1
00:00:00,040 --> 00:00:07,520
Speaker 1: M hmm, what severing my tumors. I'm Robert Evans hosted

2
00:00:07,520 --> 00:00:10,680
Speaker 1: Behind the Bastards, the podcast where you tell you everything

3
00:00:10,720 --> 00:00:12,559
Speaker 1: you don't know about the very worst people in all

4
00:00:12,600 --> 00:00:15,600
Speaker 1: of history. Uh. Here with my guest Sophia, co host

5
00:00:15,640 --> 00:00:18,440
Speaker 1: of Private Parts Unknown, and we're talking about how it's

6
00:00:18,520 --> 00:00:20,680
Speaker 1: bullshit when doctors won't let you keep the pieces of

7
00:00:20,720 --> 00:00:23,480
Speaker 1: your body that they take out of you. That's really frustrating. Yeah,

8
00:00:23,520 --> 00:00:25,880
Speaker 1: that's like the least they could do for you. It's

9
00:00:25,880 --> 00:00:28,680
Speaker 1: an infringement of your civil liberties. Like that tumor or

10
00:00:28,680 --> 00:00:30,680
Speaker 1: whatever is still a piece of you, and you deserve

11
00:00:30,880 --> 00:00:33,560
Speaker 1: to like go get drunk on a farm and shoot

12
00:00:33,600 --> 00:00:37,000
Speaker 1: it with a shotgun if that is your choice. A. Yeah,

13
00:00:37,040 --> 00:00:38,920
Speaker 1: I wanted to just keep it forever to kind of

14
00:00:38,960 --> 00:00:41,519
Speaker 1: always like pointed to it and be like, yeah, I

15
00:00:41,720 --> 00:00:44,360
Speaker 1: beat you a little bit, and they won't let me

16
00:00:44,360 --> 00:00:46,480
Speaker 1: fucking do it. They wouldn't let me keep my my

17
00:00:46,600 --> 00:00:49,919
Speaker 1: breast cancer tumor and my chemo port, which I'm like,

18
00:00:50,320 --> 00:00:53,239
Speaker 1: that's that was part of me for a year. You

19
00:00:53,720 --> 00:00:58,480
Speaker 1: why that's so frustrating? Like, Okay, this this message is

20
00:00:58,520 --> 00:01:03,880
Speaker 1: going out to Sophia's doctor. Kudos on the cancer curing

21
00:01:04,040 --> 00:01:07,399
Speaker 1: blah blah blah, dick move not letting her keeper tumor.

22
00:01:07,680 --> 00:01:12,000
Speaker 1: And I'm I'm very angry about this. Uh, please write

23
00:01:12,000 --> 00:01:16,120
Speaker 1: a campaign, listeners, just contact my doctor at I'm just

24
00:01:16,640 --> 00:01:19,880
Speaker 1: We're gonna make not important to make Sophia's tumor in

25
00:01:19,920 --> 00:01:27,240
Speaker 1: her legal possession again. It's gonna be Hardia's tumor Sophia's again. Yeah,

26
00:01:27,280 --> 00:01:31,000
Speaker 1: there we go. Well, today's subject has nothing to do

27
00:01:31,080 --> 00:01:36,200
Speaker 1: with tumors cancer. Other than that, you could argue today's

28
00:01:36,200 --> 00:01:41,479
Speaker 1: subject is a cancerous tumor metastasizing in the body politic

29
00:01:41,720 --> 00:01:47,960
Speaker 1: of our nation. Wow, talking about YouTube, that's a beautiful

30
00:01:48,000 --> 00:01:52,040
Speaker 1: metaphor for a website that most people just use for

31
00:01:52,160 --> 00:01:57,200
Speaker 1: jerking off. Hey, jerking off and not paying for music. Oh,

32
00:01:57,280 --> 00:02:01,520
Speaker 1: that's true. That's true. There's one other thing, jerking on

33
00:02:01,680 --> 00:02:07,279
Speaker 1: YouTube tutorials. It's useful for jerking off, makeup tutorials, free music,

34
00:02:07,560 --> 00:02:11,840
Speaker 1: and of course filling the world with Nazis. Again, as

35
00:02:11,880 --> 00:02:15,399
Speaker 1: a Jew, I love to hear that. That's the aspect

36
00:02:15,400 --> 00:02:17,680
Speaker 1: of YouTube we will be talking about today is it's

37
00:02:17,800 --> 00:02:22,960
Speaker 1: it's Nazi reinvigorating aspects. Now, it's so fun to leave

38
00:02:23,360 --> 00:02:26,120
Speaker 1: the former USSR because it's not great for the Jews.

39
00:02:26,680 --> 00:02:29,440
Speaker 1: And then get here and then Donald Trump becomes president

40
00:02:29,440 --> 00:02:32,920
Speaker 1: and you're like, Okay, that's a great that's a good joke,

41
00:02:33,200 --> 00:02:36,880
Speaker 1: that's very funny. And then the Nazis spread through YouTube

42
00:02:36,919 --> 00:02:41,520
Speaker 1: so they're just everywhere, and you're like, Okay, well, I

43
00:02:41,560 --> 00:02:44,440
Speaker 1: guess I'll just live in fear forever. I will say

44
00:02:44,600 --> 00:02:47,120
Speaker 1: one of the best things that, like the few things

45
00:02:47,160 --> 00:02:49,600
Speaker 1: I actually got out of college was taking Holocaust studies

46
00:02:49,600 --> 00:02:52,079
Speaker 1: courses and coming to like the dawning realizations like a

47
00:02:52,160 --> 00:02:55,200
Speaker 1: kid who was raised and like a Republican household, where

48
00:02:55,200 --> 00:02:57,400
Speaker 1: like everything you heard about the Holocaust was how awesome

49
00:02:57,400 --> 00:03:01,200
Speaker 1: it was the American soldiers stopped it, like like reading

50
00:03:01,200 --> 00:03:04,560
Speaker 1: about history and coming to like the gradual realization like oh,

51
00:03:04,600 --> 00:03:09,680
Speaker 1: it's always sucked to be Jewish everywhere, like everyone's killed

52
00:03:09,720 --> 00:03:11,880
Speaker 1: these people, Like, oh my god, like it was he

53
00:03:11,880 --> 00:03:14,760
Speaker 1: didn't start with the Nazis, Like reading about like what

54
00:03:14,840 --> 00:03:18,079
Speaker 1: happened in the in Czarist Russia, the Nitsky massacre which

55
00:03:18,120 --> 00:03:23,040
Speaker 1: killed like seven thousand people, and like, yeah, this ship

56
00:03:23,160 --> 00:03:25,880
Speaker 1: has not been good for us for a long time.

57
00:03:26,480 --> 00:03:30,239
Speaker 1: And now we're talking about digital programs. Yeah, exactly. It's

58
00:03:30,280 --> 00:03:33,280
Speaker 1: just nice to know that you cannot escape the Nazis.

59
00:03:33,440 --> 00:03:37,000
Speaker 1: Yeah yeah, that is the message YouTube has delivered to

60
00:03:37,080 --> 00:03:39,360
Speaker 1: all of us, along with allowing me to listen to

61
00:03:39,520 --> 00:03:46,680
Speaker 1: old Chris Christofferson concerts for free. Um hey man, brotherfucker

62
00:03:46,960 --> 00:03:50,040
Speaker 1: made some great music. All right, I'm gonna start with

63
00:03:50,080 --> 00:03:56,160
Speaker 1: my prepared remarks, if that's okay please. March two, sixteen,

64
00:03:56,480 --> 00:04:00,440
Speaker 1: Microsoft unveiled a new chat bot to these serried denizens Twitter.

65
00:04:00,800 --> 00:04:05,720
Speaker 1: The bot was an experiment in what Microsoft called conversational understanding. TAY,

66
00:04:05,960 --> 00:04:08,680
Speaker 1: the chat bought would engage in discussions with real people

67
00:04:08,720 --> 00:04:11,559
Speaker 1: and learn from them, evolving and changing from its interactions

68
00:04:11,600 --> 00:04:15,720
Speaker 1: just like real people do. Yeah yeah yeah. As they

69
00:04:15,800 --> 00:04:18,680
Speaker 1: released Tay into the wild, Microsoft said they hoped that

70
00:04:18,680 --> 00:04:21,560
Speaker 1: Twitter users would be happy to engage it in casual

71
00:04:21,720 --> 00:04:25,360
Speaker 1: and playful conversation. Tay entered the world at around ten

72
00:04:25,440 --> 00:04:29,599
Speaker 1: am Eastern Standard time. At two am the following morning,

73
00:04:29,680 --> 00:04:33,640
Speaker 1: less than twenty four hours later, it tweeted this Bush

74
00:04:33,680 --> 00:04:35,560
Speaker 1: did nine eleven and Hitler would have done a better

75
00:04:35,640 --> 00:04:37,839
Speaker 1: job than the monkey we have now. Donald Trump is

76
00:04:37,839 --> 00:04:42,479
Speaker 1: the only hope we've got. It just took that little

77
00:04:42,520 --> 00:04:48,440
Speaker 1: time to learn, but you knew was a bad move.

78
00:04:48,440 --> 00:04:51,599
Speaker 1: When they opened it up, you knew it's kind of America.

79
00:04:52,680 --> 00:04:57,159
Speaker 1: One of the surprising stories or arcs of the last

80
00:04:57,279 --> 00:05:01,000
Speaker 1: decade is Microsoft going from like this evil corporation in

81
00:05:01,040 --> 00:05:05,120
Speaker 1: everyone's eyes, like this innocent summer child. Like they never

82
00:05:05,200 --> 00:05:07,640
Speaker 1: tried to steal my data, They never lobbied to stop

83
00:05:07,640 --> 00:05:09,960
Speaker 1: me from being able to repair my computer. They just

84
00:05:10,080 --> 00:05:12,560
Speaker 1: they believed they could make a chatbot and the Internet

85
00:05:12,600 --> 00:05:14,640
Speaker 1: would teach it how to be a real boy and

86
00:05:15,360 --> 00:05:19,600
Speaker 1: a Nazi. And they were so horrified, Like I just

87
00:05:19,839 --> 00:05:23,239
Speaker 1: I can't I can't believe that someone had positive hopes

88
00:05:23,279 --> 00:05:26,120
Speaker 1: for that. I mean, how few people have you met

89
00:05:26,200 --> 00:05:28,760
Speaker 1: in life online that you would think that that was

90
00:05:28,760 --> 00:05:31,840
Speaker 1: going to end up? Well, I think it's because Microsoft's

91
00:05:31,839 --> 00:05:34,040
Speaker 1: team were all old heads. Like it was a bunch

92
00:05:34,040 --> 00:05:37,080
Speaker 1: of guys in their fifties who, like I, didn't know

93
00:05:37,200 --> 00:05:40,040
Speaker 1: the Internet as anything but like a series of technical thing.

94
00:05:40,120 --> 00:05:43,080
Speaker 1: They didn't they weren't active Twitter users or whatever. They

95
00:05:43,080 --> 00:05:45,880
Speaker 1: didn't go on the Graham. Well, it's very quickly that

96
00:05:45,920 --> 00:05:47,960
Speaker 1: you learned that if you upload a video of yourself

97
00:05:47,960 --> 00:05:50,640
Speaker 1: doing stand up, how many you look like a kikes,

98
00:05:50,680 --> 00:05:55,760
Speaker 1: You're gonna get right away. I mean learning curve is yeah,

99
00:05:55,800 --> 00:05:58,200
Speaker 1: and it's it's a learning curve A lot of companies have.

100
00:05:58,360 --> 00:06:00,520
Speaker 1: I can remember back in twos and twelve when Mountain

101
00:06:00,560 --> 00:06:02,520
Speaker 1: Dew decided to let the internet vote on the name

102
00:06:02,560 --> 00:06:05,160
Speaker 1: of a new soda flavor, and four chanders flooded it

103
00:06:05,200 --> 00:06:07,560
Speaker 1: in before along. The top vote getter was Hitler did

104
00:06:07,600 --> 00:06:12,680
Speaker 1: nothing wrong, which is that, I will admit right off

105
00:06:12,720 --> 00:06:16,080
Speaker 1: the tongue, it would be kind of interesting to see

106
00:06:16,120 --> 00:06:21,560
Speaker 1: that soda marketed in the seven olive. But yeah, especially

107
00:06:21,560 --> 00:06:25,120
Speaker 1: when you picture the fact that like the Sprite uh,

108
00:06:25,320 --> 00:06:31,320
Speaker 1: the Sprites spokesperson is like, uh, isn't it Vince Staples? Yeah?

109
00:06:31,400 --> 00:06:34,960
Speaker 1: I think so. Yeah. So it's not really generally ever

110
00:06:35,160 --> 00:06:40,960
Speaker 1: sold by people that are so uncool that they think

111
00:06:41,640 --> 00:06:47,240
Speaker 1: renaming soda is some kind of I don't know, forward

112
00:06:48,360 --> 00:06:53,159
Speaker 1: thinking movement of their philosophy. Yeah, and it's both of

113
00:06:53,160 --> 00:06:57,360
Speaker 1: these cases, Tay and Mountain Dew's uh new soda flavor

114
00:06:57,440 --> 00:07:00,320
Speaker 1: vote where cases were any If you'd gone to either

115
00:07:00,400 --> 00:07:02,760
Speaker 1: of us in two thousand twelve or in early two

116
00:07:03,000 --> 00:07:04,760
Speaker 1: sixteen and said we're going to do this, what do

117
00:07:04,800 --> 00:07:06,599
Speaker 1: you think will happen? I think we both would have said.

