0:02

Hello, everybody. Welcome to the

0:04

Radical Caner Podcast. I'm Kim

0:06

Scott. And I'm Jason

0:08

Rozov. And today we're going to

0:10

be discussing a philosophy that's become

0:12

almost mythical in tech circles and that

0:14

we're watching play out in real

0:16

time in the US government. This

0:19

philosophy is often referred

0:21

to as the move fast

0:23

and break things approach to

0:25

work. And when this approach is effective

0:27

and when it might lead us astray. So

0:30

with that, let's get going. I have

0:32

to say, Jason, that one of

0:34

my favorite photos, and we will put

0:36

it in the show notes, comes

0:38

from a friend of mine who used

0:41

to work at Facebook, which is

0:43

the company that coined this, Move Fast

0:45

and Break Things. And the

0:47

new sign was, slow down and fix

0:49

your shit. So

0:53

Move Fast and Break Things

0:55

is, in my book, sort

0:57

of obnoxiously aggressive. However, I

0:59

will say there is at Google,

1:01

we called it launch and

1:03

iterate. Yeah. And

1:05

that seems to me to

1:07

be a better way

1:10

to say what the good

1:12

part of this, which is that if

1:14

you're so afraid of never making a

1:16

mistake, you can't innovate, you

1:19

can't fix things. It's sort of

1:21

the ethos behind whoopsie daisy,

1:23

which is publicly saying, oh,

1:25

I messed that one

1:27

up. And I'm going to

1:29

do better next time. I mean,

1:31

another way to say the good

1:33

part of this is something that

1:35

is on the bottom of a

1:37

friend of mine's every email says

1:39

make new mistakes. So

1:42

I think that

1:44

it needs to be

1:46

OK to make a mistake.

1:48

You can't innovate if it's not OK to

1:50

make a mistake. And it even

1:52

needs to be okay to

1:54

admit mistakes in situations where it's

1:56

really not okay to make

1:59

mistakes like hospitals. This is

2:01

kind of what's

2:03

behind Amy Edmondson and

2:05

her book, The

2:07

Fearless Organization. A lot

2:09

of the research she did behind

2:11

psychological safety is if you can't

2:13

admit a mistake, if you can't

2:15

share a mistake, then you're going

2:17

to make your dooms to make

2:20

it over and over and over

2:22

again. That's right, yeah. And paradoxically,

2:24

the organizations that were the hospitals

2:26

where the most mistakes were reported

2:28

were also the safest hospitals, which

2:30

was not what you expected. So

2:32

that's my blink response. When

2:34

I hear move fast and

2:36

break things, especially in the

2:38

context of firing lots of

2:40

people, it feels evil to

2:43

me. Yeah. There

2:45

was a very concrete example of

2:47

this, which was there was

2:49

an excerpt which we'll try to

2:51

find for the show notes

2:53

of one of the cabinet meetings

2:55

where Elon Musk is talking

2:57

about making mistakes. And first of

2:59

all, what's he doing in

3:01

a cabinet meeting? He's an unelected

3:03

bureaucrat. The

3:06

fact that he was in the

3:08

room is an example of moving

3:10

fast and breaking things when they

3:12

should be slowing down and fixing

3:14

their shit. Yes. But he

3:16

said, He

3:19

said jokingly that they had

3:21

He said we're gonna make mistakes

3:23

like for example we cut

3:25

all the funding to Ebola research

3:27

and then we realize whoops

3:29

We probably should have kept the

3:31

funding for Ebola research so

3:33

we turn that funding back on

3:35

and It was like it

3:37

in the moment that clip makes

3:39

it seem sort of reasonable,

3:42

but it doesn't capture the whole

3:44

picture because what part

3:46

of what's happening is that

3:48

The uncertainty of grant money

3:50

is is unwinding programs So

3:52

it's not just like the

3:54

absence Like it's not that

3:56

a short -term absence of

3:58

funding Has no effect. It's

4:00

in fact the case that

4:03

in some cases they are

4:05

so dependent on the next

4:07

distribution of money that they

4:09

literally had to stop an

4:11

experiment midstream. And so, if there

4:13

were live samples of something,

4:15

for example, they couldn't afford to

4:17

keep those frozen, so those

4:19

samples went bad. There's

4:21

a real effect. There's

4:24

a real cost to that mistake. Correct.

4:27

And so, on

4:29

the one hand, credit

4:31

where credit's due, realizing that was

4:33

a terribly... stupid decision and it needed

4:35

to be undone. And on the

4:38

other hand, the one

4:40

of the things that I would like

4:42

to add to your list of what, you

4:44

know, things to look

4:46

out for when you're considering

4:48

a sort of ship and iterate

4:50

is what is the cost?

4:52

What are the side effects of

4:55

negative outcomes? Yes. Because if

4:57

the side effects are significant or

4:59

potentially very costly, then

5:02

the it's

5:04

much more prudent to slow

5:06

down and make sure you're making

5:08

the right decision. Yeah, exactly.

