

25:13

to consumers and would affect the growth of,

25:16

economies and businesses, it would

25:18

make it difficult to, sustain

25:20

their own operations because there'd be an administrative burden

25:23

on regulation and you know,

25:25

it would make it easy for larger banks to

25:28

sustain a place in the market because smaller

25:30

banks wouldn't be able To kind of comply in the same way.

25:33

And then what happened was all of these

25:35

regulations got introduced, right? So

25:38

you had all of these frameworks introduced

25:40

for systemic risks. You had a bunch

25:42

of transparency required around certain

25:44

markets. You had consumer protection

25:46

for certain products. And

25:49

now we all know and understand that

25:51

banks are regulated. and I wonder

25:54

if, that is a not dissimilar

25:56

place to where, where we are now,

25:59

which is compliance,

26:02

is a cost of doing business. If you want to

26:04

get into the speech space, you have to do

26:06

all of these things. And actually

26:09

it's only because there is a

26:11

big pushback against it, that we

26:13

actually think differently about, speech

26:16

and about the kind of state of online speech

26:18

right now. Your thoughts.

26:22

Oh boy. All right. So, um,

26:25

yeah, I, think there are a lot of differences. And

26:27

so one, the banking space was

26:29

very heavily regulated before that. Yes, the regulations

26:32

have changed and they have certainly advanced, but it was certainly

26:34

not a Wild West of regulation

26:36

free stuff that the, the banking

26:38

industry has long been a heavily

26:40

regulated industry. Also, many

26:43

of the concerns that were raised about some of those banking

26:45

regulations have actually proven true.

26:47

I mean, we are now a year and a half

26:49

since Silicon Valley Bank, which is

26:51

the bank with which, my company,

26:54

did business was. All

26:57

their accounts were frozen and all my

26:59

company's money was frozen.

27:00

Oh shit. I

27:03

didn't realise that.

27:04

It was, it was, quite an experience.

27:06

It was actually, this is the funny story was

27:08

the day that happened was,

27:10

a day that we had an offsite to

27:14

work on the trust

27:16

and safety tycoon game, which came

27:18

out last year with lots. And so we had

27:20

game. Great game. Go and play it. Bye.

27:22

I've gone up north, uh, to this very

27:24

nice house that a friend of ours had rented,

27:27

uh, with this amazing view, and,

27:30

in the morning before I left to drive up,

27:32

I saw the news about Silicon Valley Bank,

27:34

basically being shut down. Freaked

27:37

out and that was not a, comforting

27:39

offsite. And we did not get as much work done because

27:41

I spent a whole bunch of time in this beautiful

27:44

house on the phone, trying to figure out if

27:46

our bank accounts was

27:48

available to

27:49

Yeah,

27:50

So, you know, and since then, of course, there's

27:52

been a lot of look at, it looks into what's

27:54

happening with the U S banking system and

27:57

how poorly targeted some of those regulations

27:59

were and how they actually have, you know, the end result

28:01

of this was like Silicon Valley bank got taken over

28:04

by a larger bank and other larger banks

28:06

also have sort of, you know, there's been a lot of consolidation

28:09

and. it has actually slowed

28:11

down certain things and it has created some problems.

28:14

And on top of that, the even

28:16

more important thing is like, you know, Even though

28:18

in some cases, and I know this is one that

28:20

like sets people off in some cases,

28:23

money can be equated with speech

28:25

and expression. That's a whole other

28:27

area that we are not about to go down into.

28:30

But in general, money and speech

28:32

are not the same. The same thing. And

28:34

so regulating money and how people hold

28:36

money is a wholly different

28:38

thing than regulating how people can

28:40

speak and how they can communicate. And so

28:43

I think that the analogy breaks down

28:45

after a certain point. And as soon as you're talking

28:47

about compliance and regulations around

28:49

speech, the end result, and this is a point that

28:51

Daphne definitely makes, is that you are going

28:53

to lead to suppression of certain kinds of speech.

28:55

And that is always at the core, the

28:58

concerns that I have about these regulations.

29:00

Is what happens because as soon as

29:02

you get into that area, almost

29:05

always happens is that the speech

29:07

that is most heavily regulated and suppressed

29:10

is speech from marginalized groups, marginalized

29:12

individuals, people with less power,

29:14

and the more powerful ones, folks are fine.