118
00:07:06,600 --> 00:07:08,680
Speaker 1: I think every listener of this podcast would have said, Oh,

119
00:07:08,720 --> 00:07:11,120
Speaker 1: it's gonna get real nazi, like immediately, like it's going

120
00:07:11,200 --> 00:07:13,080
Speaker 1: to turn into a Nazi because that's just what people

121
00:07:13,080 --> 00:07:15,520
Speaker 1: on the internet I think is funny. Uh, and that's

122
00:07:15,520 --> 00:07:20,280
Speaker 1: going to happen. But like you know, older folks, people

123
00:07:20,320 --> 00:07:22,920
Speaker 1: who you know are focused more on living that life

124
00:07:22,920 --> 00:07:25,800
Speaker 1: of the mind off of the Internet, they didn't anticipate

125
00:07:25,840 --> 00:07:28,520
Speaker 1: that sort of stuff, and there really wasn't much of

126
00:07:28,560 --> 00:07:31,120
Speaker 1: a harm ever, you know, in either that Mountain Dew

127
00:07:31,160 --> 00:07:34,440
Speaker 1: contest or in the tay chat bought. Like Tay was

128
00:07:34,440 --> 00:07:37,320
Speaker 1: was a public facing AI, it was never in control

129
00:07:37,360 --> 00:07:40,920
Speaker 1: of something. But the question of its radicalization does lead

130
00:07:40,960 --> 00:07:45,040
Speaker 1: to the question, what if another company built an AI

131
00:07:45,080 --> 00:07:48,040
Speaker 1: that learned in that way that wasn't public facing, and

132
00:07:48,040 --> 00:07:51,520
Speaker 1: what if that company trusted the AI to handle a

133
00:07:51,560 --> 00:07:56,000
Speaker 1: crucial task that operated behind the scenes. And if that

134
00:07:56,040 --> 00:07:58,080
Speaker 1: were to happen, I think it might look an awful

135
00:07:58,120 --> 00:08:02,440
Speaker 1: lot like what we've seen happen to YouTube's recommendation algorithm.

136
00:08:02,520 --> 00:08:05,000
Speaker 1: Um I think I'm not the first person to make

137
00:08:05,040 --> 00:08:08,320
Speaker 1: this comparison that what had happened to YouTube's algorithm over

138
00:08:08,360 --> 00:08:11,280
Speaker 1: the last few years is what happened to that chat bought.

139
00:08:11,560 --> 00:08:15,080
Speaker 1: But since no one interacts with YouTube's algorithm directly, it

140
00:08:15,160 --> 00:08:18,400
Speaker 1: took a long time for people to realize that YouTube's

141
00:08:18,760 --> 00:08:22,760
Speaker 1: recommendation AI had turned into Joseph Gebel's, which is I

142
00:08:22,800 --> 00:08:26,640
Speaker 1: think where we are right now. Um, so that's what

143
00:08:26,640 --> 00:08:32,560
Speaker 1: today's episodes about, yea, I bring. I'm glad that it's

144
00:08:32,559 --> 00:08:35,000
Speaker 1: not about dead babies, because I know you love to

145
00:08:35,040 --> 00:08:39,040
Speaker 1: do that. Ship to me, it does in a little

146
00:08:39,040 --> 00:08:41,200
Speaker 1: bit on a hurting babies. Now, are you kidding me

147
00:08:41,840 --> 00:08:47,240
Speaker 1: a bit? Stop getting me here under false pretenses? Stop it.

148
00:08:48,000 --> 00:08:49,720
Speaker 1: I feel like at this point, you know, if you're

149
00:08:49,720 --> 00:08:52,120
Speaker 1: coming on behind the bastards, some babies are going to

150
00:08:52,200 --> 00:08:56,600
Speaker 1: get harmed. Okay, I assume sometimes maybe people just murder adults.

151
00:08:56,920 --> 00:08:58,960
Speaker 1: That's what I was hoping for coming in today. I

152
00:08:59,000 --> 00:09:02,959
Speaker 1: was like, they're still more adult murder than adult murder.

153
00:09:03,000 --> 00:09:06,199
Speaker 1: But no, we always have to get miners involved. If

154
00:09:06,240 --> 00:09:09,960
Speaker 1: it's me, don't we evans that the murders involved in

155
00:09:10,000 --> 00:09:13,720
Speaker 1: this episode we're all adults. The molest station involved in

156
00:09:13,720 --> 00:09:19,000
Speaker 1: this episode involved children so much. That's a step so much. No,

157
00:09:19,720 --> 00:09:23,320
Speaker 1: the Georgia tan One had child murder and child molestation.

158
00:09:24,240 --> 00:09:27,640
Speaker 1: Well now it's adult murder in child molestate, well child pornography.

159
00:09:27,760 --> 00:09:30,880
Speaker 1: Will you let me live a life full of just

160
00:09:31,000 --> 00:09:34,360
Speaker 1: adult murder? You know what, Sophia, I'll make this promise

161
00:09:34,400 --> 00:09:36,240
Speaker 1: to you right now over the internet. When we do

162
00:09:36,280 --> 00:09:39,320
Speaker 1: our one yearly optimistic episode about a person who's not

163
00:09:39,360 --> 00:09:42,000
Speaker 1: a bastard this upcoming Christmas, I'll have you want us

164
00:09:42,000 --> 00:09:46,319
Speaker 1: the guest for that one. Fuck yes, I can't wait. Um,

165
00:09:46,640 --> 00:09:49,840
Speaker 1: And hopefully the irony of that episode will be that

166
00:09:50,080 --> 00:09:52,880
Speaker 1: very shortly thereafter we'll find out that person is also

167
00:09:52,920 --> 00:09:55,360
Speaker 1: a bastard. Yeah, it'll be the story of the person

168
00:09:55,400 --> 00:09:59,600
Speaker 1: who saved a thousand kids by killing nine hundred. Still

169
00:10:00,040 --> 00:10:03,439
Speaker 1: it like net game that will be exactly your pitch

170
00:10:03,440 --> 00:10:08,360
Speaker 1: when that happens, I'm already googling. Yeah, you're like, how

171
00:10:08,400 --> 00:10:11,640
Speaker 1: can we let's get back to YouTube. As I write this,

172
00:10:11,720 --> 00:10:14,240
Speaker 1: the Internet is still reeling from the shock waves caused

173
00:10:14,240 --> 00:10:16,800
Speaker 1: by a gigantic battle over whether or not YouTube should

174
00:10:16,840 --> 00:10:19,720
Speaker 1: ban conservative comedian And I put that in air quotes

175
00:10:20,320 --> 00:10:22,960
Speaker 1: Stephen Crowder. Now, if you're lucky enough to not know

176
00:10:23,000 --> 00:10:25,000
Speaker 1: about him, Crowder is a bigot who spends most of

177
00:10:25,000 --> 00:10:27,400
Speaker 1: his time verbally attacking people who look different than him.

178
00:10:27,400 --> 00:10:30,400
Speaker 1: He spent several months harassing Carlos Maza, who makes YouTube

179
00:10:30,440 --> 00:10:33,440
Speaker 1: videos for Vox, calling Maza a list be queer, and

180
00:10:33,480 --> 00:10:36,320
Speaker 1: a number of other horrible things. Uh. Crowder has not

181
00:10:36,360 --> 00:10:39,240
Speaker 1: explicitly directed his fans to attack Carlos in real life,

182
00:10:39,240 --> 00:10:41,520
Speaker 1: but Crowder's fans don't need to be told to do that.

183
00:10:41,720 --> 00:10:44,400
Speaker 1: When he directs his iro at an individual, Crowder fans

184
00:10:44,400 --> 00:10:49,880
Speaker 1: swarm that individual. Uh. Carlos is regularly bombarded with text messages, emails, tweets, etcetera,

185
00:10:49,920 --> 00:10:52,880
Speaker 1: calling him horrible names, asking him, demanding that he debates

186
00:10:52,880 --> 00:10:55,360
Speaker 1: Stephen Crowder, telling him to kill himself, doing all the

187
00:10:55,440 --> 00:10:58,440
Speaker 1: kind of things that sociopathic internet trolls like to do

188
00:10:58,600 --> 00:11:02,680
Speaker 1: to the targets of their I now Carlos on Twitter

189
00:11:02,800 --> 00:11:05,880
Speaker 1: asked YouTube to ban Crowder, and he pointed out specific

190
00:11:05,920 --> 00:11:09,480
Speaker 1: things Crowder had said and highlighted specific sections of YouTube's

191
00:11:09,559 --> 00:11:13,000
Speaker 1: terms of service that Crowder had violated. UH. YouTube opted

192
00:11:13,080 --> 00:11:15,920
Speaker 1: not to ban Crowder because Crowder has nearly four million

193
00:11:15,960 --> 00:11:18,880
Speaker 1: followers and makes YouTube a lot of money. UH. There

194
00:11:18,920 --> 00:11:22,360
Speaker 1: has been more dumb fallout. YouTube demonetized Crowder's channel and

195
00:11:22,400 --> 00:11:25,120
Speaker 1: then randomly demonetized a bunch of other people so conservatives

196
00:11:25,120 --> 00:11:27,760
Speaker 1: couldn't claim they were being oppressed. And it's all a big, gross,

197
00:11:27,840 --> 00:11:31,200
Speaker 1: ugly mess. But the real problem here, the issue at

198
00:11:31,240 --> 00:11:34,160
Speaker 1: the core of this latest eruption in our national culture war,

199
00:11:34,559 --> 00:11:37,440
Speaker 1: has nothing to do with YouTube's craven refusal to enforce

200
00:11:37,480 --> 00:11:40,400
Speaker 1: their own rules. Stephen Crowder would not be a figure

201
00:11:40,440 --> 00:11:42,880
Speaker 1: in our nation's political discourse if it weren't for a

202
00:11:42,920 --> 00:11:46,160
Speaker 1: series of changes YouTube started making to their algorithm in

203
00:11:46,200 --> 00:11:51,080
Speaker 1: two thousand ten. Now, YouTube's recommendation algorithm is what you know.

204
00:11:51,120 --> 00:11:53,320
Speaker 1: It recommends the next video that you should watch. It's

205
00:11:53,360 --> 00:11:55,400
Speaker 1: why if you play enough music videos while logged in,

206
00:11:55,600 --> 00:11:58,520
Speaker 1: YouTube will gradually start to learn your preferences and suggest

207
00:11:58,600 --> 00:12:02,400
Speaker 1: new music that often you like. It's also why teenagers

208
00:12:02,400 --> 00:12:05,080
Speaker 1: who look up the Federal Reserve for a school report

209
00:12:05,080 --> 00:12:08,280
Speaker 1: will inevitably find themselves recommended something that's basically the protocols

210
00:12:08,320 --> 00:12:13,240
Speaker 1: of the Elders of Zion with better animation. Oh my god, yeah, yeah,

211
00:12:13,320 --> 00:12:15,720
Speaker 1: it's it's both of those things. Well, but the animation

212
00:12:15,840 --> 00:12:20,240
Speaker 1: is good, Okay, playing some money behind that anti semitism, Yeah,

213
00:12:20,840 --> 00:12:23,400
Speaker 1: it's a it's it's it's a mixed bag. On one hand,

214
00:12:23,480 --> 00:12:25,640
Speaker 1: I learned about the music of Tom Russell, who's a

215
00:12:25,760 --> 00:12:29,120
Speaker 1: musician I very much enjoy. Now. On the other hand, uh,

216
00:12:29,240 --> 00:12:35,720
Speaker 1: there's thousands more nazis so really pretty even exchange, I say, yeah,

217
00:12:36,760 --> 00:12:41,240
Speaker 1: fair mix, Yeah, it's good trade. Yeah. Now, I do

218
00:12:41,360 --> 00:12:45,200
Speaker 1: really like Tom Russell's music, but that's not the important thing.