5:10

I mean, launch and iterate was

5:12

about search results. It

5:14

did not apply to

5:16

a nuclear power plant,

5:18

for example. You

5:21

would not want to launch

5:23

and iterate at a nuclear power

5:25

plant. You want to make

5:27

damn sure that you're not going

5:29

to You know blow up

5:32

a major metropolitan area or any

5:34

area if you if you

5:36

launch you you got a test

5:38

and test and test and

5:40

even when the even when the

5:42

consequences of failure are not

5:44

that dire but like Apple was

5:47

not a launch and iterate

5:49

kind of culture because they were

5:51

making Hardware it wasn't so

5:53

you couldn't just push a fix

5:55

out and and once you

5:57

have sold this phone to someone

5:59

you You can't really fix

6:02

it very easily, very cheaply anyway.

6:04

It's not just a matter

6:06

of pushing a software patch out.

6:09

And so they tested it and

6:12

tested it. Apple was much

6:14

more of a measure 100 times

6:16

cut once kind of culture, which

6:19

is not to say that

6:21

it wasn't innovative. Apple's obviously incredibly

6:23

innovative. So you don't even

6:25

have to launch and iterate. to

6:28

be innovative and you certainly

6:30

don't have to move fast and

6:32

break things. Go

6:36

ahead, you were going to say something. No,

6:39

I think the

6:41

cost of failure

6:43

calculation, I think

6:45

you're right. There

6:48

are some times

6:50

in human history

6:52

where, in

6:54

fact, We

6:59

took great risks with people's lives

7:01

because the cost of failure was even

7:03

higher. I think it's important to

7:05

put everything on a spectrum. When

7:08

we were testing aircraft, for example,

7:10

there are all these people who

7:12

put their lives at risk to

7:14

test this aircraft. Even though they

7:17

were testing to try to get

7:19

it right, the cost

7:21

of failure was still high,

7:23

like test pilots died in

7:25

testing these aircraft. But

7:28

the work that they did wind

7:30

up putting us in a position

7:32

to be able to turn the

7:34

tide of World War II. So

7:36

the potential for success was great.

7:38

But the thing that they did

7:40

really well was they were very

7:42

clear headed about the cost of

7:44

failure. Meaning they understood that they

7:46

were putting people's lives at risk.

7:49

And they said, we're measuring that

7:51

against the good that we think

7:53

this can do. And it's not

7:55

that when people do that, they

7:57

always get it right. But I

7:59

think that's a pretty big difference

8:01

than the sort of like, har

8:03

har we, you know, we undid

8:05

Ebola funding. You know what I'm

8:07

saying? Like the joke. Right.

8:09

Exactly. Because he's not going

8:11

to bear that cost. Musk

8:14

is not going to. And there's

8:16

and part of the problem

8:18

with what's happening is that, you

8:20

know, part of the American

8:22

experiment was checks and balances, right?

8:24

Yeah. And so we were

8:26

putting checks and balances in place.

8:29

Our governmental system put checks and

8:31

balances in place so that

8:33

the president didn't have too much

8:36

power. The Congress didn't have

8:38

too much power. The judiciary didn't

8:40

have too much power. They

8:42

could check each other's power. And

8:45

unfortunately, those checks and balances

8:47

are being undone before our

8:49

very eyes. And then

8:51

it becomes impossible to

8:53

hold people accountable for part

8:55

of the cost of

8:58

failure means that if you

9:00

fail, there should be

9:02

some accountability for failure. And

9:06

in this case, there's

9:08

not. That's the problem of

9:10

having an unelected bureaucrat

9:12

in these cabinet meetings. One

9:15

of many.

9:18

I think the other thing that's

9:20

on my mind is just

9:22

to bring back a topic we've

9:24

talked about many times, which

9:26

is one of the issues with

9:29

move fast and break things,

9:31

even if you're aware of the

9:33

potential cost, the

9:35

potential negative impact of

9:37

failure, is that

9:39

it's often quite hard

9:42

to measure the full cost

9:44

of failure. And so

9:46

if your attitude is like,

9:49

there isn't really, there doesn't need to be

9:51

accountability for failure. You know, if you

9:53

have this like sort of careless attitude. Yes,

9:57

like careless people. Yes,

9:59

you wind up, you wind up being

10:01

much more susceptible to the measurement

10:03

problem, which is that it is very

10:05

hard to measure the things that

10:07

really matter when you're doing these calculations.

10:10

And so it actually takes it

10:12

does take slowing down to really

10:14

consider what the externalities of what

10:16

we're doing actually are for you

10:18

to realize, oh, it

10:21

may not be so simple as

10:23

just turning the funding back on. We

10:26

may be setting ourselves back if

10:28

we do this in this way. I

10:30

think there's also size matters. I

10:32

think a lot of the launch and

10:34

iterate and even the move fast

10:36

and break things kind of culture, Came

10:39

from a world in which

10:41

these companies were small right

10:43

right when they started saying

10:45

launch and iterate like it

10:47

didn't really matter if Facebook

10:49

made a mistake in the

10:51

early days because Facebook didn't

10:53

really matter in the early

10:55

days and now as we

10:58

have seen it's having a

11:00

huge impact on the psychological

11:02

well -being of all of

11:04

us polarization You know and

11:07

Myanmar genocide

11:09

was planned on

11:11

Facebook. And

11:14

it really does matter.