29:17

But people who, have more controversial

29:20

opinions, but important ones, are

29:22

kept out of it. I mean, we've talked about this where

29:24

there are, Attempts to regulate kinds

29:26

of disinformation that what happens. It

29:28

leads to pulling down information on LGBTQ

29:31

content. It leads to pulling down information

29:34

on health care, uh, things around

29:36

abortion. All of these kinds of

29:38

topics get targeted. By these

29:40

speech regulations. And from the company standpoint,

29:42

if the idea is just to comply, the easiest

29:45

thing to do is just say, we're not going to do

29:47

that. And we're seeing that to some extent, not,

29:49

you know, this is not getting all the way there, but

29:51

there is a story that we had talked about doing, which we didn't do

29:54

though now I'm bringing it up around,

29:56

like how meta Instagram and

29:59

threads are handling, political

30:01

content and they're sort of pushing it, they're

30:03

down ranking it and trying to not have it

30:05

apply, you know, not show up as

30:07

much within the conversations. And that is

30:10

a direct result from all the shit that

30:12

meta is getting for how

30:14

it moderates political content. And so they've

30:16

sort of thrown their hands up and said, it's easier

30:18

for us not to deal with this content at all.

30:21

And therefore we're going to sort of push it down the algorithm.

30:23

We're not going to block it, but we're not going to have it promoted

30:25

as widely. And that's a direct response

30:28

to like people complaining about which content

30:30

they did promote in the past. And then just saying,

30:32

we just don't want to deal with this anymore. The further

30:34

and further you get towards creating

30:36

regulations and compliance burdens for the

30:38

speech, the more it's going to lead to this kind

30:41

of speech being suppressed because it's just not worth

30:43

it.

30:43

yeah. And I do, I do buy the idea that,

30:45

There is going to be an unnecessary

30:48

blocking, deranking, downranking

30:50

of speech and that concerns me.

30:53

but I imagine that probably Banks said similar

30:55

things in 2008. And

30:58

my question is like, in 10 years

31:00

time, You know, 15 years time,

31:02

however far we are from from those regulations,

31:05

kind of context that users are banking

31:07

in better than then? And

31:09

can you see a situation when

31:12

this regulation, this compliance

31:14

agenda when that actually takes

31:16

hold and it looks like it will, because we're seeing

31:19

more and more, regulation, government regulation about

31:21

speech all the time. Can you see a situation

31:23

where the people would adjust to the context and in

31:25

10 or 15 years time have processes

31:28

to better understand, mitigate

31:30

in a way that isn't just about downranking

31:33

or blocking speech as we're worried about?

31:35

Yeah, I mean, you know, we'll see, like, people always

31:37

adjust to the reality that they have around them.

31:39

and so I don't think that that is,

31:42

you know, yeah, I mean, people will adjust and you don't

31:44

know what you don't have. So, you

31:46

know, whatever, whatever world we live in,

31:48

people will say like, yes, this is normal. And this

31:50

is okay. But I think, the idea

31:53

of That, all the regulations

31:55

around finance, and I'm not,

31:58

let me be clear, because it was about to sound like I was

32:00

saying, like, oh, get rid of banking regulations. I am

32:02

not saying that. Not saying that.

32:04

Do you keep all of your money under the floorboards now because

32:06

of what happened?

32:08

yeah, but, we are still

32:10

living in a, in a world today where

32:12

financial scams are massive

32:14

and all over the place. And,

32:17

You can say that maybe some of the

32:19

regulations on banking led to

32:21

the rise in like crypto scams. And

32:23

now we're talking about pig butchering, which, you know,

32:25

sort of relies on crypto scamming. there

32:27

are a bunch of different things like this stuff sort

32:30

of oozes out somewhere else in

32:32

the system. The idea that we can sort of perfectly

32:34

regulate these things is not true because

32:37

humanity is messy, and society

32:39

is messy, and the way that people interact

32:41

is messy. and, so much of these

32:43

regulations, and again, I'm not making

32:45

the argument for no regulations. I'm not saying get

32:47

rid of them, but I'm saying recognize that there are

32:50

consequences to these regulations and

32:52

the unintended consequences lead to, you

32:54

know, the problems ooze out somewhere

32:56

else. And, and we should be very

32:58

careful about where and how those problems

33:01

ooze out. And don't assume that just because

33:03

like the world itself doesn't collapse,

33:05

that these approaches are necessarily the right

33:07

way to do it.

33:08

Yeah. Okay. As you can see from the discussion

33:10

we've had, it's a great piece. I definitely

33:12

recommend listeners go and have a read. Daphne's done a

33:14

great job of adding something new.

33:17

And we could go on and talk about, it

33:19

for, a lot longer, but we, we

33:21

have other stories to get to. We've,

33:23

we've gone, Mike, from there, from a, uh,

33:26

a very thoughtful, well argued

33:28

piece by somebody who's really trying to add to

33:30

the debate to somebody who

33:33

is the antithesis of that. We're

33:35

back talking about Elon Musk. Um,

33:38

we're back talking about Elon Musk. And what's he done

33:40

this week? What should we, what can we be angry about?

33:42

Well, I was about to say, because, you know,

33:44

I wrote an article about this and I think we're going to

33:46

link to my article in the show notes. So I thought you were about

33:49

to say, we went from a thoughtful piece by Daphne to

33:51

a terrible, stupid piece by you, Mike.