219
00:12:45,280 --> 00:12:48,160
Speaker 1: Is Tom Russell not be offended? Okay, let's make sure

220
00:12:48,320 --> 00:12:52,880
Speaker 1: he's fine. Yeah. Now. YouTube's recommendation engine was not always

221
00:12:52,920 --> 00:12:55,760
Speaker 1: a core part of the site's functionality. In the early

222
00:12:55,840 --> 00:12:57,920
Speaker 1: days of YouTube and two thousand six or seven, or

223
00:12:57,960 --> 00:13:00,600
Speaker 1: eight or nine, most of the content was focused around

224
00:13:00,720 --> 00:13:03,480
Speaker 1: channels a lot like television. People would search for what

225
00:13:03,520 --> 00:13:05,520
Speaker 1: they wanted to see, and they would tune into stuff

226
00:13:05,520 --> 00:13:08,560
Speaker 1: they knew that they liked. Unfortunately, that meant people would

227
00:13:08,640 --> 00:13:11,360
Speaker 1: leave YouTube when they were done watching stuff. I'd like

228
00:13:11,400 --> 00:13:13,400
Speaker 1: to quote now from a very good article in The

229
00:13:13,520 --> 00:13:16,840
Speaker 1: Verge by Casey Newton. He interviewed Jim McFadden, who joined

230
00:13:16,880 --> 00:13:19,240
Speaker 1: YouTube in two thousand eleven and worked as the technical

231
00:13:19,320 --> 00:13:23,120
Speaker 1: lead for YouTube recommendations. Quote. We knew people were coming

232
00:13:23,120 --> 00:13:24,720
Speaker 1: to YouTube when they knew what they were coming to

233
00:13:24,800 --> 00:13:26,880
Speaker 1: look for. We also wanted to serve the needs of

234
00:13:26,880 --> 00:13:29,080
Speaker 1: people when they didn't necessarily know what they wanted to

235
00:13:29,080 --> 00:13:32,080
Speaker 1: look for. Casey goes on to write, I first visited

236
00:13:32,080 --> 00:13:33,839
Speaker 1: the company in two thousand eleven, just a few months

237
00:13:33,880 --> 00:13:36,600
Speaker 1: after McFadden joined. Getting users to spend more time watching

238
00:13:36,679 --> 00:13:40,040
Speaker 1: videos was then, as now YouTube's primary aim. At the time,

239
00:13:40,080 --> 00:13:43,080
Speaker 1: it was not going particularly well. YouTube dot com as

240
00:13:43,120 --> 00:13:45,960
Speaker 1: a homepage was not driving a ton of engagement. McFadden says,

241
00:13:46,040 --> 00:13:48,040
Speaker 1: we said, well, how do we turn this thing into

242
00:13:48,080 --> 00:13:52,120
Speaker 1: a destination? So YouTube tried a bunch of different things.

243
00:13:52,120 --> 00:13:54,640
Speaker 1: They tried buying professional gear for their top creators to

244
00:13:55,000 --> 00:13:57,880
Speaker 1: increase the quality of YouTube content, but that just made

245
00:13:57,920 --> 00:14:01,319
Speaker 1: YouTube more enjoyable. It didn't make this service more addictive.

246
00:14:01,800 --> 00:14:05,600
Speaker 1: So in two thousand eleven, they launched lean Back. Now,

247
00:14:05,720 --> 00:14:08,600
Speaker 1: lean back would automatically pick a new video at random

248
00:14:08,640 --> 00:14:10,839
Speaker 1: for you to watch after you finished your old video.

249
00:14:11,200 --> 00:14:13,680
Speaker 1: Lean Back became the heart of the algorithm we all

250
00:14:13,720 --> 00:14:16,640
Speaker 1: know and many of us hate today. At first, lean

251
00:14:16,720 --> 00:14:19,000
Speaker 1: Back would select new videos for people to watch based

252
00:14:19,040 --> 00:14:21,560
Speaker 1: on what seemed like a reasonable metric, the number of

253
00:14:21,640 --> 00:14:24,760
Speaker 1: views those videos had received. So if more people watched

254
00:14:24,760 --> 00:14:27,040
Speaker 1: a video, it was more likely to wind up recommended

255
00:14:27,080 --> 00:14:29,880
Speaker 1: to new people. But it turned out lean back didn't

256
00:14:29,880 --> 00:14:33,000
Speaker 1: actually impact the amount of time spent on site per user.

257
00:14:33,360 --> 00:14:36,360
Speaker 1: So in two thousand twelve, YouTube started basing recommendations on

258
00:14:36,400 --> 00:14:40,000
Speaker 1: how long people spent watching videos, so it's engine switched

259
00:14:40,040 --> 00:14:42,560
Speaker 1: from recommending videos a lot of people have watched to

260
00:14:42,640 --> 00:14:45,560
Speaker 1: recommending videos people had spent a lot of time on. Now,

261
00:14:45,600 --> 00:14:48,120
Speaker 1: this seemed like a great idea at first. According to

262
00:14:48,160 --> 00:14:51,000
Speaker 1: The Verge, nearly overnight, creators who had profited for misleading

263
00:14:51,000 --> 00:14:54,360
Speaker 1: headlines and thumbnails saw their view counts plummet. Higher quality videos,

264
00:14:54,360 --> 00:14:58,120
Speaker 1: which are strongly associated with longer watch times, surged. Watchtime

265
00:14:58,200 --> 00:15:01,200
Speaker 1: on YouTube grew fifty a year for the next three years.

266
00:15:01,920 --> 00:15:05,760
Speaker 1: So that sounds great, right, not that evil yet horrible.

267
00:15:07,360 --> 00:15:11,040
Speaker 1: Let's read the next paragraph during this period of time

268
00:15:12,520 --> 00:15:15,080
Speaker 1: really quickly. Yeah, I was waiting for you to start

269
00:15:15,080 --> 00:15:17,520
Speaker 1: talking to them, interrupt you. No, I wanted to know

270
00:15:17,760 --> 00:15:22,600
Speaker 1: if if part of the lean back algorithm was that

271
00:15:23,480 --> 00:15:27,920
Speaker 1: they would just automatically play lean back by Fat Joe.

272
00:15:28,920 --> 00:15:32,680
Speaker 1: If that had been the YouTube algorithm, the numbers, that's

273
00:15:32,760 --> 00:15:34,760
Speaker 1: what they would do after any video you would watch.

274
00:15:35,200 --> 00:15:37,440
Speaker 1: We if that had been what had happened Sophia. We

275
00:15:37,480 --> 00:15:40,640
Speaker 1: would live in a paradise. Climate change would have been

276
00:15:40,640 --> 00:15:43,960
Speaker 1: dealt with. The president would be a being of pure light.

277
00:15:44,720 --> 00:15:47,520
Speaker 1: It would be there would be peace in Ukraine and Syria.

278
00:15:47,560 --> 00:15:50,520
Speaker 1: It would be a perfect world. If only if only

279
00:15:51,280 --> 00:15:55,280
Speaker 1: YouTube's lean Back be backing people to the music video

280
00:15:55,400 --> 00:15:59,480
Speaker 1: for lean Back. I mean that's a chill ass Jamay,

281
00:15:59,640 --> 00:16:03,200
Speaker 1: that is chill ass jam There would be no Nazis

282
00:16:03,280 --> 00:16:06,200
Speaker 1: in twenty nineteen. If that's the change YouTube and made.

283
00:16:06,760 --> 00:16:12,920
Speaker 1: It's true, Fat Joe transcends the boundaries of country, religion,

284
00:16:13,640 --> 00:16:17,840
Speaker 1: skin color, anything. You could have saved the world YouTube,

285
00:16:17,880 --> 00:16:21,440
Speaker 1: if you just pushed fat Joe on a welcoming nation

286
00:16:22,840 --> 00:16:26,600
Speaker 1: into the into the longing arms of a nation. Yeah,

287
00:16:26,640 --> 00:16:29,360
Speaker 1: God damn, I wish that. I wish that's the path

288
00:16:29,400 --> 00:16:33,960
Speaker 1: things have taken. Tragically, it's not now. During this period

289
00:16:34,000 --> 00:16:38,240
Speaker 1: after lean Back was instituted, Gillim Chesslo was a software

290
00:16:38,240 --> 00:16:42,400
Speaker 1: engineer for Google. I'm sorry, Gillim Chasslow Chasslo. It's spelled

291
00:16:42,440 --> 00:16:44,440
Speaker 1: c h s l o T. I found I think

292
00:16:44,440 --> 00:16:47,120
Speaker 1: I'm pronouncing gilhim right, because I found some pronunciation guides

293
00:16:47,160 --> 00:16:50,120
Speaker 1: for the name gillim Um, but I've not found a

294
00:16:50,120 --> 00:16:53,000
Speaker 1: pronunciation guide for c H A S L O T.

295
00:16:53,200 --> 00:16:55,160
Speaker 1: I think Chaslow is. I think he's a French guy.

296
00:16:56,400 --> 00:16:58,800
Speaker 1: It's a very good name. It is a great name.

297
00:16:58,840 --> 00:17:02,080
Speaker 1: I think I'm pronouncing sort of correct Gillim Chaslow. But

298
00:17:02,480 --> 00:17:08,520
Speaker 1: I'm doing my best here to get dommed in the evenings. Yeah, yeah,

299
00:17:08,840 --> 00:17:11,879
Speaker 1: Gillim Chaslow for sure, stuffy bank owner. But in this

300
00:17:11,920 --> 00:17:16,400
Speaker 1: case he's actually an engineer whose expertise is an artificial

301
00:17:16,440 --> 00:17:19,720
Speaker 1: intelligence uh and The Guardian interviewed him for an article

302
00:17:19,800 --> 00:17:23,040
Speaker 1: titled how YouTube's algorithm distorts reality. I'm gonna quote from

303
00:17:23,040 --> 00:17:25,600
Speaker 1: that now. During the three years he worked at Google,

304
00:17:25,880 --> 00:17:27,560
Speaker 1: he was placed for several months with a team of

305
00:17:27,600 --> 00:17:30,840
Speaker 1: YouTube engineers working on the recommendation system. The experience led

306
00:17:30,920 --> 00:17:33,720
Speaker 1: him to conclude that the priorities YouTube gives its algorithms

307
00:17:33,720 --> 00:17:37,080
Speaker 1: are dangerously skewed. YouTube is something that looks like reality,

308
00:17:37,119 --> 00:17:39,520
Speaker 1: but it is distorted to make you spend more time online,

309
00:17:39,560 --> 00:17:41,840
Speaker 1: he tells me when we meet in Berkeley, California. The

310
00:17:41,920 --> 00:17:45,439
Speaker 1: recommendation algorithm is not optimizing for what is truthful, or

311
00:17:45,480 --> 00:17:49,440
Speaker 1: balanced or healthy for democracy. Chaslow explains that the algorithm

312
00:17:49,520 --> 00:17:51,880
Speaker 1: never stays the same it is constantly changing the way

313
00:17:51,920 --> 00:17:54,400
Speaker 1: it gives to different signals, the viewing patterns of a user,

314
00:17:54,440 --> 00:17:56,280
Speaker 1: for example, or the length of time a video was

315
00:17:56,320 --> 00:17:59,239
Speaker 1: watched before someone clicks away. The engineers he worked for,

316
00:17:59,280 --> 00:18:02,520
Speaker 1: we're responsible for continuously experimenting with new formulas that would

317
00:18:02,520 --> 00:18:05,520
Speaker 1: increase advertising revenue by extending the amounts of time people

318
00:18:05,520 --> 00:18:09,159
Speaker 1: watched videos. Watch time was the priority. Everything else was

319
00:18:09,200 --> 00:18:14,399
Speaker 1: considered a distraction. So YouTube builds this robot to decide

320
00:18:14,400 --> 00:18:16,520
Speaker 1: what you're gonna listen to next, and the robots only

321
00:18:17,000 --> 00:18:20,440
Speaker 1: concern is that you spend as much time as possible

322
00:18:20,520 --> 00:18:23,600
Speaker 1: on YouTube. And that's the seed of all of the

323
00:18:23,680 --> 00:18:26,520
Speaker 1: problems that we're going to be talking about today. So

324
00:18:27,200 --> 00:18:29,840
Speaker 1: Gillham was fired in two thousand thirteen, and Google says

325
00:18:29,880 --> 00:18:32,399
Speaker 1: it's because he was bad at his job. Chaslow claims

326
00:18:32,400 --> 00:18:34,840
Speaker 1: that they instead fired him because he complained about what

327
00:18:34,880 --> 00:18:37,240
Speaker 1: he saw is the dangerous potential of the algorithm to

328
00:18:37,320 --> 00:18:40,440
Speaker 1: radicalize people. Uh. He worried that the algorithm would lock

329
00:18:40,480 --> 00:18:43,399
Speaker 1: people into filter bubbles that only reinforce their beliefs and

330
00:18:43,480 --> 00:18:47,040
Speaker 1: make conservatives more conservative, liberals more liberal, and people who

331
00:18:47,080 --> 00:18:49,840
Speaker 1: like watching documentaries about aliens more convinced that the Jews

332
00:18:49,840 --> 00:18:58,240
Speaker 1: are fluoridating their water, et cetera. Uh, thank you for laughing. Yeah.

333
00:18:58,440 --> 00:19:00,880
Speaker 1: Chaslow said there are many ways YouTube can change its

334
00:19:00,880 --> 00:19:03,199
Speaker 1: algorithms to suppress fake news and improve the quality and

335
00:19:03,200 --> 00:19:05,720
Speaker 1: diversity of videos people see. I try to change YouTube

336
00:19:05,720 --> 00:19:09,120
Speaker 1: from the inside, but it didn't work. YouTube's masters, of course,

337
00:19:09,160 --> 00:19:12,159
Speaker 1: had no desire to diversify the kind of content people saw.

338
00:19:12,480 --> 00:19:14,640
Speaker 1: Why would they do that if it meant folks would

339
00:19:14,640 --> 00:19:18,719
Speaker 1: spend less time on the site, So it doesn't. Fifteen

340
00:19:18,760 --> 00:19:23,680
Speaker 1: YouTube integrated Google Brain, a machine learning program, into its algorithm.