11:16

Mistakes really have real

11:18

world, terrible human consequences.

11:21

And I've been

11:24

thinking, I've been

11:26

doing a fair amount of

11:28

retrospection about my career

11:30

in tech and what has

11:32

gone wrong. in

11:34

recent years. And I think,

11:37

in fact, I may even have to

11:39

go back to the first job

11:41

I had out of business goals at

11:43

the FCC in 1996. And

11:45

this was when the

11:47

Telecom Act came out and

11:49

Section 230 is a

11:52

section of the Telecom Act

11:54

of 1996. Section

11:56

230 explicitly lets social

11:58

media platforms, lets tech

12:00

platforms off the hook

12:02

for For any content

12:05

that is for any accountability for

12:07

content moderation and in 1996 I

12:09

mean even then it was sort

12:11

of questionable writing this blank check

12:13

to this new industry But I

12:15

could you know, I I understand

12:18

what we were thinking at the

12:20

time We were thinking it doesn't

12:22

make sense to regulate a tech

12:24

a company companies that don't even

12:26

exist yet. I mean in 1996

12:28

Google hadn't even been founded

12:31

and I think Zuckerberg was in like

12:33

third grade or something. And

12:35

so these platforms

12:38

didn't exist at

12:40

the time. And

12:42

so there was some sense

12:45

to it. But I wish

12:47

that we hadn't, in retrospect,

12:50

written this blank check for

12:52

all time because now this

12:54

check is coming due and

12:56

we can't afford to pay

12:58

it. And so at this

13:00

point, it is time to

13:02

hold these companies accountable for

13:04

the harm that the content

13:06

on these platforms does. And

13:08

I think also you raised one of

13:10

my favorite topics of all time, the

13:12

measurement problem. Part

13:15

of the issue is

13:17

that if you

13:19

just are measuring engagement,

13:22

which is what? Sort of

13:24

which is all they

13:26

really measure. I shouldn't say

13:28

that's all they measure

13:30

But that's a that's very

13:32

important to Facebook's business

13:34

is engagement because the more

13:36

engaged users are the

13:38

more likely people are to

13:40

contribute content to Facebook

13:42

and to Instagram and to

13:44

all these other and

13:46

and and and and people

13:48

tend to engage more

13:50

with negative content with negative

13:52

emotions, which is why

13:54

these platforms tend to sort

13:56

of spew more BOMO,

13:58

fear of missing out, more

14:01

deeply enraging content that

14:03

polarizes us all, more content

14:05

about eating disorders, all

14:07

of these kinds of things,

14:09

because we pay attention

14:11

to these things. And so

14:13

they're just kind of

14:15

tracking what they can measure.

14:17

very hard to measure

14:19

the value of a well

14:21

-functioning society. Believe

14:24

me, it doesn't factor in

14:26

clearly. Even

14:28

though in the end, if

14:31

society dissolves, it'll kill the

14:33

goose that lay the golden

14:35

egg. It seems like you

14:37

would factor it in. I

14:40

spent many of the formative

14:42

years of my career in Silicon

14:44

Valley working at a nonprofit.

14:46

On the other side, at Khan

14:48

Academy. At Khan Academy, yeah.

14:51

Yeah. And I will say, we

14:53

thought a lot about

14:56

these externalities. What does

14:58

it mean to give this away? Can

15:00

I interrupt? Can you explain to people who

15:02

don't know what an externality is? Oh,

15:04

sure. Yeah. But just a side

15:06

effect is maybe a simpler way

15:08

to say it. The side effects of

15:10

what we were doing. So

15:12

Khan Academy was providing

15:14

freeze or world -class

15:17

educational content. And

15:20

we were mostly fine with that

15:22

because most of the other people who

15:24

were providing content were educational publishers. And

15:27

we didn't think the educational publishers were

15:29

doing a great job of providing great

15:31

content for teachers and students to use.

15:34

But we were conscious of

15:36

the fact that there

15:38

may be some side effects.

15:41

So for

15:43

example, As much

15:45

as we might not like to

15:47

think about it this way, a

15:49

textbook you can use for many

15:52

years and many students can use

15:54

a textbook and you don't need

15:56

a computer and you don't need

15:58

internet access to use a textbook.

16:02

We were conscious of the fact

16:04

that if we were successful

16:06

and we undermined this other way

16:08

of getting access to stuff,

16:10

it had the potential of creating

16:13

of exacerbating inequities in the

16:15

system. Even though

16:17

it would be done for all the best

16:19

reasons, and there wasn't

16:21

a profit motive, so we were giving

16:23

away for free, but there were these

16:25

potential side effects of what we were doing, and

16:28

we thought about that a lot.