33:55

Never. I'd never say that. Don't do it,

33:58

But Elon has continued his,

34:00

belief in changing what used

34:02

to be called Twitter to his liking and

34:04

without any thought whether or not it actually makes sense. And

34:06

so he did two big

34:08

things this week. One is changing the terms of service,

34:11

that will go into effect in mid November,

34:14

in ways that have, um, For the time being, I'm going to

34:16

just raise some eyebrows. But the bigger one, the more noticeable

34:18

one is that, while he's talked about

34:20

this in the past, they've announced officially,

34:22

that they are changing how the block feature

34:25

works on x. Where

34:28

formally, you know, when you block someone, they can't

34:30

see what you're doing and they can't interact

34:32

with you. And if anyone else interacts with,

34:35

you know, it's all, it's all blocked. He's

34:37

never liked that. And so now the way

34:39

it is going to happen is that if someone

34:41

blocks you, you can still see their content.

34:44

You just won't be able to interact with it. You won't be able

34:46

to reply. You won't be able to retweet it. But

34:49

you can still see it. And there is

34:52

to give him a very, very,

34:54

very tiny amount of

34:56

Mike. Don't do it. no

34:59

need.

34:59

there is a non

35:02

crazy way to think about

35:04

this. And that is basically

35:07

when people are posting to a public. system,

35:10

that content is public and anyone

35:12

can see it, right? If

35:15

I post something to tech dirt, I can't

35:17

ban someone from seeing what

35:19

I've written. If you write something and everything in moderation,

35:22

you can't ban someone from seeing it.

35:24

It is public. And therefore there's just a general

35:26

sense that anyone can see it. And this is the point

35:28

that everyone always raises with the way that Twitter does

35:30

blocks or did blocks, which

35:33

is that. You know, if somebody blocks you, you can

35:35

just go into incognito mode and

35:37

you can see that content because it's public.

35:39

And so block is sort of a weird

35:42

function on social media.

35:45

So the theory is like, well, this changes

35:47

that because it's public, it remains public,

35:50

but the reality is very different than the theory

35:52

on this. And the reality is that that

35:54

friction that the block function adds

35:57

makes a real difference in slowing

35:59

down abuse and harassment a

36:01

variety of different kinds of.

36:03

Yeah, it's a first, it's a first line of defense in

36:05

a lot of cases, isn't it, to, people being,

36:08

targeted in some way.

36:09

Exactly. And it's not perfect by any

36:11

means, but it's not meant to be. It is just an

36:13

attempt to add some friction. And often,

36:15

not always, often that amount of friction

36:17

is enough to make a difference in

36:20

very important ways. And so, it

36:23

works for some element of

36:25

what it is trying to deal with, and

36:28

not everything. And what Musk is doing

36:30

is sort of removing it for a large

36:32

segment of the population for which it actually

36:34

does work.

36:35

Mmm.

36:36

And so, the history here is also

36:38

important, which is that 11

36:41

years ago, Dick Costello, who was

36:43

CEO of Twitter at the time, had the

36:45

exact same idea and had

36:47

the exact same thought that Block is stupid and

36:50

we should And

36:52

he did the same thing. He turned it so that you

36:54

couldn't interact with people, but

36:56

you could still see people who

36:58

blocked you. It lasted for,

37:00

I think, two hours before

37:03

the revolt was so loud

37:05

and so angry and so

37:07

public that they rolled it back and

37:09

never spoke about it again. As far as I know,

37:11

there was never any like, you know, postmortem.

37:14

There was no discussion of it. There was a, we're

37:16

rolling it back. We're sorry. We're going to rethink this.

37:18

And then it was never mentioned again.

37:20

Right. Elon doesn't do postmortems.

37:22

no, no, no. And, I mean,

37:24

Elon at times will roll stuff back.

37:26

So it'll be interesting to see if that. I

37:30

am pretty sure that Elon has no

37:32

idea that Twitter tried this before and

37:34

that it failed miserably and was a complete

37:37

disaster and an embarrassment for

37:39

Twitter at the time. and we are starting

37:41

to see there is pushback. I have even

37:43

seen, people who are normally big supporters

37:45

of Elon have been screaming about

37:48

what a stupid idea this is. and

37:50

we're seeing a big exodus

37:52

from Twitter to other sites.

37:55

And so people are recognizing

37:57

that this is a problem and they don't like the way

37:59

the site works.

38:00

yeah, you were saying before we started recording that blue

38:02

skies had this massive, uh, Surgeon

38:04

users, which we're kind of tallying with

38:07

the changes to block. Right. And we, this is where

38:09

I ring the, uh, you know, Mike

38:10

Yes, yes, yes, yes. Disclaimer,

38:13

disclaimer, disclaimer. I

38:15

am on the board of blue sky. I am associated

38:17

with blue sky. Consider me biased. Do

38:19

not take what I'm saying as impartial

38:22

for the next section here.