341
00:19:23,720 --> 00:19:26,320
Speaker 1: According to an engineer interview by The Verge, one of

342
00:19:26,359 --> 00:19:28,560
Speaker 1: the key things it does is it's able to generalize.

343
00:19:28,720 --> 00:19:31,000
Speaker 1: Whereas before, if I watch a video from a comedian,

344
00:19:31,000 --> 00:19:33,360
Speaker 1: our recommendations were pretty good at saying here's another one

345
00:19:33,440 --> 00:19:35,760
Speaker 1: just like it. But the Google Brain model figures out

346
00:19:35,800 --> 00:19:38,240
Speaker 1: other comedians who are similar but not exactly the same,

347
00:19:38,320 --> 00:19:41,359
Speaker 1: even more adjacent relationships. It's able to see patterns that

348
00:19:41,400 --> 00:19:44,239
Speaker 1: are less obvious, and Google Brain is a big part

349
00:19:44,280 --> 00:19:47,160
Speaker 1: of why Stephen Crowder and others like him are now millionaires.

350
00:19:47,440 --> 00:19:49,680
Speaker 1: It's why if you watch a Joe Rogan video, you'll

351
00:19:49,720 --> 00:19:53,240
Speaker 1: start being recommended videos by Ben Shapiro or Paul Joseph Watson,

352
00:19:53,520 --> 00:19:56,280
Speaker 1: even though Joe Rogan is not an explicitly political guy

353
00:19:56,280 --> 00:19:59,360
Speaker 1: in Ben Shapiro and Paul Joseph Watson are. It's why

354
00:19:59,480 --> 00:20:02,520
Speaker 1: for year, whenever conservative inclined to people would start watching,

355
00:20:02,560 --> 00:20:05,480
Speaker 1: say a Fox News clip critical of Obama, they'd wind

356
00:20:05,560 --> 00:20:08,760
Speaker 1: up being shuffled gently over to Info Wars and Alex Jones.

357
00:20:09,119 --> 00:20:12,639
Speaker 1: It's why if you watched a video about Obama's birth certificate,

358
00:20:12,680 --> 00:20:15,520
Speaker 1: YouTube would next serve you Alex Jones claiming that Michelle

359
00:20:15,520 --> 00:20:18,320
Speaker 1: Obama is secretly a man. It's why if you watched

360
00:20:18,320 --> 00:20:21,000
Speaker 1: a video criticizing gun control, YouTube would serve you up

361
00:20:21,040 --> 00:20:24,399
Speaker 1: Alex Jones claiming the New World Order under Obama was

362
00:20:24,440 --> 00:20:27,000
Speaker 1: going to confiscate your guns so we could carry out genocide.

363
00:20:27,240 --> 00:20:29,119
Speaker 1: And it's why if you watched coverage of the Sandy

364
00:20:29,160 --> 00:20:33,080
Speaker 1: Hook massacre, YouTube would hand you Alex Jones claiming the

365
00:20:33,080 --> 00:20:35,640
Speaker 1: massacre was a false flag and all the children involved

366
00:20:35,640 --> 00:20:38,800
Speaker 1: were crisis actors. I bring up Alex Jones so many

367
00:20:38,840 --> 00:20:41,480
Speaker 1: times in this because it's probable that no single person

368
00:20:41,520 --> 00:20:45,600
Speaker 1: benefited as much from YouTube's Google brain algorithm changes as

369
00:20:45,640 --> 00:20:49,239
Speaker 1: Alex Jones. That's what Gilim Chaslow seems to think. On

370
00:20:49,280 --> 00:20:54,040
Speaker 1: February eighteen, he tweeted this, the algorithms I worked out

371
00:20:54,080 --> 00:20:59,120
Speaker 1: on Google recommended Alex Jones videos more than fifteen billion times.

372
00:20:59,200 --> 00:21:01,760
Speaker 1: Vulnerable be pull in the nation. Yeah. Fifth, that's the

373
00:21:01,800 --> 00:21:07,800
Speaker 1: scale of this thing because it recognizes that people who

374
00:21:07,800 --> 00:21:09,720
Speaker 1: are going to start like watching just sort of a

375
00:21:09,760 --> 00:21:14,919
Speaker 1: conservative take on whatever issue gun control, uh, the Sandy

376
00:21:14,960 --> 00:21:18,000
Speaker 1: Hook shooting, uh, fluoride in the water or whatever, people

377
00:21:18,000 --> 00:21:20,680
Speaker 1: who might just want like a Fox News take on that.

378
00:21:21,440 --> 00:21:25,280
Speaker 1: Alex Jones is much more extreme. But because he's much

379
00:21:25,359 --> 00:21:28,960
Speaker 1: more extreme, he's like compelling to those people, and if

380
00:21:28,960 --> 00:21:31,480
Speaker 1: you serve him them, they watch his stuff all the

381
00:21:31,520 --> 00:21:33,879
Speaker 1: way through. And his videos are really really long. It

382
00:21:34,000 --> 00:21:35,960
Speaker 1: is like a four hour show, So people stay on

383
00:21:36,000 --> 00:21:37,840
Speaker 1: the site a long time. If they get served up

384
00:21:37,880 --> 00:21:40,639
Speaker 1: a four hour Alex Jones video, they just keep playing

385
00:21:40,640 --> 00:21:43,000
Speaker 1: it while they're doing whatever they're doing, and they sink

386
00:21:43,080 --> 00:21:46,760
Speaker 1: deeper and deeper into that rabbit hole. And a regular

387
00:21:46,800 --> 00:21:48,960
Speaker 1: person would look at this and be like, oh, Google's

388
00:21:49,000 --> 00:21:51,280
Speaker 1: taking people who believe, I don't know that a flat

389
00:21:51,280 --> 00:21:53,760
Speaker 1: tax is a good idea and turning them into people

390
00:21:53,800 --> 00:21:56,680
Speaker 1: who think that Floride is turning frogs gay and that

391
00:21:56,920 --> 00:21:59,760
Speaker 1: Sandy Hook was an inside job and that's a bad thing.

392
00:22:00,119 --> 00:22:03,480
Speaker 1: But YouTube algorithm didn't think that way. It just thought like, oh,

393
00:22:04,119 --> 00:22:06,399
Speaker 1: as soon as these people find Alex Jones videos, they

394
00:22:06,440 --> 00:22:09,600
Speaker 1: spend fifty more time on YouTube. So I'm just gonna

395
00:22:09,600 --> 00:22:11,679
Speaker 1: serve Alex Jones up to as many fucking people as

396
00:22:11,680 --> 00:22:18,639
Speaker 1: I possibly can. And that's what starts happening in So

397
00:22:19,920 --> 00:22:24,480
Speaker 1: that's where we are in the story right now, and

398
00:22:24,520 --> 00:22:26,840
Speaker 1: then we're going to continue from that point. But you know,

399
00:22:26,880 --> 00:22:31,280
Speaker 1: it is time for next, Sophia, No tell me, it's

400
00:22:31,320 --> 00:22:37,760
Speaker 1: time for products. And maybe maybe I'm not gonna make promises.

401
00:22:37,880 --> 00:22:40,120
Speaker 1: I'm not gonna write checks. My ass can't catch here.

402
00:22:40,600 --> 00:22:46,720
Speaker 1: But maybe there's services you're ask in cash. Well, my

403
00:22:46,840 --> 00:22:51,320
Speaker 1: ass is all about products. I hope it's a chair

404
00:22:51,359 --> 00:22:53,800
Speaker 1: company that comes up next. Otherwise that's a non sequitor.

405
00:22:53,920 --> 00:22:57,760
Speaker 1: I hope it's a it's a squatty potty. It's probably

406
00:22:57,760 --> 00:22:59,840
Speaker 1: gonna be Dick Pills because we just we just signed

407
00:22:59,840 --> 00:23:02,560
Speaker 1: a deal with Dick Pills. I'm very, very proud of

408
00:23:02,560 --> 00:23:05,879
Speaker 1: our of our Dick Pills sponsorship. It's not even a

409
00:23:06,080 --> 00:23:09,840
Speaker 1: job I love. I love selling Dick pills. I can

410
00:23:09,880 --> 00:23:12,359
Speaker 1: see your hard right now. I can just see your head,

411
00:23:12,520 --> 00:23:15,600
Speaker 1: your head, of your body and not your penis head.

412
00:23:15,760 --> 00:23:18,280
Speaker 1: But I can tell you're hard from the pills. Thank you.

413
00:23:18,640 --> 00:23:23,200
Speaker 1: You have a very uh taught Dick energy. You know.

414
00:23:23,400 --> 00:23:28,439
Speaker 1: Thank you is what this show aims to present to

415
00:23:28,480 --> 00:23:32,480
Speaker 1: the world. Um you know, I said, speaking on the

416
00:23:32,480 --> 00:23:34,560
Speaker 1: subject of YouTube when we when we filled out our

417
00:23:34,600 --> 00:23:37,440
Speaker 1: ad things, I won't sell brain pills because I don't

418
00:23:37,440 --> 00:23:40,000
Speaker 1: want to be like Paul, Joseph Watson or Big Shapiro.

419
00:23:40,640 --> 00:23:43,800
Speaker 1: But I will hundred sell Dick pills. And it's mainly

420
00:23:43,960 --> 00:23:46,000
Speaker 1: so that I can say the phrase Dick pills over

421
00:23:46,040 --> 00:23:50,960
Speaker 1: and over again. Um. So, meet my son, Dick Pills Evans,

422
00:23:51,400 --> 00:23:53,560
Speaker 1: Dick Pills Evans. I am gonna I'm gonna name my

423
00:23:53,680 --> 00:23:55,800
Speaker 1: I'm gonna have a son just to name him Dick Pills.

424
00:23:55,840 --> 00:23:58,280
Speaker 1: And then sort of it's gonna be like a boy

425
00:23:58,359 --> 00:24:01,800
Speaker 1: named Sue, but with a boy name Dick PILs. And

426
00:24:01,880 --> 00:24:04,480
Speaker 1: instead of like me explaining to him that I gave

427
00:24:04,560 --> 00:24:06,359
Speaker 1: him the name Sue so that he'd be like, it

428
00:24:06,400 --> 00:24:08,560
Speaker 1: would harden him up and he'd become like a tough person.

429
00:24:08,640 --> 00:24:10,399
Speaker 1: He could survive the rough world. And like, oh no,

430
00:24:10,520 --> 00:24:11,879
Speaker 1: I got paid a lot of money to call you

431
00:24:11,920 --> 00:24:18,159
Speaker 1: Dick Piles. No, you're just sponsored by Dick Pills. This

432
00:24:18,240 --> 00:24:20,919
Speaker 1: has gone very off the rails, Sophie, is this a

433
00:24:20,960 --> 00:24:27,800
Speaker 1: good idea? No, she's doing a hard no hard No, okay,

434
00:24:27,880 --> 00:24:41,960
Speaker 1: well speaking hard products. We're back. We're back, And Sophia

435
00:24:42,080 --> 00:24:44,680
Speaker 1: just said the sentence we got to mold our own

436
00:24:44,720 --> 00:24:49,200
Speaker 1: genitals at the Dick Johnson factory or Dog Johnson factory,

437
00:24:48,560 --> 00:24:52,199
Speaker 1: Doc Johnson. I loved that sentence, which is why I

438
00:24:52,240 --> 00:24:54,800
Speaker 1: brought us back in mid conversation from the ad break,

439
00:24:55,000 --> 00:25:00,359
Speaker 1: because that's a wonderful sentence. Um, I have that Agram

440
00:25:00,400 --> 00:25:03,199
Speaker 1: stories saved on my Instagram. I want to get that

441
00:25:03,359 --> 00:25:07,679
Speaker 1: sentence tattooed on my back where some people will have

442
00:25:07,760 --> 00:25:11,480
Speaker 1: Jesus like I got my genitals molded at the Doc

443
00:25:11,600 --> 00:25:14,520
Speaker 1: Johnson factory. Yeah, and it was the most fun ever.

444
00:25:15,200 --> 00:25:20,800
Speaker 1: That sounds great. That sounds so much better than YouTube's algorithm.