16:30

And what that caused us to do

16:32

was to build partnerships with people

16:34

who were bringing free or very inexpensive

16:36

internet access, free or very inexpensive

16:38

computing. Two schools in the US

16:40

and around the world like we

16:42

thought about like how do we make

16:44

sure that if we. If

16:47

we're successful that we don't wind

16:49

up just sort of you know

16:51

giving really great access to stuff

16:53

to rich kids who already had

16:55

all the things that they already

16:57

needed. And I

16:59

and I think that. How

17:01

much time do you think Facebook

17:04

spends worrying about that? I

17:06

mean I'm sure there are people I'm

17:08

sure there are but let me pause

17:10

I'm sure there are people at Facebook

17:12

who care deeply deeply about this like

17:14

I do not mean to dismiss everyone

17:16

who works at Facebook I have friends

17:18

who work at Facebook or meta So

17:22

it certainly don't mean to

17:24

cast dispersion on all these people.

17:26

But if you're making that

17:28

argument, and meanwhile these other metrics

17:30

that are making all the

17:32

money, the argument that

17:35

this thing that you care passionately about

17:37

is likely to get lost in

17:39

the noise. Yes. And I

17:41

was going to say the

17:43

exact same thing. I think,

17:45

in part, there's a

17:47

benefit of... You know,

17:49

the Khan Academy relied on

17:51

donations, so that limited

17:53

our scope and our ability

17:55

to grow. And

17:58

that meant it was easier to keep

18:00

people aligned around the mission and it

18:02

was easier to sort of bubble up

18:04

or center these conversations around making sure

18:06

that we were actually achieving the mission

18:08

and not just sort of reinforcing inequities

18:10

that were already there. And

18:13

so I do imagine that there

18:15

are some people at Facebook who are

18:17

every day having a discussion about

18:19

how can they help to fix some

18:21

of the problems that social media

18:23

as a medium has created and all

18:25

this other stuff. As

18:27

there are at Google, this great

18:29

program Grow with Google where

18:31

Google is trying to offer content

18:33

to people to learn skills

18:35

that will help them get jobs.

18:38

Right. It exists

18:40

everywhere. It's sort

18:42

of not at the core of

18:45

the machinery. It exists on

18:47

the periphery as these

18:49

organizations grow. I

18:52

think that back to the

18:54

measurement problem, what's measured is

18:56

managed. If short -term

18:58

profits are measured, if

19:01

engagement is measured, those

19:03

are the things that actually wind

19:05

up being managed against. Yeah,

19:07

I think you're totally right that the

19:09

long -term play is a bad one, which

19:11

is like if it actually undermines society and

19:14

like the health of people become so. Disgusted

19:17

with it that they, you know, they

19:19

leave these platforms, then they have no users

19:21

left so they can't make any more

19:23

money. Yeah, but. Again,

19:26

I think a consideration of

19:28

the periphery as opposed to like at the

19:30

core. Yeah, well, I mean, I

19:32

think part of the part of the

19:34

problem is. to be fair

19:36

is not only the

19:38

metrics that drive Facebook's business,

19:40

but also the market.

19:42

If the market rewards quarterly

19:44

earnings, it's really hard to

19:47

worry about the downstream impact

19:49

of your product. Although

19:51

it's not impossible. Again,

19:53

I don't mean... for

19:56

this to be a

19:58

Facebook bashing Google promoting podcast.

20:01

But like in the S1 letter, when I

20:03

took the job at Google in 2004,

20:05

and you can feel free to push back

20:07

and tell me I'm being, you

20:09

know, I had drunk the Kool -Aidems. But, you

20:11

know, Larry and Tergay said, we are

20:14

not a normal company. We don't intend to

20:16

become one. And we're going to invest

20:18

a lot of money in things that are

20:20

not going to make short term return

20:22

for shareholders. And if you're not comfortable

20:24

with that, you know, don't don't buy our

20:26

stock. You know, and we're

20:28

going to continue to reward and treat our

20:30

people well. And if you're not, if

20:32

you're not happy with that, don't, you know,

20:34

don't buy the stock. And that to me

20:36

was really important. And it

20:38

had, so let's talk about

20:40

sort of the importance of

20:42

debate. Like another problem of move

20:44

fast and break things is

20:46

that there's no time for discussion.

20:48

And in order to in

20:51

order to really innovate you need to

20:53

create time and space for discussion. Yeah, everything

20:55

we're just talking about is this is

20:57

exactly was on my mind like all there

20:59

are people who want to have these

21:01

debates who have really good arguments for why,

21:04

you know, should or shouldn't

21:06

do something and and

21:08

to your point like.