38:25

I know he can't block people

38:27

on the platform for you or unban.

38:29

You don't get in touch with them.

38:31

Yes. Yes. Uh, yeah, please don't. but

38:34

yes, blue sky has seen a massive influx

38:36

in users. like even bigger,

38:39

you know, blue sky saw a huge influx

38:41

of users when, Brazil,

38:43

uh, Band twitter and

38:45

a whole, you know, basically, you

38:47

know, we jumped up about 4 million users

38:49

the course of a couple of weeks, from that

38:51

band. This is larger

38:54

than that as of right now. I don't know if it will

38:56

last as long, but, right before we

38:58

started recording, I looked at the U

39:00

S app store, iPhone

39:02

downloads and blue sky

39:04

was the number four app. four

39:07

free app. being downloaded,

39:09

which is pretty impressive. We were one ahead

39:11

of TikTok, ahead of Instagram,

39:13

ahead of WhatsApp, ahead of Gmail. the

39:16

only ones that we were behind were threads, which

39:18

also, I believe lots of people

39:20

are probably going to threads, uh, ChatGPT

39:23

and Google. So, having BlueSky, this

39:25

very small player in the space, Being

39:28

the fourth most downloaded app almost,

39:30

and, it was reported, this is not me revealing

39:33

anything, uh, internal that,

39:35

that I'm aware of, but it was reported publicly

39:37

that within BlueSky, they

39:40

refer to, EME, which

39:42

are Elon Musk events that

39:44

suddenly lead to a massive influx

39:46

of users on the platform. And

39:49

so it's interesting. to see

39:51

like how these traces and I'll note again,

39:53

bias, all of that, all the caveats.

39:56

you know, blue sky implements block in a very

39:59

different way. And in fact goes further than,

40:01

Twitter did before, you know, it is referred

40:03

to as the nuclear block on blue

40:05

sky, where when you block someone,

40:07

it obliterates every

40:10

interaction that you have with that person. Whereas

40:12

with blocks on Twitter, the way they, you know, have

40:14

worked for, A very long time is that

40:16

even if you block someone, other people can still track,

40:19

the conversation that you had, they can

40:21

sort of see what happened on blue sky.

40:24

Basically disappears and it is effectively

40:27

impossible to track down and it stops

40:29

harassment campaigns cold. Like

40:31

it is a very, very

40:33

impactful way of doing it. There are some

40:36

ways around some of it, but

40:38

they throw a lot more friction into the gears

40:40

and it has been, it's really powerful.

40:43

And it's. sort of clued me into

40:45

how these very small changes and how

40:47

these things are implemented have huge impact

40:50

in terms of how harassment

40:52

campaigns can and cannot work on

40:54

different platforms.

40:56

Yeah, and are we taking the influx

40:58

of users to blue sky to mean that people

41:00

are interested in how platforms think about safety?

41:03

it something that we think is a kind of conscious decision

41:05

for users or do you think there's people are just shopping

41:08

around for alternatives?

41:09

I don't know the answer. And so this would

41:11

be purely speculation. You know, I think

41:13

that it's just a sense that, Yes,

41:16

that some of the safety features

41:18

matter. I mean, we saw this early on with blue

41:20

sky, you know, when blue sky first launched as

41:22

a beta, it was a private beta invite

41:24

only the first sort of controversy

41:27

was that they hadn't yet implemented block

41:29

at all. And this was, they were just sort of testing

41:31

out the system. And there were like 5,

41:33

000 people on the platform and people were

41:35

so angry. How could you launch this

41:37

without block? Like block was like a, is

41:40

now considered a fundamental feature

41:42

of social media, which is kind of surprising

41:44

because it was not originally, that was a, a

41:46

later addition to all of this. And

41:48

as we see, even Twitter sort of was rethinking

41:51

how it works, but in our minds now people

41:53

have sort of learned to associate

41:55

blocking as a fundamental

41:57

feature. Key safety feature

41:59

of social media that social media needs,

42:02

then how it's implemented really matters.

42:05

And whereas blue sky is gone for this nuclear

42:07

block, Twitter or X is

42:09

now moving in this opposite direction of

42:11

the loosest level of block that

42:13

you could possibly imagine. Something is much more akin to

42:15

the concept of muting rather than

42:17

blocking. And so. it's really

42:20

interesting, and I hope that we'll

42:22

find out more of the reasons why people

42:24

are switching, but, that seems to be the clear

42:26

reason why we're seeing this

42:28

massive, massive influx to BlueSky,

42:30

and I am assuming that Threads and also Mastodon

42:33

are probably seeing a similar influx of

42:36

people. And, uh, You know, I think

42:38

it's great. I think, exploring all these

42:40

different systems and recognizing that

42:42

now you have choices in terms of which

42:44

systems do you think are going to do the most to

42:47

keep you safe and make you feel comfortable on a platform.