445
00:25:21,080 --> 00:25:24,040
Speaker 1: That's a really smooth transition, thank you. That was like

446
00:25:24,280 --> 00:25:29,200
Speaker 1: jazz fucking saxophone smooth. I am as good at transitions

447
00:25:29,240 --> 00:25:35,480
Speaker 1: as Dick Pills are. Getting your exactly exactly fun. Yeah, hymns,

448
00:25:36,200 --> 00:25:41,280
Speaker 1: good times. Okay. So you know we were one of

449
00:25:41,280 --> 00:25:43,760
Speaker 1: the big sources for this podcast and one of the

450
00:25:43,760 --> 00:25:46,080
Speaker 1: big sources for the articles that have covered the problems

451
00:25:46,080 --> 00:25:50,440
Speaker 1: with YouTube's algorithm is Gillim Cheslow. And he's not just

452
00:25:50,600 --> 00:25:53,200
Speaker 1: a former employee with an axe to grind or someone

453
00:25:53,240 --> 00:25:56,159
Speaker 1: who feels guilty about the work he participated in. For

454
00:25:56,680 --> 00:25:59,800
Speaker 1: years now, he has turned into something of an activist

455
00:25:59,840 --> 00:26:02,080
Speaker 1: against what he sees as the harms of his former

456
00:26:02,119 --> 00:26:06,199
Speaker 1: employer UM and obviously as a guy with potentially an

457
00:26:06,200 --> 00:26:07,960
Speaker 1: ax to grind, he's someone that you've got to approach

458
00:26:07,960 --> 00:26:11,200
Speaker 1: a little bit critically. But Chaslow hasn't just like complained

459
00:26:11,240 --> 00:26:13,760
Speaker 1: about Google. He's built like he has a team of

460
00:26:13,800 --> 00:26:16,720
Speaker 1: people that have built like systems in order to test

461
00:26:16,760 --> 00:26:19,800
Speaker 1: the way Google's algorithm works and show the way that

462
00:26:19,840 --> 00:26:25,080
Speaker 1: it picks new content and like, uh, document with hard numbers,

463
00:26:25,119 --> 00:26:28,040
Speaker 1: like here's the kind of things that it's serving up,

464
00:26:28,119 --> 00:26:30,920
Speaker 1: Here's the sort of videos that it recommends people towards,

465
00:26:30,960 --> 00:26:33,880
Speaker 1: Like here's how often it's doing them. So he's he's

466
00:26:33,880 --> 00:26:37,840
Speaker 1: not just making claims. He has reams and reams of

467
00:26:37,920 --> 00:26:42,600
Speaker 1: documentation on how Google's algorithm works behind him. Um, he's

468
00:26:42,640 --> 00:26:44,400
Speaker 1: really put a lot of work into this, and from

469
00:26:44,400 --> 00:26:47,640
Speaker 1: everything I can tell, he's someone who's deeply concerned about

470
00:26:47,640 --> 00:26:50,879
Speaker 1: the impact YouTube's algorithm has had on our democracy and

471
00:26:50,960 --> 00:26:54,119
Speaker 1: someone who's trying to do something about it. So just

472
00:26:54,280 --> 00:26:55,879
Speaker 1: digging into the guy a bit. I have a lot

473
00:26:55,920 --> 00:26:58,520
Speaker 1: of respect for what he's trying to do. UM. On

474
00:26:58,600 --> 00:27:03,280
Speaker 1: November and sixteen, shortly after the election, while we were

475
00:27:03,280 --> 00:27:07,520
Speaker 1: all drinking heavily, uh, gilham Chaslo published a medium post

476
00:27:07,680 --> 00:27:11,680
Speaker 1: titled YouTube's AI was Divisive in the US presidential Election.

477
00:27:12,160 --> 00:27:14,359
Speaker 1: In it, he included the results of a study he

478
00:27:14,359 --> 00:27:18,439
Speaker 1: he and a team of researchers conducted. UM. They were

479
00:27:18,480 --> 00:27:21,280
Speaker 1: essentially trying to measure which candidate was recommended the most

480
00:27:21,320 --> 00:27:24,800
Speaker 1: by YouTube's AI during the presidential election, and the code

481
00:27:24,840 --> 00:27:26,439
Speaker 1: that they used to do this in All of the

482
00:27:26,440 --> 00:27:28,840
Speaker 1: methodology behind it is available on the website. If you're

483
00:27:28,880 --> 00:27:30,439
Speaker 1: someone who knows how to do the coding, you can

484
00:27:30,520 --> 00:27:34,199
Speaker 1: check it all up. But they're very transparent, uh, he says. Quote. Surprisingly,

485
00:27:34,200 --> 00:27:36,359
Speaker 1: a Clinton search on the eve of the election led

486
00:27:36,440 --> 00:27:39,760
Speaker 1: to mostly anti Clinton videos. The pro Clinton videos were

487
00:27:39,840 --> 00:27:42,840
Speaker 1: viewed many times and had high ratings. But represent only

488
00:27:42,920 --> 00:27:48,200
Speaker 1: less than twenty of all recommended videos. Chaslow's research found

489
00:27:48,200 --> 00:27:51,200
Speaker 1: that the vast majority of political videos recommended by YouTube

490
00:27:51,359 --> 00:27:55,280
Speaker 1: were anti Clinton and pro Trump, because those videos got

491
00:27:55,280 --> 00:28:00,400
Speaker 1: the best engagement now. Chaslow explained that because Google Brain

492
00:28:00,480 --> 00:28:03,879
Speaker 1: was optimized to maximize time users spent on site or engagement,

493
00:28:04,160 --> 00:28:06,800
Speaker 1: it's also happy to route people to content that say

494
00:28:07,000 --> 00:28:10,000
Speaker 1: proposes the existence of a flat earth because those videos

495
00:28:10,040 --> 00:28:14,080
Speaker 1: improve engagement too. Gillen found that searching is the Earth

496
00:28:14,119 --> 00:28:17,840
Speaker 1: flat around and following Google's recommendations sent users to flat

497
00:28:17,840 --> 00:28:20,560
Speaker 1: earth conspiracy videos more than nine percent of the time.

498
00:28:21,000 --> 00:28:23,360
Speaker 1: So if you're wondering why flat earth is taken off

499
00:28:23,400 --> 00:28:26,280
Speaker 1: as a conspiracy, it's because simply asking the question is

500
00:28:26,359 --> 00:28:29,320
Speaker 1: the Earth flatter around? Of the time leads you to

501
00:28:29,400 --> 00:28:33,400
Speaker 1: videos that say it's flat, homie, that's how come all

502
00:28:33,440 --> 00:28:41,160
Speaker 1: those basketball players think the earth this flat? And also what, yeah,

503
00:28:41,280 --> 00:28:45,239
Speaker 1: you can you can see in your head? How like

504
00:28:46,240 --> 00:28:48,800
Speaker 1: that change happens? Is like some guys having a conversation

505
00:28:48,840 --> 00:28:51,960
Speaker 1: with a friend who is kind of dumb and it's like, no, dude,

506
00:28:51,960 --> 00:28:54,080
Speaker 1: you know the Earth flat? And you're like, what that's bullshit?

507
00:28:54,120 --> 00:28:56,200
Speaker 1: And you type, is the Earth flat in the YouTube,

508
00:28:56,560 --> 00:28:58,800
Speaker 1: and then it serves you up before hour documentary about

509
00:28:58,840 --> 00:29:02,760
Speaker 1: how the Earth is flat, and yeah, it's uh, probably

510
00:29:02,760 --> 00:29:07,680
Speaker 1: your first is typing it into YouTube. It's probably not

511
00:29:07,760 --> 00:29:11,040
Speaker 1: the place you want to get that answer. Yeah. No,

512
00:29:11,880 --> 00:29:15,200
Speaker 1: But it's not like schools in America teach people critical

513
00:29:15,240 --> 00:29:17,920
Speaker 1: thinking or how to functionally do research. It's like going

514
00:29:17,960 --> 00:29:21,400
Speaker 1: to Yahoo answers to be like, am I pregnant? Which

515
00:29:21,680 --> 00:29:26,560
Speaker 1: happens all the time. The answer is yes. If you

516
00:29:26,600 --> 00:29:29,760
Speaker 1: are asking whether or not you're pregnant, you are in

517
00:29:29,800 --> 00:29:36,240
Speaker 1: fact pregnant for sure, probably second or third trimester. You

518
00:29:36,320 --> 00:29:39,360
Speaker 1: should you should at least stop smoking for a while

519
00:29:39,440 --> 00:29:42,080
Speaker 1: until you find out for sure. Maybe maybe put down

520
00:29:42,080 --> 00:29:46,920
Speaker 1: a bottle for a second. Yeah. Now. Further reporting using

521
00:29:46,920 --> 00:29:49,800
Speaker 1: additional sources from within Google seems to support most of

522
00:29:49,880 --> 00:29:53,080
Speaker 1: Chaslo's main contentions. In fact, it suggests that he, if anything,

523
00:29:53,200 --> 00:29:57,040
Speaker 1: understated the problem. Chaslo left YouTube in two thousand twelve,

524
00:29:57,400 --> 00:29:59,320
Speaker 1: and while he knew about Google Brain, he did not

525
00:29:59,400 --> 00:30:02,560
Speaker 1: know about an new AI called Reinforce that Google had

526
00:30:02,640 --> 00:30:05,560
Speaker 1: just instituted or institute I think in twos in fifteen

527
00:30:05,640 --> 00:30:08,200
Speaker 1: to YouTube, its existence was revealed by a new York

528
00:30:08,280 --> 00:30:10,760
Speaker 1: Times article published just a few days before I wrote this,

529
00:30:11,080 --> 00:30:14,360
Speaker 1: the making of a YouTube radical. That article claims that

530
00:30:14,440 --> 00:30:17,640
Speaker 1: Reinforce focused on a new kind of machine learning called

531
00:30:17,720 --> 00:30:21,360
Speaker 1: reinforcement learning. The new AI, known as Reinforced, was a

532
00:30:21,440 --> 00:30:24,520
Speaker 1: kind of long term addiction machine. It was designed to

533
00:30:24,640 --> 00:30:28,360
Speaker 1: maximize users engagement over time by predicting which recommendations would

534
00:30:28,400 --> 00:30:30,640
Speaker 1: expand their tastes and get them to watch not just

535
00:30:30,760 --> 00:30:34,480
Speaker 1: one video but many more. Reinforced was a huge success.

536
00:30:34,720 --> 00:30:37,520
Speaker 1: In a talk at an AI conference in February, Men

537
00:30:37,560 --> 00:30:40,480
Speaker 1: mentionin a Google Brain researcher said it was YouTube's most

538
00:30:40,480 --> 00:30:43,760
Speaker 1: successful launch in two years. Site wide views increased by

539
00:30:43,800 --> 00:30:46,360
Speaker 1: nearly one percent. She said a game that at YouTube

540
00:30:46,440 --> 00:30:49,400
Speaker 1: scale can amount to millions more hours of daily watch

541
00:30:49,520 --> 00:30:52,680
Speaker 1: time and millions more dollars in advertising revenue per year.

542
00:30:53,080 --> 00:30:55,360
Speaker 1: She added that the new algorithm was already starting to

543
00:30:55,400 --> 00:30:59,080
Speaker 1: alter users behavior. We can really lead users toward a

544
00:30:59,120 --> 00:31:03,320
Speaker 1: different state versus recommending content that is familiar. Miss Chen said,

545
00:31:04,720 --> 00:31:08,120
Speaker 1: it's another example of like if you take that quote

546
00:31:08,840 --> 00:31:10,920
Speaker 1: out of context and just read it back to her

547
00:31:10,960 --> 00:31:14,920
Speaker 1: and say, ms Chien, this sounds incredibly sinister when you're

548
00:31:14,960 --> 00:31:18,200
Speaker 1: talking about leading people towards a different state, Like, is me,

549
00:31:18,320 --> 00:31:21,520
Speaker 1: ma'am um? Are you in fact a villain? A super

550
00:31:23,000 --> 00:31:27,920
Speaker 1: You sound like a supervillain? Sounds like this might be evil?

551
00:31:28,240 --> 00:31:33,040
Speaker 1: Is this a James body has in the tech industry? Yeah,

552
00:31:33,240 --> 00:31:35,640
Speaker 1: it's that no one ever has in the tech industry.

553
00:31:35,680 --> 00:31:40,120
Speaker 1: That Are we the baddies? Moment We're like, oh, we're

554
00:31:40,120 --> 00:31:44,440
Speaker 1: addicting people to our service? Is that maybe bad? Are

555
00:31:44,480 --> 00:31:50,080
Speaker 1: we the Nazis this whole time? I thought for the Americans? Nope.

556
00:31:51,560 --> 00:31:55,719
Speaker 1: Now YouTube claims that reinforces a good thing, uh, fighting

557
00:31:55,760 --> 00:31:59,280
Speaker 1: YouTube's biased towards popular content and allowing them to provide

558
00:31:59,360 --> 00:32:03,920
Speaker 1: more accurate recommendations. But Reinforce once again presented an opportunity

559
00:32:03,960 --> 00:32:07,120
Speaker 1: for online extremists. They quickly learned that they could throw

560
00:32:07,160 --> 00:32:10,240
Speaker 1: together videos about left wing bias in movies or video games,

561
00:32:10,240 --> 00:32:12,800
Speaker 1: and YouTube would recommend those videos to people who were

562
00:32:12,840 --> 00:32:16,479
Speaker 1: just looking for normal videos about these subjects. As a result,

563
00:32:16,520 --> 00:32:19,480
Speaker 1: extremists were able to red pill viewers by hiding rants

564
00:32:19,520 --> 00:32:22,400
Speaker 1: about the evils of feminism and immigration, as reviews of

565
00:32:22,440 --> 00:32:25,880
Speaker 1: Star Wars. In far right lingo, red pilling refers to

566
00:32:25,920 --> 00:32:27,800
Speaker 1: the first moment that sort of set someone off on

567
00:32:27,840 --> 00:32:32,360
Speaker 1: their journey towards embracing Nazism and so prior to Reinforce,

568
00:32:32,480 --> 00:32:34,560
Speaker 1: if you were looking up I want to see gameplay

569
00:32:34,680 --> 00:32:37,160
Speaker 1: videos about Call of Duty, or I want to see

570
00:32:37,160 --> 00:32:39,320
Speaker 1: your review of Star Wars The Force Awakens, it would

571
00:32:39,360 --> 00:32:42,160
Speaker 1: just take you two reviews and gameplay videos. Now it

572
00:32:42,200 --> 00:32:45,480
Speaker 1: would also take you to somebody talking about like how

573
00:32:45,520 --> 00:32:48,840
Speaker 1: Star Wars is part of the social justice warrior agenda,

574
00:32:48,960 --> 00:32:52,120
Speaker 1: or how Star Wars you know, embraces white genocide or

575
00:32:52,160 --> 00:32:54,440
Speaker 1: something like that, and so then you know, and it

576
00:32:54,480 --> 00:32:56,320
Speaker 1: will recommend that to millions of people, and most of

577
00:32:56,320 --> 00:32:57,840
Speaker 1: them will be like, what the funk is this bullshit?