21:11

I don't think it's just the attitude like

21:13

to some degree. It's

21:15

also size, it makes a

21:17

difference here. Like the larger

21:19

the organization gets, the harder

21:21

it is to have debates

21:23

with the people who really

21:26

matter, like whose arguments are

21:28

going to move the needle. Maybe it's a better

21:30

way to say it. It's not about the

21:32

people who matter, but it's about the arguments that

21:34

they're able to make. And so as a

21:36

result, you wind up with, and

21:38

I think this was some of what

21:40

we, What was in careless

21:42

people is like you wind up with

21:44

these sort of like echo echo

21:46

chambers like tiny echo chambers inside the

21:48

company where there's like reinforcement of

21:50

bad behavior In part because they're not

21:52

listening to you know what I'm

21:55

saying there like they're not inviting the

21:57

Disagreement they're not inviting the debate

21:59

and for folks who are not familiar

22:01

with careless people This is a

22:03

book you want to talk about careless

22:05

people for a second Oh, no,

22:07

you go ahead. You got it. It's

22:09

a book. It's

22:11

sort of a memoir

22:13

written by a

22:15

former Facebook employee. And

22:18

she really describes in

22:21

great detail some of

22:23

the problems with the

22:25

way the systems worked

22:27

and the Again,

22:30

the negative externalities, the negative

22:33

impact on all of us, the

22:35

negative side effects that all

22:37

of us are bearing the cost

22:39

of these negative externalities that

22:41

are created by the way that

22:43

Facebook system works. And

22:46

Meta sort of prevented, you

22:48

know, sued the author and

22:50

prevented her from talking about

22:52

her book. So we're trying

22:54

to talk about her book

22:57

for her. And

22:59

by the way, it also prompted

23:01

me to read Francis Hogan's book,

23:03

The Power of One, which

23:05

I hadn't read before is

23:08

also really a great explanation

23:10

of how these systems work

23:12

and what we could do

23:14

to make them not create

23:16

these terrible negative side effects

23:18

for society. And begs a

23:20

lot of questions like why Facebook

23:22

isn't already doing these things. And I

23:24

think part of the answer is

23:26

that there's no public debate. about

23:28

how their algorithm should

23:30

work. And there's no sense

23:32

that there should be

23:34

a public debate about that.

23:37

And I think even to some

23:39

extent, there's very limited internal

23:42

debate. Debate, yes. And

23:44

so if

23:46

you're a

23:48

free market

23:50

capitalist believer,

23:53

in theory, the idea is like

23:55

is that people can have

23:57

an opinion about this and they

23:59

can vote with their money.

24:01

They can basically say, I'm not

24:03

going to give my money.

24:05

But that, to some extent, is

24:07

undone by the business models

24:09

that they've created, which is essentially

24:12

monetizing. They don't

24:14

charge us to access

24:16

the content. They

24:18

charge advertisers, essentially, to

24:20

support the platforms. And

24:23

so it's very hard. Right

24:25

before we got on the podcast,

24:27

we were talking about, you know, should

24:29

we continue to post on Facebook?

24:31

Should we continue to contribute content to

24:33

Facebook and Instagram? And

24:36

I think part of

24:38

what's interesting about that

24:40

debate is like, it's

24:42

hard to affect a

24:45

real protest when the

24:47

consequences are fairly delayed

24:49

for a company like

24:51

Facebook. One

24:53

massive, you know, movements

24:56

of people away from

24:58

the platform to start

25:00

to really have a

25:02

noticeable negative impact on

25:04

their revenue. And that's

25:06

different, right? Like, you

25:08

know, when Tylenol had to

25:10

recall, this is like in the

25:12

70s or something, this thing,

25:15

people stopped buying Tylenol, like

25:17

the money dried up,

25:19

like it went away fairly

25:21

immediately. So there was

25:23

actually like a market response.

25:25

There was like a consequence. Yeah,

25:28

although if everybody, let's

25:30

say if even 10 % of

25:33

the people who contribute, I

25:35

mean, most Facebook users are readers,

25:37

not writers. That's right. If

25:39

only 10 % of the people,

25:41

and we're writers on the thing,

25:43

we're giving away our content

25:45

to this platform. If

25:47

only 10 % of the people

25:49

who actually post to Facebook

25:51

quit posting, that would

25:54

have a huge and

25:56

very quick problem. In fact,

25:58

this is the thing.

26:00

This was at two AHAs

26:02

when I read these

26:04

two books, Careless People and

26:06

the Power of One.

26:08

One AHA was that part

26:10

of the reason why

26:13

Facebook started advocating, Francis Hogan

26:15

explains this, started sort

26:17

of pushing more polarizing content

26:19

in their platform is that

26:21

it got more engagement. People

26:24

who are contributing content were

26:26

slowing down. They weren't contributing

26:28

as much content because they

26:30

weren't getting so many reactions

26:32

from people, lights and whatnot.