42:49

I think that's, that's actually a really good thing.

42:51

yeah, no, it's, it's great that there's alternatives

42:54

out there that people feel they can even

42:56

try. obviously last week we had a story

42:58

about threads and Instagram moderation

43:00

being below par. there being a

43:02

lot of oddities to the moderation

43:04

process as a result of presumed moves

43:07

to AI. And actually this is a kind of

43:09

nice segue onto what will be a, probably

43:11

a final story for this week, Mike, but. a

43:14

bit of an update to the explanation

43:16

for that, those technical glitches.

43:18

Yes. So they came, they came out and said

43:20

that, you know, we, did our whole discussion on it

43:22

and talked about how it was, obviously

43:25

AI because it was so stupidly done

43:27

and like blocking everyone who, who used

43:29

the word cracker, you know, without

43:31

understanding any of the context and how like

43:34

cracker jacks are a snack food and not

43:36

a racial insult. So we, assumed

43:38

naturally that it was an AI driven thing, because

43:40

that was the only thing that made sense. But Instagram

43:42

came out and Adam Asari came out and said

43:44

that it was a human reviewer problem and

43:47

that they had not properly

43:49

instructed human reviewers on how to handle certain

43:51

kinds of context and things like that. And

43:54

I'm not sure I believe them.

43:55

Well, I mean, there's, there's a few parts to this, like

43:58

don't believe him, right? And so you

44:00

think he's putting the emphasis on humans because

44:02

it makes, it's in his interest to

44:05

like disown The human part of this moderation

44:07

process, I think actually there's something

44:09

here about the kind of tooling that the humans were, enabled

44:12

with, like he mentioned about that

44:14

the tools didn't give it the human's context.

44:16

Humans aren't going to be able to kind of moderate effectively

44:18

if they don't have context about something. whether

44:21

or not you're right it's a tooling aspect,

44:24

it's not the fact that there's people involved in the process.

44:26

Right.

44:27

Yeah, that is true. And, and, you

44:29

know, we know that the tools do matter, but

44:32

again, like, you know, kind of what we said last week was

44:34

that it is insane that, meta,

44:36

the sort of most experienced

44:38

and the, you know, largest resourced

44:41

player in trust and safety would

44:43

have tools that were so bad. And we were talking about

44:46

on the AI front, but if that's also true

44:48

in terms of like the tools for

44:50

the moderators, that they don't have the

44:52

context. I understand like. Tools

44:55

for context are a difficult thing, and it's,

44:57

something I've actually spent a lot of time thinking about it's

44:59

difficult because context

45:01

can come in all different ways and could be

45:03

external to, the content

45:06

that you're looking at could be external to the platform

45:08

itself, but not knowing

45:10

that Crackerjack is

45:14

a It's not a slur

45:17

Yeah. The tool has to have been designed very

45:19

badly in the first place for

45:22

that to happen.

45:23

yes, yes. That is my take on it

45:25

is, is I don't like, that seems

45:27

like a huge mistake. And for a company as

45:29

big, like I could see like a very small

45:31

player pulling a kind of off the shelf

45:33

tool somewhere and not really

45:35

implementing it. Right. Like having that issue.

45:38

I have that issue a little bit now with like moderating

45:41

comments on TechDirt. Our tools are not.

45:43

Perfect. And there are a couple of things where like context

45:46

would be really useful that I don't have, and I have to take

45:48

a few extra steps to look at like, Oh,

45:50

who's this person replying to? Because that actually matters,

45:53

Yeah.

45:53

but you know, we're tiny, right?

45:56

And so like meta is

45:58

not right. I I'm, I am

46:01

amazed that they would not have tools in place that

46:03

could handle basic context

46:04

Yeah. Agreed. And I'm not, I'm not a

46:06

big fan of the headline of this article, which

46:08

will include in the show notes. Cause it does, it does

46:10

suggest that he doesn't row back on,

46:12

on that initial claim and he, he

46:15

does. So anyway. it's a complex

46:17

issue as always with content moderation

46:19

and trust and safety. I don't know if we've solved

46:21

any of the issues today, Mike, in, in the

46:24

time we've been talking, but it's been

46:25

I will say, this is actually a really good lead

46:27

into the bonus chat that I have with Mike Pappas,

46:30

where this question of buy versus build,

46:32

there are some issues about, how

46:35

do you choose which things and like, what

46:37

features do you want and how customized

46:39

do they need to be? And that's some of

46:41

what Mike and I talk about in this bonus

46:43

chat, because, these are not easy questions.