578
00:32:58,120 --> 00:32:59,920
Speaker 1: But a few thousand of them will be like, oh

579
00:33:00,040 --> 00:33:02,920
Speaker 1: my god, this guy's right, like Star Wars is part

580
00:33:02,960 --> 00:33:05,440
Speaker 1: of a conspiracy to destroy white men, and then they'll

581
00:33:05,440 --> 00:33:08,120
Speaker 1: click on the next video that Stefan Malineu puts out,

582
00:33:08,200 --> 00:33:10,440
Speaker 1: or they'll they'll they'll go deeper down that rabbit hole,

583
00:33:10,680 --> 00:33:14,040
Speaker 1: and that's how this starts happening. Uh. Star Wars is

584
00:33:14,080 --> 00:33:18,280
Speaker 1: a conspiracy, though, just take your money, that's all it is.

585
00:33:18,400 --> 00:33:21,440
Speaker 1: Not to take your money. It's like any other conspiracy

586
00:33:22,080 --> 00:33:25,520
Speaker 1: that involves movies. It's the only thing is to take

587
00:33:25,560 --> 00:33:29,760
Speaker 1: your money. Yeah, not to destroy white people. They want

588
00:33:29,800 --> 00:33:32,280
Speaker 1: white people because white people spend the most money on

589
00:33:32,360 --> 00:33:36,200
Speaker 1: Star Wars. Yeah. If they killed that, that's that's the

590
00:33:36,320 --> 00:33:40,280
Speaker 1: number one customer that's killing your whole customers that a

591
00:33:40,360 --> 00:33:43,800
Speaker 1: cigarette company wants you to breed. Didn't want teenagers to

592
00:33:43,840 --> 00:33:48,240
Speaker 1: start smoking. It's like, yeah, you need to replenish the flocks. Yeah, yeah,

593
00:33:48,280 --> 00:33:51,040
Speaker 1: you you want people to start smoking in their twenties

594
00:33:51,120 --> 00:33:54,240
Speaker 1: as they have children who grow up watching dad smoke. Yes,

595
00:33:54,520 --> 00:33:58,240
Speaker 1: that's like the plans. Yes, yeah, they want they want

596
00:33:58,320 --> 00:34:00,560
Speaker 1: kids like me to grow up who every now and

597
00:34:00,600 --> 00:34:02,800
Speaker 1: then we'll buy a pack of cigarettes just to smell

598
00:34:02,960 --> 00:34:05,200
Speaker 1: the open pack of cigarettes because it takes me back

599
00:34:05,240 --> 00:34:08,880
Speaker 1: to moments in my childhood. It's such a soothing smell,

600
00:34:09,800 --> 00:34:16,000
Speaker 1: unsmoked cigarettes, a little bit sweet, a little bit fruity. Yeah,

601
00:34:16,160 --> 00:34:18,280
Speaker 1: this is going to trigger somebody to buy a cigarette.

602
00:34:18,320 --> 00:34:22,600
Speaker 1: Right now, someone's pulling over to seven eleven. Fuck it yea,

603
00:34:23,400 --> 00:34:25,680
Speaker 1: And I feel terrible about that. And they're like, also,

604
00:34:25,680 --> 00:34:27,759
Speaker 1: I just thought dick pills. They're like, I don't know

605
00:34:27,840 --> 00:34:31,640
Speaker 1: what's happening to me by dick pills. Fucking it's good

606
00:34:31,680 --> 00:34:34,200
Speaker 1: for your health. It's good for your heart. Uh, it's

607
00:34:34,320 --> 00:34:39,160
Speaker 1: it's it's great. Fucking is all benefits. Uh, cigarettes are

608
00:34:39,200 --> 00:34:42,120
Speaker 1: almost all downsides other than the wonderful smell of a

609
00:34:42,120 --> 00:34:46,359
Speaker 1: freshly opened and looking really fucking cool. Yeah, well they

610
00:34:46,360 --> 00:34:51,520
Speaker 1: do make you look incredibly coolblivably cool. So yeah, nothing

611
00:34:51,520 --> 00:34:54,480
Speaker 1: looks cooler than its damn it. Now something does? Smoking

612
00:34:54,480 --> 00:34:57,480
Speaker 1: a joint looks cooler? You're right, smoking a joint does

613
00:34:57,520 --> 00:35:00,640
Speaker 1: look cooler, and the coolest thing of all smoking a

614
00:35:00,719 --> 00:35:06,799
Speaker 1: joint on a unicycle on a yacht. Wow, you just

615
00:35:06,840 --> 00:35:09,680
Speaker 1: took it to another level. I just would want to

616
00:35:09,719 --> 00:35:16,279
Speaker 1: see how good your balance is on. Yeah, one of

617
00:35:16,280 --> 00:35:19,200
Speaker 1: our many millionaire listeners is going to message me tomorrow

618
00:35:19,280 --> 00:35:22,160
Speaker 1: being like, my husband tried to smoke a joint while

619
00:35:22,160 --> 00:35:24,600
Speaker 1: writing a unicycle in our yacht and now he's dead.

620
00:35:24,840 --> 00:35:27,600
Speaker 1: You you killed the love of my life. And or

621
00:35:27,640 --> 00:35:30,879
Speaker 1: we'll get some dope fan art of you on aycle

622
00:35:31,400 --> 00:35:35,080
Speaker 1: smoking a joint on a yacht. Yeah, burning a fat one.

623
00:35:36,000 --> 00:35:39,040
Speaker 1: Speaking of fat ones, The New York Times interviewed a

624
00:35:39,080 --> 00:35:42,319
Speaker 1: young man who was identified in their article on radicalization

625
00:35:42,400 --> 00:35:44,960
Speaker 1: as Mr Kane, and Mr Kane claims that he was

626
00:35:45,000 --> 00:35:47,560
Speaker 1: sucked down one of these far right YouTube rabbit holes

627
00:35:47,560 --> 00:35:50,600
Speaker 1: thanks to YouTube's algorithm. He is scarred by his experience

628
00:35:50,600 --> 00:35:53,320
Speaker 1: of being radicalized by what he calls a decentralized cult

629
00:35:53,360 --> 00:35:56,120
Speaker 1: of far right YouTube personalities who convinced him that Western

630
00:35:56,160 --> 00:35:59,640
Speaker 1: civilization was under threat from Muslim immigrants and cultural Marxists,

631
00:35:59,640 --> 00:36:02,400
Speaker 1: that an eight i Q differences explained racial disparities, and

632
00:36:02,400 --> 00:36:05,520
Speaker 1: that feminism was a dangerous ideology. I just kept falling

633
00:36:05,560 --> 00:36:07,319
Speaker 1: deeper and deeper into this, and it appealed to me

634
00:36:07,360 --> 00:36:09,560
Speaker 1: because it made me feel a sense of belonging, he said,

635
00:36:09,719 --> 00:36:12,719
Speaker 1: I was brainwashed. There's a spectrum on YouTube between the

636
00:36:12,760 --> 00:36:16,279
Speaker 1: calm section, the Walter Cronkite, Carl Sagan part, and crazy town,

637
00:36:16,280 --> 00:36:18,879
Speaker 1: where the extreme stuff is, said Tristan Harris, a former

638
00:36:18,920 --> 00:36:22,520
Speaker 1: design ethicist at Google, YouTube's parent company. If I'm YouTube

639
00:36:22,560 --> 00:36:24,799
Speaker 1: and I want you to watch more, I'm always going

640
00:36:24,840 --> 00:36:29,359
Speaker 1: to steer you toward crazy town. Um, and I will

641
00:36:29,400 --> 00:36:32,800
Speaker 1: say I'm very hard on the on the tech industry

642
00:36:33,120 --> 00:36:36,080
Speaker 1: regularly on this podcast. It speaks well of a lot

643
00:36:36,120 --> 00:36:39,400
Speaker 1: of engineers that the most vocal people in trying to

644
00:36:39,440 --> 00:36:43,440
Speaker 1: fight YouTube's algorithm are former Google engineers who realized what

645
00:36:43,480 --> 00:36:46,640
Speaker 1: the company was doing and like stepped away and have

646
00:36:46,760 --> 00:36:49,239
Speaker 1: been hammering it ever since, being like we made a

647
00:36:49,320 --> 00:36:52,520
Speaker 1: Nazi engine. Guys, like we weren't trying to, but we

648
00:36:52,600 --> 00:36:54,800
Speaker 1: made a Nazi engine, and we have to deal with that.

649
00:36:55,120 --> 00:36:58,000
Speaker 1: Mainly alarm on this one. Yeah, I gotta really gotta

650
00:36:58,000 --> 00:37:00,840
Speaker 1: reign the alarm on this one. You know, so at Google,

651
00:37:00,920 --> 00:37:04,000
Speaker 1: I worked at Google for two years. I didn't know that. Yeah,

652
00:37:04,040 --> 00:37:08,160
Speaker 1: what did you do? I My job title won't explain

653
00:37:08,200 --> 00:37:12,239
Speaker 1: what I did, But basically it was like a quality

654
00:37:12,360 --> 00:37:15,920
Speaker 1: Uh yeah, it has nothing to do with anything. But

655
00:37:15,960 --> 00:37:20,720
Speaker 1: basically I got to um in Russian like help build

656
00:37:21,040 --> 00:37:26,480
Speaker 1: um binary engine that can um well like train it,

657
00:37:26,520 --> 00:37:29,080
Speaker 1: dot build it, train it to be able to tell

658
00:37:29,120 --> 00:37:32,719
Speaker 1: whether something is um a restricted category or not, like

659
00:37:32,800 --> 00:37:35,759
Speaker 1: something is porn or not, gambling or not that kind

660
00:37:35,800 --> 00:37:39,520
Speaker 1: of stuff. So um, yeah it was. It was crazy.

661
00:37:40,680 --> 00:37:44,880
Speaker 1: Well that sounds different. Uh yeah, I saw some of

662
00:37:44,920 --> 00:37:46,759
Speaker 1: the most fucked up stuff on the internet, you know,

663
00:37:47,280 --> 00:37:51,000
Speaker 1: like I've reported child porn before. Then you will have

664
00:37:51,040 --> 00:37:53,640
Speaker 1: a lot to say this latter part, because we do

665
00:37:53,760 --> 00:37:56,480
Speaker 1: talk about content moderators for a little bit. That's kind

666
00:37:56,520 --> 00:37:58,520
Speaker 1: of somebody asking a couple of questions about that at

667
00:37:58,560 --> 00:38:03,200
Speaker 1: the end. Yeah, yeah, yeah, now that that New York

668
00:38:03,200 --> 00:38:06,080
Speaker 1: Times article in full disclosure actually cites me in it

669
00:38:06,120 --> 00:38:08,200
Speaker 1: because of a study that I published with the research

670
00:38:08,239 --> 00:38:11,160
Speaker 1: collective Belling Cat last year where I trawled through hundreds

671
00:38:11,200 --> 00:38:14,440
Speaker 1: and hundreds of leaf conversations between fascist activists and found

672
00:38:14,440 --> 00:38:17,839
Speaker 1: seventy five self reported stories of how these people got

673
00:38:17,920 --> 00:38:21,320
Speaker 1: red pilled. In that study, I found that thirty four

674
00:38:21,440 --> 00:38:24,680
Speaker 1: of the seventy five people I looked at cited YouTube

675
00:38:25,000 --> 00:38:28,120
Speaker 1: videos as the things that red pilled them. Um, I'm

676
00:38:28,160 --> 00:38:30,480
Speaker 1: not the only source on this though. The New York

677
00:38:30,520 --> 00:38:34,719
Speaker 1: Times also cited a research report published by a European

678
00:38:34,760 --> 00:38:38,719
Speaker 1: research firm called vox Poll. They conducted an analysis of

679
00:38:38,760 --> 00:38:42,000
Speaker 1: thirty thousand Twitter accounts affiliated with the far right, and

680
00:38:42,040 --> 00:38:45,000
Speaker 1: they found that those accounts linked to YouTube videos more

681
00:38:45,000 --> 00:38:48,360
Speaker 1: than they linked to any other thing. So there's a

682
00:38:48,400 --> 00:38:52,920
Speaker 1: lot of evidence that YouTube is the primary reason why.