26:36

They found that when

26:38

they promoted in their

26:40

algorithm, more extreme content it

26:42

got more engagement and then people started

26:44

and it was really a vicious

26:46

cycle because then even if you didn't

26:48

believe these extreme things or if

26:50

you were writing headlines for example you

26:52

started writing these clickbait headlines and

26:54

so that's a real that that was

26:56

one real problem the other thing

26:59

and maybe this is just my own

27:01

stupidity that I didn't know I've

27:03

always sort of wondered like why do

27:05

they allow all these political ads

27:07

like it's not that it's not that

27:09

much money. Why don't they just

27:11

disallow them? And what

27:13

I realized was they allow

27:16

them because now they are

27:18

king makers. Now they

27:20

can help you get elected

27:22

or unelected. And so no

27:24

official dares regulate Facebook because

27:26

Facebook can prevent them from

27:29

getting elected. So this is

27:31

a huge dampening on democratic

27:33

debate about the ways that

27:35

we should regulate um the

27:38

content on on on facebook

27:40

flash instagram slash meta Yeah,

27:42

and that was like i

27:44

don't know why i didn't

27:46

realize that before but i'm

27:48

like oh of course you

27:51

know uh so that i

27:53

think is is important to

27:55

think about Yeah,

28:04

so like there's multiple layers of which

28:06

we needed debate and and we didn't have

28:08

it like internally I'm sure there were

28:11

people who were like we should not be

28:13

doing this like you know The engagement

28:15

thing is is good in the in the

28:17

sense that we're getting more people to

28:19

write But it's bad in the sense that

28:21

with like content is worse and people

28:23

feel worse about it Like i'm sure there

28:25

are people making that argument inside of

28:28

this book as they were deciding to yes

28:30

of course to do this and then

28:32

there was like to your point the public

28:34

debate where like like the public having

28:36

an opinion about whether or not this is

28:38

uh or the more broader market having

28:40

an opinion about whether or not what they're

28:43

doing is good or should continue that

28:45

doesn't really happen or doesn't feel like it

28:47

it should happen um and then there's there's

28:50

also like the government level

28:52

debate. Yes, the public, the

28:54

democracy level debate. It

28:56

seems increasingly, it feels

28:58

to me anyway, like there's

29:00

a move to just

29:02

do away with, I don't

29:04

want to have the

29:06

debate at all. So let's

29:08

in, how about we

29:10

just arm democracy? And that

29:12

obviously is very worrisome. But

29:15

I want to go back to,

29:17

there was a key moment in

29:19

my career where somebody raised the

29:22

issue of the importance of having

29:24

sort of public debate about decisions

29:26

that this is when I was

29:28

working at Google. I

29:31

went, so I was

29:33

managing AdSense, the AdSense

29:35

online team. And

29:38

we, in

29:41

addition to, sort

29:44

of trying to grow the

29:46

business. I was also in

29:48

charge of policy enforcement and

29:50

creating policy. And

29:52

that meant that at the

29:55

same time that I

29:57

was trying to grow revenue,

29:59

I was also in

30:01

charge of terminating AdSense publishers

30:03

violated Google's policies. Got

30:05

it. And this goes back

30:07

to the measurement problem. I

30:10

don't think it would have mattered. In fact,

30:12

I can tell you for sure, it did not

30:14

matter how fast it grew, how

30:17

much money AdSense

30:19

made, did not specifically

30:21

impact my compensation. And

30:25

that may seem sort

30:27

of nutty to people

30:29

who was in charge

30:31

of sales and operations,

30:34

but there was an understanding at

30:36

Google that we couldn't If we

30:38

measured things that narrowly, we were

30:40

going to get the wrong kinds

30:43

of behaviors. And so

30:45

I was equally as

30:47

passionate about taking down

30:49

the bad sites, or

30:52

the sites that were violating policy, I

30:54

should call them bad sites, as

30:56

I was about growing revenue. And

30:58

I was really excited about

31:00

growing revenue, believe me. But

31:02

I really believed that

31:04

you weed your garden. And

31:06

if you allow your

31:08

garden to get overrun by

31:10

weeds, that's not good

31:13

for your garden long term.

31:15

And I don't understand why

31:17

that doesn't happen more at

31:20

Facebook, because it's possible to

31:22

create a system where people

31:24

are caring about both of

31:26

these things. So I

31:28

thought I was doing a great job. That's

31:31

the TLDR there. But

31:35

then I was invited to

31:37

speak at this class called

31:39

Liberation Technology. It was

31:41

a class taught at Stanford, and it was a

31:43

class taught by Josh Cohen, who is an old

31:45

friend of mine, a person I like a lot.

31:48

And I was describing to him

31:50

content moderation challenges and this

31:52

big debate I had had. And

31:55

we had had the debate

31:57

at Google. This was actually around

31:59

Blogger, which I also managed for

32:01

a while, the policy enforcement. And

32:04

there was somebody had

32:07

written something calling for

32:09

genocide, basically, kill all

32:11

the ex people. And

32:14

I shouldn't even say ex,

32:16

kill all the ABC people.

32:18

And I wanted to take

32:20

it down. I believed that

32:22

that kind of content had

32:24

no part in the AdSense

32:26

network. And I was just

32:28

going to pull it down.

32:31

But I didn't have unilateral decision -making

32:33

authority. This was something that had

32:35

to be discussed more broadly. So

32:37

I was in this big meeting

32:39

at one point, you know, Eric

32:41

Schmidt sort of agreed with me.

32:43

He said, you know, if we

32:46

had a dance hall and not

32:48

blogger, you know, we would not

32:50

rent it out to the KKK.