46:46

these are really, really important ones. And so,

46:48

I, I think Mike and I get into that

46:50

in this bonus chat. And so actually we

46:52

didn't plan it this way, but it leads in really nicely

46:55

to the bonus chat between, uh, myself

46:57

and Mike Pappas from Modulate. So

47:09

Mike, welcome back to the show. Always

47:11

good to have you here. Always good to catch up. we

47:13

wanted to talk today. You, you recently wrote this

47:15

blog post on the modulate site on

47:17

the issue of building versus buying,

47:20

technology for trust and safety purposes.

47:22

And in particular around voice moderation,

47:24

which I think is an issue that really comes up a lot

47:27

in all sorts of contexts around trust

47:29

and safety right now, and it's, a topic that. I

47:31

heard a lot about at TrustCon and I hear about in

47:33

all sorts of, other contexts as well

47:35

when thinking about this stuff. And I was

47:37

just kind of curious, like, what made you

47:40

want to write about that topic in the first place?

47:42

thanks for having me back, Mike. Always enjoy having

47:44

these conversations. the answer to this

47:46

is in some sense pretty mundane. We are

47:48

a trust and safety tools vendor. So

47:51

this is an inherent sort of existential

47:53

question for us is, does

47:55

it make sense for people to buy

47:57

externally? we're very

47:59

sort of mission driven, ethics driven company.

48:01

We have a lot of deep internal conversations

48:04

about how do we want to think about marketing

48:06

and sales strategy. We don't want

48:08

to trick someone into buying our product

48:10

if it's not actually a good fit for them. So

48:13

I think what's more interesting here is

48:15

really looking at it from that lens of.

48:18

Can we add a kind of value

48:21

by being an external organization

48:23

that you cannot get from

48:26

sort of spinning up an internal team to build

48:28

the same tool? And I think what's interesting is that

48:31

we do feel like there's a couple of areas where

48:33

we can provide value that way.

48:35

Yeah, I mean, I think one of the things, and, and, and, This

48:37

comes up in all sorts of contexts, again,

48:39

like not just within trust and safety field. I mean,

48:41

I think there's lots of areas where everyone is always

48:43

having that build versus buy conversation.

48:46

And one of the things that comes up a lot is

48:48

this idea of like how central is the

48:50

thing to your business. and

48:52

the general sense being, you know, if something is more

48:55

central to your business, you tend

48:57

to lean more towards building versus

48:59

buying. But not always sure that's true,

49:01

right? I mean, I think there are times where, even

49:04

if something is central, there is value

49:06

in. if there is an expert who

49:09

has the ability to do more because of

49:11

scale or because of expertise or

49:13

something else, it actually makes more sense

49:15

to partner with that vendor. And so

49:17

sort of curious where you, how do you, how do

49:20

you feel about that? Or how do you think through those

49:22

things yourself?

49:23

Yeah, I mean, electricity is vital

49:25

to my business. That doesn't mean I build it myself,

49:28

right? To, to your point, there's a lot

49:30

of things that are essential components.

49:33

And because of that essentialness,

49:36

you cannot afford to get it

49:38

wrong. And I think as

49:40

we've seen over the last several years, you know, trust

49:42

and safety has long been

49:44

seen as kind of this cost center,

49:46

this must we do it. And we'll try

49:48

to do it as little as possible. And

49:51

as. Both sort of regulation, consumer

49:54

sentiment, and frankly, just platforms

49:56

that care that are being built

49:58

around that perspective of caring

50:00

are all sort of looking at this and saying, no,

50:03

actually, we really should treat this as

50:05

essential. I actually

50:07

see that coinciding with more

50:09

of them starting to perk their ears

50:11

up a little bit and say, Let's take

50:13

a closer look at what else is out there

50:15

and think about is it worth

50:17

getting that really premium offering

50:19

as opposed to just trying to slap something

50:22

together to check a box?

50:24

Yeah, it's, it's interesting. I, I've

50:26

been thinking about this in a slightly different context

50:28

and I didn't realize I was going to go down

50:30

this path when we started this conversation, but You

50:32

bringing up the idea of people thinking of it as a cost

50:35

center. This is something that I've actually thought a lot

50:37

about lately as well, where,

50:40

years ago, I remember there was this whole debate

50:42

over like call centers and

50:45

customer support and a lot of companies

50:47

viewing that as a cost center and

50:49

then somebody. You know, had the bright idea,

50:51

like, wait, this is the main touch point

50:54

that many of our customers have with our

50:56

company or their service or products. Maybe

50:59

we should realize that like, customer support

51:02

is actually a marketing function rather

51:04

than a cost center that is just, causing,

51:06

headaches for us and then we try and keep as, cheap

51:08

as possible. And in that

51:11

situation too, right. I mean, a lot of people choose

51:13

to still outsource it because they know that

51:15

even, though it is. really important

51:17

feature of their business. It still makes sense to outsource

51:19

it because you can do that at

51:22

a much better rate. do you think of it as

51:24

sort of similar to that?