683
00:38:53,239 --> 00:38:55,880
Speaker 1: If you look at people who were researching the KKK

684
00:38:56,040 --> 00:38:58,560
Speaker 1: and neo Nazis in America in two thousand, four tiers

685
00:38:58,600 --> 00:39:02,000
Speaker 1: and five two six big gathering would be twenty people,

686
00:39:02,600 --> 00:39:06,720
Speaker 1: and then in two thousands seventeen four or five hundred

687
00:39:06,760 --> 00:39:09,359
Speaker 1: of them. However many it was showed up at Charlottesville. Like,

688
00:39:10,200 --> 00:39:13,239
Speaker 1: there's a reason their numbers increased so much over a

689
00:39:13,280 --> 00:39:16,560
Speaker 1: pretty short period of time, and it's because these videos

690
00:39:16,560 --> 00:39:19,200
Speaker 1: made more of them. Um. And there's there's a lot

691
00:39:19,200 --> 00:39:21,840
Speaker 1: of evidence of that. So while Google is raking in

692
00:39:21,880 --> 00:39:24,400
Speaker 1: more and more cash and increasing time spent on site,

693
00:39:24,680 --> 00:39:28,160
Speaker 1: they're also increasing the amount of people who think Hitler

694
00:39:28,200 --> 00:39:31,880
Speaker 1: did nothing wrong. Um. And that's that's the tale of today.

695
00:39:32,719 --> 00:39:36,080
Speaker 1: So Mr Kane, the New York Time source for that article,

696
00:39:36,280 --> 00:39:38,759
Speaker 1: claims his journey started in two thousand and fourteen when

697
00:39:38,760 --> 00:39:42,439
Speaker 1: YouTube recommended a self help video by Stefon Malineux. Mr Mallin,

698
00:39:42,480 --> 00:39:44,759
Speaker 1: who was a great candidate for an episode of this podcast.

699
00:39:44,800 --> 00:39:47,719
Speaker 1: But in short, he's a far right YouTube philosopher self

700
00:39:47,719 --> 00:39:50,240
Speaker 1: help grew who advises his listeners to cut all ties

701
00:39:50,280 --> 00:39:53,080
Speaker 1: with their family. He runs a community called Free Domain

702
00:39:53,239 --> 00:39:55,480
Speaker 1: Radio that some people accused of being a cult that

703
00:39:55,600 --> 00:39:59,799
Speaker 1: you know, tells people to cut off contact with their family. Yeah, no,

704
00:40:00,239 --> 00:40:03,440
Speaker 1: cool club is going to be like, hey, please join us,

705
00:40:03,480 --> 00:40:07,240
Speaker 1: but also never speak to anyone you love ever again.

706
00:40:08,080 --> 00:40:10,480
Speaker 1: They never talk to your mom again. Like that's not

707
00:40:10,800 --> 00:40:13,640
Speaker 1: That's not how a cool club starts, you know, That's

708
00:40:13,640 --> 00:40:18,120
Speaker 1: all that's how a cool club starts, that's always bad news. Yeah, yeah,

709
00:40:18,160 --> 00:40:20,640
Speaker 1: cool clubs say, never talked to the cops again, which

710
00:40:20,640 --> 00:40:26,160
Speaker 1: cool clubs absolutely now. Molan New has been on YouTube

711
00:40:26,160 --> 00:40:29,239
Speaker 1: since forever, but his content has radicalized sharply over the years.

712
00:40:29,280 --> 00:40:32,080
Speaker 1: At the beginning, he identified as an anarcho capitalist, and

713
00:40:32,120 --> 00:40:35,520
Speaker 1: he mostly focused on his ideas about how everyone was

714
00:40:35,560 --> 00:40:37,880
Speaker 1: bad at being parents and people should cut ties with

715
00:40:37,920 --> 00:40:41,000
Speaker 1: toxic family members. In recent years, he's made bro, just

716
00:40:41,120 --> 00:40:44,080
Speaker 1: call your dad, call your dad. Bro, you probably need

717
00:40:44,080 --> 00:40:46,719
Speaker 1: to have a convo. Yeah, you guys, probably just talk

718
00:40:46,800 --> 00:40:50,120
Speaker 1: some feelings out. Maybe you'll calm the funk down. I

719
00:40:50,120 --> 00:40:52,359
Speaker 1: don't know, Like I don't want to say, like there's

720
00:40:52,400 --> 00:40:54,520
Speaker 1: actually are a lot of people with toxic family members,

721
00:40:54,520 --> 00:40:56,000
Speaker 1: so they do need to cut out of their lives,

722
00:40:56,000 --> 00:40:58,160
Speaker 1: which I think is part of why Molini was able

723
00:40:58,200 --> 00:41:00,279
Speaker 1: to get a following. Like there's not nothing in what

724
00:41:00,320 --> 00:41:02,040
Speaker 1: he's saying. There's a lot of people who have fucked

725
00:41:02,120 --> 00:41:04,799
Speaker 1: up family backgrounds and you get told like, well, you

726
00:41:04,840 --> 00:41:06,480
Speaker 1: just need to make things right with your mom, And

727
00:41:06,480 --> 00:41:10,040
Speaker 1: it's like, no, if if your mom like sent you

728
00:41:10,080 --> 00:41:13,560
Speaker 1: to gay conversion therapy, maybe cut all ties with her forever.

729
00:41:14,040 --> 00:41:16,040
Speaker 1: I totally agree. No, no, no, I'm not trying to

730
00:41:16,080 --> 00:41:18,400
Speaker 1: say that. What I'm trying to say is that he

731
00:41:18,560 --> 00:41:23,520
Speaker 1: himself uh to pursue a life where you tell people

732
00:41:23,560 --> 00:41:26,920
Speaker 1: to cut contact off with their family. You clearly have

733
00:41:27,040 --> 00:41:31,160
Speaker 1: unresolved issues with your family, and if you resolve those

734
00:41:31,239 --> 00:41:34,399
Speaker 1: by say, calling your parents and talking to them, I'm

735
00:41:34,440 --> 00:41:36,240
Speaker 1: not saying you have to make up with them. I'm saying,

736
00:41:36,280 --> 00:41:39,400
Speaker 1: somehow get closure for yourself so then you don't spend

737
00:41:39,440 --> 00:41:41,600
Speaker 1: the rest of your life trying to get people to

738
00:41:41,680 --> 00:41:48,160
Speaker 1: quit their families. Yeah, that's just like seems yeah, you

739
00:41:48,239 --> 00:41:51,960
Speaker 1: get some ship to deal with bro Um. But you know,

740
00:41:52,000 --> 00:41:56,239
Speaker 1: he Mollynew didn't stay on that sort of thing. Like,

741
00:41:56,760 --> 00:42:00,600
Speaker 1: he made a switch over to pretty hard core nationalism,

742
00:42:00,640 --> 00:42:03,080
Speaker 1: particularly in the last two years. There's like a video

743
00:42:03,120 --> 00:42:06,480
Speaker 1: of him where he's in um uh Poland during like

744
00:42:06,520 --> 00:42:10,520
Speaker 1: a far right march to commemorate like Poland's uh like

745
00:42:10,600 --> 00:42:14,400
Speaker 1: independence day, and he like said, like starts crying and

746
00:42:14,440 --> 00:42:16,640
Speaker 1: has like this big realization of how like I've been

747
00:42:16,680 --> 00:42:19,920
Speaker 1: against nationalism and stuff for years and I realized it

748
00:42:19,920 --> 00:42:22,560
Speaker 1: can really be beautiful and like the unseid things, like

749
00:42:22,600 --> 00:42:26,319
Speaker 1: I realized that white nationalism can be beautiful and that like, uh,

750
00:42:26,360 --> 00:42:30,120
Speaker 1: instead of you know, being an independent libertarian type, I'm

751
00:42:30,120 --> 00:42:33,360
Speaker 1: gonna focus on uh, fighting for my people, which is

752
00:42:33,400 --> 00:42:35,880
Speaker 1: like white people and stuff like that. That's how Stefan

753
00:42:35,960 --> 00:42:39,600
Speaker 1: Molineux is now. Like he's essentially a neo Nazi philosopher

754
00:42:39,640 --> 00:42:41,399
Speaker 1: at this point, and he spends most of his time

755
00:42:41,440 --> 00:42:44,680
Speaker 1: talking about race and i Q and you know, talking

756
00:42:44,719 --> 00:42:47,520
Speaker 1: about how black people are not as good as white people.

757
00:42:47,600 --> 00:42:51,800
Speaker 1: Like That's that's the thrust of modern day Stefal Stefon Molineux.

758
00:42:52,000 --> 00:42:54,600
Speaker 1: He also believes global warming as a hoax, So maybe

759
00:42:54,640 --> 00:42:58,200
Speaker 1: nobody should have much respect for Molineu's own i Q um,

760
00:42:58,239 --> 00:43:00,560
Speaker 1: but a lot of people get turned onto to Fawn's

761
00:43:00,880 --> 00:43:05,520
Speaker 1: straight up fascist propaganda because of their interest in Joe Rogan. Uh.

762
00:43:05,640 --> 00:43:08,120
Speaker 1: Rogan has had stefawn On as a guest several times,

763
00:43:08,120 --> 00:43:11,400
Speaker 1: and YouTube has decided that people who like Rogan should

764
00:43:11,400 --> 00:43:14,719
Speaker 1: have Stefens Channel recommended to them. This maybe why Mr

765
00:43:14,800 --> 00:43:17,719
Speaker 1: Kane saw Mollineu pop into his recommendations, which is what

766
00:43:17,800 --> 00:43:22,080
Speaker 1: he credits as radicalizing him in two thousand and fourteen. Um,

767
00:43:22,160 --> 00:43:25,759
Speaker 1: so yeah, he wound up watching like a lot of

768
00:43:25,800 --> 00:43:29,200
Speaker 1: members of what some people call the intellectual dark Web,

769
00:43:29,280 --> 00:43:32,759
Speaker 1: Joe Rogan, Dave Rubin, guys like Stephen Crowder and of

770
00:43:32,800 --> 00:43:37,719
Speaker 1: course Stefan Molineux uh and over time, like he went

771
00:43:37,920 --> 00:43:41,400
Speaker 1: further and further and further to the right, until eventually

772
00:43:41,440 --> 00:43:44,239
Speaker 1: he starts watching videos by Lauren Southern, who is a

773
00:43:44,320 --> 00:43:48,480
Speaker 1: Canadian activist who's essentially like he called her his fascist crushed,

774
00:43:48,880 --> 00:43:52,920
Speaker 1: like his fashy bay so like. By like two thousand sixteen,

775
00:43:53,440 --> 00:43:56,360
Speaker 1: this guy who starts watching Joe Rogan and like gets

776
00:43:56,360 --> 00:43:59,239
Speaker 1: turned in the Stefan Molinus videos about global warming as

777
00:43:59,239 --> 00:44:02,360
Speaker 1: a hoax and i Q and race. By two sixteen,

778
00:44:02,960 --> 00:44:07,840
Speaker 1: he's like identifying YouTube Nazi as his fascist like crush.

779
00:44:08,200 --> 00:44:10,879
Speaker 1: Like that's how this proceeds for this dude. And that's

780
00:44:10,880 --> 00:44:14,319
Speaker 1: a pretty standard path. But you know what's not a

781
00:44:14,360 --> 00:44:19,160
Speaker 1: standard path? No what the path that our listeners will

782
00:44:19,200 --> 00:44:22,600
Speaker 1: blaze if they buy the products and services that we

783
00:44:22,640 --> 00:44:27,359
Speaker 1: advertise on this program. You seem like your breath has

784
00:44:27,360 --> 00:44:32,840
Speaker 1: been taken away by these skill and ingenuity of that transition. Truly,

785
00:44:32,880 --> 00:44:35,440
Speaker 1: there was nothing I could add. It was a perfect,

786
00:44:35,680 --> 00:44:39,120
Speaker 1: perfect work. I'm the best at this, I'm the best around.

787
00:44:39,239 --> 00:44:41,880
Speaker 1: Nothing's going to ever keep me down. Yeah, I'm not

788
00:44:41,920 --> 00:44:44,399
Speaker 1: able to put a bumper sticker on a Rolls Royce.