32:52

Like why would we allow this

32:54

kind of content? And

32:56

yet both Eric and

32:58

I got overruled by kind

33:01

of the free speech crowd

33:03

and in retrospect I think I

33:05

was right and I wish

33:07

I had fought harder but these

33:09

are hard questions because their

33:11

point of view was that you're

33:13

better off knowing who believes

33:15

these things and who's saying it

33:17

than not knowing and you

33:19

know forcing this kind of stuff

33:21

to go underground. And

33:23

so in the end, I

33:25

think, if memory serves, which it

33:28

often doesn't, the older I

33:30

get. But I think what

33:32

happened was we put a content

33:34

warning. We left the site

33:36

up, but we put a

33:38

content warning saying something like,

33:40

this is bad. And although

33:42

I'm sure that's not what it said. And

33:45

so I was talking about this

33:47

in the class and thinking that we

33:49

had a pretty good debate process. And

33:51

I'll never forget Josh looked at

33:53

me and he said, you are making

33:55

those decisions. And at first, I

33:58

was kind of insulted, you know,

34:00

I'm like, but what's wrong? What? Of

34:02

course I like, why am I

34:04

not? And he was like, you are

34:07

totally unqualified to make this. And,

34:09

and, you know, once I got

34:11

over feeling kind of offended, because

34:13

this was the most interesting part

34:15

of my job hands down, you

34:17

know, I really did care about

34:19

it. But he's. Josh said, and

34:21

now I think he's right, like,

34:23

these decisions, there needs to

34:26

be some democratic oversight for these

34:28

kinds of decisions because they have

34:30

such a huge impact on our

34:32

whole society. So this

34:34

is a big, Josh was right,

34:36

I was wrong. But at the time, I

34:39

was like, oh, Josh, you don't

34:41

understand, the government could never be

34:43

involved in these kinds of decisions.

34:45

You know, like, I have people,

34:47

we had a policy about no

34:49

porn sites. And some clever person

34:51

took a picture of himself in

34:53

front of a toaster. But

34:55

of course, the toaster was very shiny

34:57

and he was naked. And

35:00

it's complicated to manage

35:02

all the content. And

35:04

there was another moment

35:06

where there were these

35:08

ads for bestiality showing

35:11

up on a, this

35:13

is all a long

35:15

segue, but they're funny

35:17

stories. But there was

35:19

an ad for bestiality

35:21

that kept showing up

35:23

on this parenting magazine

35:25

that was an AdSense

35:27

publisher. And obviously they

35:29

were very upset about

35:31

this bestiality ad. And

35:34

I called up

35:36

the ad content moderation

35:38

team, which was

35:40

in another country, but

35:42

anyway, they They

35:44

were reviewing all the ads and I was

35:46

saying, why are these ads even showing up

35:48

anywhere? Like we have a policy against porn

35:50

ads. And this person

35:53

claimed that bestiality didn't count as

35:55

porn. I was like, gosh,

35:57

that was not on my bingo

35:59

card today. The argument about

36:01

whether. So anyway, I told these

36:03

stories and I. Yeah, they

36:05

all got a big laugh and

36:07

nobody really stopped to think

36:09

about should we have democratic oversight

36:11

over some of these decisions.

36:14

And I now believe we should.

36:16

Josh Cohen was right. I'm not

36:18

arguing against it, but I'll

36:20

just say like I think

36:23

that it's so much harder

36:25

in practice to achieve the

36:27

goal of oversight in a

36:29

way. Here

36:31

just to give your team credit for

36:33

a second like I think your team

36:35

probably thought about these things very seriously

36:37

and maybe even put like days or

36:39

hours of time and it went all

36:41

the way I mean, you know the

36:43

CEO of the company was willing to

36:45

spend his time on this like Google

36:47

took this very seriously right and I

36:49

guess like my experience with Like I

36:52

don't know exactly how you encourage that

36:54

kind of debate and at what level

36:56

and who gets involved outside of these

36:58

companies I'm like I'm open to the

37:00

idea, but I'm just recognizing a practical

37:02

challenge of the fact that everybody else

37:04

who would be contributing to that conversation

37:06

would either need to be paid by

37:08

Google or would be doing it on

37:10

a volunteer bait. You know what I'm

37:12

saying? They'd be volunteering their time to

37:14

do this thing. And that, I

37:16

think, is part of the reason why the

37:18

public debate hasn't been vigorous about this

37:20

stuff is because it's like, how do you

37:23

make the time to really understand what's

37:25

going on? And instead of public debate, what

37:27

we get is hot takes. You

37:29

know I'm saying? What we

37:31

get is like one person saying,

37:33

you know, something which is missing a

37:35

whole bunch of context, but it's

37:37

sort of punchy. And so as

37:39

a result, a bunch of people are like, yeah,

37:41

I agree with that person. And then the

37:43

other person sort of fires back and their thing

37:45

also misses a bunch of context. And as

37:47

a result, we're not really having a debate. We're

37:49

just sort of like throwing bars at each

37:51

other. The debate certainly should not happen on social

37:53

media. Yeah, for

37:55

sure it needs to happen in a different

37:57

in a different way And I think

37:59

I guess what I would say is like

38:01

what you're it seems to me what

38:04

you're describing is sort of the third pillar

38:06

of Avoiding the pitfalls of move fast

38:08

and break things which is like they're having

38:10

the right kind of culture now. Did

38:12

you get the systems were imperfect? There should

38:14

probably should have been some external involvement,

38:16

but I do think this idea that we're

38:20

going to have a public debate. We're

38:22

not going to give unilateral decision -making

38:24

power to you, to the person who's

38:27

technically in charge of this. We're going

38:29

to force a public debate on this.