51:25

Yeah, I think that's right. And I think

51:27

it's, partly, you know, why,

51:29

why we really enjoy working with the kinds of

51:31

businesses that we do is because

51:34

they have that conception

51:36

of this is not just a cost center.

51:38

I mean, games are often built

51:40

by people who are, you

51:42

know, visionaries trying to create this particular

51:45

experience that it's really emotionally

51:47

important to them to realize the experience

51:49

that they had when they Wanted to create

51:52

and this is part of it. If as soon as

51:54

you step into that wondrous world,

51:56

someone's calling you the N word. It's

51:58

not so wondrous anymore, right?

52:01

and I think even with some of the enterprise

52:03

platforms that we've been working with more recently,

52:05

we see similarly. It

52:07

comes from the culture and mentality

52:10

of the business that if you're the

52:12

kind of organization that says we want

52:14

to win by treating people

52:16

well and by allowing that

52:18

to elevate them and help them

52:21

create something even more wonderful,

52:23

then you're going to start treating

52:26

trust and safety more as

52:28

a core component of your business.

52:30

Yeah. And, you know, I think it's interesting,

52:32

this comes out in the blog post that you wrote,

52:35

that there are some things that are very specific

52:37

to the, voice space

52:39

and the work that you do. And you know,

52:41

it's funny cause I'm, I read your blog post

52:44

on, building versus buying and I'm just

52:46

like, All of the things that you

52:48

have to do to do voice

52:51

moderation. Well, it seems horrifying

52:54

to me, right? It's just like this huge

52:56

list of things. And so not that

52:58

I'm in the space where I need voice moderation,

53:00

but like, for me, it's an easy call. There's

53:02

no way I want to build, so

53:05

do you want to talk a little bit about some of the specifics

53:07

that you sort of discussed in that blog post about, what

53:10

companies should be thinking about specifically

53:12

in the voice moderation space.

53:14

sure. Yeah. I, I think broader

53:16

than voice moderation, it really comes

53:18

down the builder by decision to three major

53:21

components. There's the

53:23

data. What decisions

53:25

do you want to make? How can you make the best decisions?

53:28

There's the technical capability to actually,

53:30

build those tools and have them function. And

53:33

then there's sort of the user interface. What's

53:35

the experience of this? Do we know how to make that

53:37

efficient? In voice, each

53:39

of those things kind of takes on an extra level,

53:42

right? In order to get good data, you

53:44

You first of all need it to be voice

53:46

native data. If you're just looking at transcripts,

53:49

we've worked with some platforms that took

53:51

text transcripts and then sent

53:53

them to, you know, Amazon Mechanical Turk

53:55

or something and had someone try to read the transcript,

53:58

you're not getting the emotion, you're not getting

54:00

any of what we're actually looking

54:02

for in that voice to really understand

54:04

that. So if you want to build a proper

54:07

voice moderation tool, you need

54:09

data that's native to.

54:11

Real conversations between

54:13

people. This is also where I think

54:16

you get a benefit from working with an

54:18

external partner, because we

54:20

see what's happening across

54:22

so many different platforms. We

54:25

can see emerging new trends that haven't

54:27

reached your ecosystem yet. We

54:29

can get much broader data sets to

54:31

be more robust to different kinds of accents

54:33

and languages. We just have

54:35

a lot more ability to do

54:37

that kind of stuff because of the cross section

54:40

we see. The technical sort of

54:42

capability. I don't think I need

54:44

to spell out too much there. Processing voice

54:46

is really complicated. You need a specialized

54:48

team on that. but just on the user

54:50

interface piece, I'll call out for

54:52

voice and for a lot of content moderation,

54:55

but it really becomes extra true in voice.

54:57

Moderator mental health. is a consideration

55:00

you have to make, right? if you're not

55:02

just seeing this awful stuff, but hearing

55:04

it over and over again over the course of

55:06

a day, that can be really brutal, and

55:08

it can actually lead to physical issues, such

55:11

as tinnitus. So being able

55:13

to actually process that audio

55:15

and make sure that if people are

55:17

listening to it, they're doing it as little

55:19

as possible. You're pre processing

55:21

it so it's not causing physical damage.

55:23

So, That kind of extra stuff

55:25

is also easily missed when someone

55:28

is just thinking, oh, how can I, you know,

55:30

spin something up as quickly as possible here?

55:33

Yeah. I think that's, that's interesting. There's, I mean,

55:35

there's a few things that are obviously specific to

55:37

the voice world that I just never even would have thought

55:39

of, as well. Um, it's kind of interesting.

55:41

You know, a lot of this discussion is really, and

55:43

obviously, as you mentioned, you're in the vendor

55:46

space. And so this is an existential

55:48

question for you, but I'm sort of curious.

55:50

not as a devil's advocate kind of question,

55:52

but in which situations do

55:54

you think it actually does make sense for companies

55:57

build, build their own

55:58

I think the, biggest reasons are sort

56:00

of where they get an edge on one of

56:02

those dimensions. So, uh,

56:05

On the data dimension, if you have a really

56:07

unique kind of experience,

56:09

or you have some way that people are interacting

56:11

that really is not reflected

56:14

in other situations, maybe

56:16

it makes more sense for you to say, Hey, we need

56:18

to train on our own data because that has

56:21

truly nothing to do with anything else. I'm honestly

56:23

struggling to come up with a good example of

56:25

that, but I've, I've looked at some very

56:27

niche apps of, you know, this is specifically.

56:30

Chess camp only for like four

56:32

to six year olds, which is a very different

56:34

vibe of how to have a conversation. Maybe

56:37

something very niche like that. I think

56:39

it's more commonly true that it's a

56:41

technical advantage. that's not

56:44

true for things like text and voice that

56:46

are standardized, images

56:48

as well, but UGC,

56:50

obviously coming from gaming, this kind of moderation

56:53

is a big topic that we hear about a lot,

56:55

and usually when you're thinking about what are people building

56:58

in my game, You can't just print

57:00

out an image of it and say,

57:02

Oh, run that through the Is It Genitalia

57:04

filter. You have to look at it from different

57:07

angles. You have to think about how things can compose

57:09

together. You have to think about are they actually

57:11

writing something in the game

57:13

universe? Or all of

57:16

those factors are going to be very

57:18

specific to your ecosystem

57:20

and require access to

57:22

your ecosystem at a deeper level

57:24

than just grabbing the text or grabbing

57:27

the audio clips of what's being sort of

57:29

said. So I think those are the

57:31

situations where it makes more sense to try

57:33

to build in house when just you have access

57:35

to that in a deeper way than an

57:37

external partner would be able to.

57:40

very interesting. One sort of final question

57:42

here. And this, this is something that has come

57:44

up when I've had this discussion with people as

57:46

well, in terms of thinking about it is

57:48

one of the fears that some

57:50

people have around doing

57:53

the outsourced by decision

57:56

is that it leads to sort of Homogenization

57:59

of the tools that if everybody's using

58:01

sort of the same tool, then

58:03

it, gets stuck with just, you

58:05

know, one or maybe two, two or three

58:07

players in the space. Do you have any thoughts

58:09

on, on that or how to think

58:11

through that issue? Right.

58:15

I really buy it to be honest. Um,

58:17

you know, If you're building

58:19

one thing internally, that

58:22

itself is an echo chamber.

58:25

If you're pulling from external

58:27

tools, if there was only one external

58:30

tool that everyone was using, yes,

58:32

I would have some concerns. Now, of course, there's

58:34

configuration options and all that, but I would still

58:36

have some concerns. But we're so

58:38

amazingly far from that world.

58:40

We are very much in a competitive ecosystem

58:43

where there's a lot of different tools out there,

58:45

bringing different approaches, different

58:48

mentalities, handling different

58:50

kinds of harms, some focused

58:52

on the financial stuff, some focused on interpersonal.

58:55

I don't think we're in any risk

58:58

of everything calcifying into

59:00

the exact same environment there.

59:02

And I think by. Having more

59:04

platforms talking openly to

59:07

folks like vendors that do sort

59:09

of operate in that in between

59:11

space. That actually

59:13

sort of up levels the whole conversation

59:15

we're having as a collective industry because

59:18

now we're seeing all those different

59:20

perspectives being gathered together

59:22

in one place.

59:23

Cool. Very, very interesting. Yeah. This

59:26

is such a fascinating topic. And

59:28

I think it's something that a lot of people are thinking about.

59:30

So, I'm glad you wrote the blog post.

59:32

I'm glad you're coming on the podcast to

59:34

talk about it as well. And I'm sure there will be many

59:36

more conversations along this line but,

59:38

thanks for joining us today.

59:40

Yeah, thanks again for having me. Always happy to chat.

59:45

Thanks for listening to Ctrl-Alt-Speech.

59:48

Subscribe now to get our weekly episodes

59:50

as soon as they're released. If your

59:52

company or organization is interested in sponsoring

59:55

the podcast, contact us by visiting

59:57

ctrlaltspeech.Com. That's

59:59

C T R L Alt Speech. com.

1:00:02

This podcast is produced with financial support

1:00:05

from the Future of Online Trust and Safety Fund,

1:00:07

a fiscally sponsored multi donor fund

1:00:09

at Global Impact that supports charitable

1:00:11

activities to build a more robust, capable,

1:00:13

and inclusive trust and safety ecosystem.