38:31

And there's going to be a record

38:33

of that debate. We're actually going to

38:35

record this thinking for posterity. Those

38:37

are important cultural rituals that

38:39

I think don't exist in

38:42

a lot of organizations. And

38:45

going back to your original point, it's one

38:47

of the reasons why Without

38:49

those types of rituals without

38:51

like ritualizing debate for

38:53

example and saying this is

38:55

important part of how

38:57

we make decisions as an

38:59

organization Without removing unilateral

39:01

decision -making power for example

39:03

and setting that as a

39:05

cultural touchstone. I think

39:07

it's very easy for You

39:09

know the necessity of

39:11

the moment to overtake good

39:13

like good thinking It

39:17

feel like things feel urgent. And as

39:19

a result, people don't slow down to have

39:21

the debate. And I think more than

39:23

that, what you were saying at the very

39:25

top of the podcast was like, it

39:27

also without creating those

39:30

cultural norms, it also

39:32

discourages people from talking

39:34

about mistakes that they

39:36

make. And

39:38

when you combine those things together,

39:40

when you have no culture of

39:42

debate and no discussion of mistakes, It

39:45

becomes very easy to see how you

39:47

could go very deep down a rabbit

39:49

hole of bad things happening. And people

39:51

sort of looking around and like, whose

39:53

job is it to put the brakes

39:55

on this? What

39:57

is the way that

39:59

we respond collectively to

40:01

these bad decisions or

40:04

bad behavior that we're

40:06

observing? And I think

40:08

a lot of cultures

40:10

would benefit from... know

40:12

the removal of unilateral decision

40:14

-making power and a push

40:16

toward public You know as

40:18

public as you can make

40:20

it debate on important decisions.

40:22

Yeah, there's there is a

40:24

conference room at Metta right

40:26

next to Zuckerberg's office that

40:28

says Good news only or

40:30

only good news like that's

40:32

a disaster. That's an example

40:34

of The wrong kind of

40:36

culture like don't tell me

40:38

what's wrong, you know, like

40:40

You've got to have that

40:42

culture where leaders are soliciting

40:45

feedback and are eager to

40:47

hear the bad news and

40:49

that contrary point of view, not

40:51

the good news only

40:53

kind of culture. And

40:56

I think it's really important

40:58

that companies be willing, and

41:00

this is something that

41:02

tech companies have not traditionally

41:04

been willing to do,

41:06

to be held accountable by

41:08

the public, by the

41:10

government. There is a reason

41:13

why we have all

41:15

these NDAs and agreements that

41:17

don't allow you to

41:19

sue, that forced arbitration agreements.

41:21

That is an example

41:23

of a culture that is

41:25

trying to avoid being

41:27

held accountable by our government,

41:29

by the systems that

41:31

we have in place to

41:33

hold wealthy big companies

41:35

accountable. All right.

41:38

Well, let's let's let's try

41:40

to summarize our guidance here

41:42

for people who want to

41:44

be able to move fast

41:46

But don't want to break

41:49

things in an irreparably bad

41:51

way Whoopsie Daisy is one

41:53

thing like destroying democracy is

41:55

another All right, so tip

41:58

number one consider the cost

42:00

of failure don't skip the

42:02

debate phase and build a

42:04

team culture that supports

42:07

both speed and

42:09

learning. Yeah. Tip number

42:11

two, moving fast and breaking

42:13

things isn't inherently good or

42:15

bad. It's about applying a

42:17

thoughtful approach in the right

42:19

context with the right process.

42:21

So making sure that you

42:23

slow down enough to debate

42:25

something to make sure that

42:27

you're still on the right

42:29

track. And tip

42:31

number three. always focus on

42:33

learning. The whole point of

42:35

moving fast and potentially breaking

42:38

things is to learn fast. It's

42:40

not to create

42:43

a land grab

42:45

and establish a

42:47

monopoly that then destroys

42:49

democracy. And

42:51

with that... We invite you to head

42:54

over to radicalcander.com slash podcast to see the

42:56

show notes for this episode. Remember, praise

42:58

in public and criticize in private. If you

43:00

like what you hear, please rate review

43:02

and follow us on whatever platform you like

43:04

to listen to your podcasts. If

43:06

you have feedback, please

43:08

email it to us

43:11

at podcastradicalcander.com. Until next

43:13

time. And by the way, if

43:15

you have feedback on whether you think

43:17

we're going to invite some public

43:19

debate, do you think that we should

43:21

stop posting on Instagram and Facebook?

43:23

Let us know your thoughts. We are

43:26

eager for them. Thank you.