789
00:44:44,600 --> 00:44:47,479
Speaker 1: You know what I'm saying. The bumper sticker is gonna

790
00:44:47,520 --> 00:44:53,920
Speaker 1: say I got my genitals molded. Yeah, I want that

791
00:44:53,960 --> 00:44:57,320
Speaker 1: bumper sticker. It's actually any hologram. And like, when you

792
00:44:57,360 --> 00:45:01,320
Speaker 1: look at it one way, I U wearing a skirt,

793
00:45:01,400 --> 00:45:02,799
Speaker 1: and when you look at it the other way, you

794
00:45:02,960 --> 00:45:06,840
Speaker 1: see my vagina mold. It's it's really cool. Man. That

795
00:45:07,080 --> 00:45:10,480
Speaker 1: put a lot of thought into into it. So that's

796
00:45:10,520 --> 00:45:13,520
Speaker 1: quite a bumper sticker. And you know, I have thought

797
00:45:13,560 --> 00:45:15,960
Speaker 1: for a long time that what traffic is missing is

798
00:45:16,120 --> 00:45:21,120
Speaker 1: explicitly pornographic bumper stickers. Like if truck nuts are okay,

799
00:45:21,120 --> 00:45:24,760
Speaker 1: why isn't that Seriously, it's actually a lot more pleasant

800
00:45:24,760 --> 00:45:27,760
Speaker 1: to look at than truck nuts. Yes, yes, nobody actually

801
00:45:27,800 --> 00:45:32,319
Speaker 1: likes truck nuts. Uh no one? All right, Well this

802
00:45:32,360 --> 00:45:44,600
Speaker 1: has been a long digression. Yeah, let's let's products we're back. Boy,

803
00:45:44,640 --> 00:45:50,399
Speaker 1: howdy what a day we've had today? M m so

804
00:45:51,040 --> 00:45:53,760
Speaker 1: at this point, YouTube's roll in radicalizing a whole generation

805
00:45:53,760 --> 00:45:57,719
Speaker 1: of fascists is very well documented, but YouTube is sort

806
00:45:57,760 --> 00:45:59,839
Speaker 1: of stuck when it comes to admitting that they've ever

807
00:46:00,040 --> 00:46:04,680
Speaker 1: done anything wrong. Of their traffic comes from the recommendation engine.

808
00:46:04,880 --> 00:46:08,000
Speaker 1: It is the single thing that drives the platform's profitability

809
00:46:08,040 --> 00:46:10,919
Speaker 1: more than anything else. Back in March, the New York

810
00:46:10,920 --> 00:46:15,360
Speaker 1: Times interviewed Neil Mohan, YouTube's chief product officer. His responses

811
00:46:15,400 --> 00:46:18,080
Speaker 1: were pretty characteristic of what the company says when confronted

812
00:46:18,120 --> 00:46:21,920
Speaker 1: about their little Nazi issue. The interviewer asked, I hear

813
00:46:21,960 --> 00:46:23,839
Speaker 1: a lot about the rabbit hole effect where you start

814
00:46:23,840 --> 00:46:26,400
Speaker 1: watching one video and you get nudged with recommendations towards

815
00:46:26,400 --> 00:46:28,480
Speaker 1: a slightly more extreme video and so on, and all

816
00:46:28,480 --> 00:46:30,880
Speaker 1: of a sudden you're watching something really extreme. Is that

817
00:46:30,920 --> 00:46:35,279
Speaker 1: a real phenomenon, to which Neil responded, Yeah, So I've

818
00:46:35,280 --> 00:46:36,960
Speaker 1: heard this before, and I think that there are some

819
00:46:37,040 --> 00:46:38,600
Speaker 1: myths that go into that description that I think it

820
00:46:38,600 --> 00:46:40,839
Speaker 1: would be useful for me to debunk. The first is

821
00:46:40,840 --> 00:46:42,600
Speaker 1: this notion that it's somehow in our interests for their

822
00:46:42,600 --> 00:46:45,120
Speaker 1: recommendations to shift people in this direction because it boost

823
00:46:45,200 --> 00:46:47,640
Speaker 1: watch time or what have you. I can say categorically

824
00:46:47,640 --> 00:46:50,359
Speaker 1: that's not the way our recommendation systems are designed. Watch

825
00:46:50,440 --> 00:46:52,320
Speaker 1: time is one signal that they use, but they have

826
00:46:52,360 --> 00:46:55,280
Speaker 1: a number of other engagement and satisfaction signals from the user.

827
00:46:55,480 --> 00:46:57,640
Speaker 1: It is not the case that extreme content drives a

828
00:46:57,680 --> 00:46:59,880
Speaker 1: higher version of engagement or watch time than content of

829
00:47:00,040 --> 00:47:03,919
Speaker 1: their types. So he basically has a blanket denial there.

830
00:47:04,239 --> 00:47:08,160
Speaker 1: Uh yeah, that's a huge just like blanket. Now, we

831
00:47:08,200 --> 00:47:12,160
Speaker 1: don't do that. No, that doesn't happen. Doesn't happen, And

832
00:47:12,239 --> 00:47:14,360
Speaker 1: he goes on it's a bit bit of a rambling answer,

833
00:47:14,400 --> 00:47:16,640
Speaker 1: and later in his answer, Mohan called the idea of

834
00:47:16,680 --> 00:47:20,560
Speaker 1: a YouTube radicalization rabbit hole purely a myth. The interviewer,

835
00:47:20,600 --> 00:47:23,080
Speaker 1: to his credit, presses Neil Mohan on this a bit

836
00:47:23,120 --> 00:47:25,560
Speaker 1: more later and asks if he's really sure he wants

837
00:47:25,600 --> 00:47:28,920
Speaker 1: to make that claim. Mohan responds, what I'm saying is

838
00:47:28,960 --> 00:47:30,600
Speaker 1: that when a video is watched, you will see a

839
00:47:30,680 --> 00:47:33,320
Speaker 1: number of videos that are recommended. Some of those videos

840
00:47:33,400 --> 00:47:35,719
Speaker 1: might have the perception of skewing in one direction or

841
00:47:35,760 --> 00:47:38,280
Speaker 1: you know, call it more extreme. There are other videos

842
00:47:38,280 --> 00:47:40,840
Speaker 1: that skew in the opposite direction. And again, our systems

843
00:47:40,840 --> 00:47:42,680
Speaker 1: are not doing this because that's not a signal that

844
00:47:42,680 --> 00:47:45,799
Speaker 1: feeds into the recommendations. That's just the observation that you

845
00:47:45,840 --> 00:47:47,960
Speaker 1: see in the panel. I'm not saying that a user

846
00:47:48,000 --> 00:47:49,839
Speaker 1: couldn't click on one of those videos that are quote

847
00:47:49,880 --> 00:47:52,600
Speaker 1: unquote more extreme, consume that, and then get another set

848
00:47:52,600 --> 00:47:54,640
Speaker 1: of recommendations that sort of keep moving in one path

849
00:47:54,760 --> 00:47:57,319
Speaker 1: or the other. All I'm saying is that it's not inevitable.

850
00:47:58,080 --> 00:48:01,320
Speaker 1: So because everybody doesn't choose to watch more extreme videos,

851
00:48:01,640 --> 00:48:04,759
Speaker 1: there's no YouTube radicalization rabbit hole. Yeah, and also it's

852
00:48:04,880 --> 00:48:07,920
Speaker 1: kind of acknowledging there that it does happen. Yeah, nothing

853
00:48:07,960 --> 00:48:11,360
Speaker 1: is inevitable. I mean except for like death and whatever.

854
00:48:11,480 --> 00:48:14,000
Speaker 1: You know, it's just to be like, yeah, yeah, no,

855
00:48:14,080 --> 00:48:16,759
Speaker 1: it's not. A meteor rite could hit your house before

856
00:48:16,800 --> 00:48:18,360
Speaker 1: you get to click on the video that turns you

857
00:48:18,400 --> 00:48:20,680
Speaker 1: into a Nazi, So of course it's not inevitable. Yeah,

858
00:48:20,960 --> 00:48:24,120
Speaker 1: just be like, it's not a hundred it's not true.

859
00:48:24,560 --> 00:48:27,319
Speaker 1: Is not a good answer when someone A percentage of

860
00:48:27,360 --> 00:48:31,120
Speaker 1: our users die of heart before the next video plays, Yeah,

861
00:48:31,400 --> 00:48:35,520
Speaker 1: pretty high percentage of people. Yeah, that's not what we're

862
00:48:35,560 --> 00:48:38,640
Speaker 1: asking Neil. Now. The reality, of course, is that Neil

863
00:48:38,719 --> 00:48:43,480
Speaker 1: Mohan is uh, shall we say not entirely honest. I

864
00:48:43,480 --> 00:48:45,760
Speaker 1: think I wrote a damn liar in the original draft,

865
00:48:45,800 --> 00:48:48,160
Speaker 1: but I'm not sure where the legally actionable line is.

866
00:48:49,080 --> 00:48:53,560
Speaker 1: Big video, Yeah, pocket of big big video. For just

867
00:48:53,640 --> 00:48:57,680
Speaker 1: one example, Jonathan Albright at Columbia University researcher recently carried

868
00:48:57,680 --> 00:48:59,719
Speaker 1: out a test where he seated a YouTube account with

869
00:48:59,719 --> 00:49:03,440
Speaker 1: a for the phrase crisis actor. The upnext recommendation led

870
00:49:03,520 --> 00:49:08,000
Speaker 1: him to nine thousand different videos promoting crisis actor conspiracy theories.

871
00:49:08,400 --> 00:49:10,640
Speaker 1: So again, someone who heard the term and wanted to

872
00:49:10,640 --> 00:49:13,600
Speaker 1: search for factual information about the conspiracy theory would be

873
00:49:13,640 --> 00:49:17,640
Speaker 1: directed by YouTube two hundreds of hours of conspiratorial nonsense

874
00:49:17,640 --> 00:49:21,040
Speaker 1: about how the Sandy Hook shooting was fake. Now, I'm

875
00:49:21,040 --> 00:49:23,480
Speaker 1: gonna guess you remember last year's mass shooting at the

876
00:49:23,520 --> 00:49:27,760
Speaker 1: Marjorie Stoneman Douglas High School. Um By the Wednesday after

877
00:49:27,800 --> 00:49:30,440
Speaker 1: that shooting, less than a week after all of those

878
00:49:30,520 --> 00:49:33,839
Speaker 1: kids died, the number one trending video on YouTube was

879
00:49:33,960 --> 00:49:37,200
Speaker 1: David Hogg the Actor, which is obviously a video accusing

880
00:49:38,000 --> 00:49:40,120
Speaker 1: one of the kids who's been most prominent of being

881
00:49:40,160 --> 00:49:43,200
Speaker 1: a crisis actor. According to a report from adage, it

882
00:49:43,280 --> 00:49:46,160
Speaker 1: and many others claimed to expose Hogg as a crisis actor.

883
00:49:46,239 --> 00:49:48,800
Speaker 1: YouTube eventually removed that particular video, but not before to

884
00:49:48,880 --> 00:49:52,000
Speaker 1: mass nearly two hundred thousand views. Other videos targeting hog

885
00:49:52,200 --> 00:49:54,920
Speaker 1: remain up. One that appears to show Hogg struggling with

886
00:49:55,000 --> 00:49:57,719
Speaker 1: his words during an interview after the shooting suggests because

887
00:49:57,760 --> 00:50:00,960
Speaker 1: he forgot his lines. YouTube autoso us certain search terms

888
00:50:01,000 --> 00:50:03,000
Speaker 1: that would lead people directly to the clips. If a

889
00:50:03,000 --> 00:50:06,040
Speaker 1: person typed David Hogg and YouTube search bar midday Wednesday,

890
00:50:06,040 --> 00:50:09,400
Speaker 1: for example, some of the suggestions would include exposed and

891
00:50:09,600 --> 00:50:13,839
Speaker 1: crisis actor. When reporters asked YouTube how that video made

892
00:50:13,840 --> 00:50:16,360
Speaker 1: it to the top of their coveted trending chart, YouTube

893
00:50:16,400 --> 00:50:19,320
Speaker 1: explained that since the video included edited clips from a

894
00:50:19,360 --> 00:50:22,200
Speaker 1: CNN report, it's algorithm had believed that it was a

895
00:50:22,280 --> 00:50:25,520
Speaker 1: legitimate piece of journalism and allowed it to spread as

896
00:50:25,560 --> 00:50:29,680
Speaker 1: an authoritative news report would. Um. So again, that's there.

897
00:50:29,760 --> 00:50:32,799
Speaker 1: That's there, like justification, Like we couldn't have known that

898
00:50:32,840 --> 00:50:35,719
Speaker 1: this was fake news because it was fake news that

899
00:50:36,000 --> 00:50:39,200
Speaker 1: used clips from a legitimate news site. So like we're

900
00:50:39,239 --> 00:50:41,040
Speaker 1: clearly not at fault here for the fact that we

901
00:50:41,120 --> 00:50:43,480
Speaker 1: let a robot select all these things and no human

902
00:50:43,480 --> 00:50:46,239
Speaker 1: being watched the top trending video on the side at

903
00:50:46,239 --> 00:50:50,120
Speaker 1: the moment to see if like it was something terrible. Um. Also,

904
00:50:50,239 --> 00:50:55,400
Speaker 1: that's bullshit. Yeah, yeah, that's total bullshit. Now YouTube's or

905
00:50:55,480 --> 00:50:58,120
Speaker 1: Nazi propaganda and conspiracy theories aren't the only things that

906
00:50:58,160 --> 00:51:01,800
Speaker 1: spread like wildfire on YouTube. Of course, pedophilia is also

907
00:51:01,840 --> 00:51:04,840
Speaker 1: a big thing on the site. Yeah. Yeah, this is

908
00:51:04,840 --> 00:51:06,719
Speaker 1: where we get to that part of the story. So

909
00:51:06,760 --> 00:51:09,719
Speaker 1: this broken February of twenty nineteen when a YouTuber named

910
00:51:09,719 --> 00:51:12,200
Speaker 1: Matt Watson put together a video exposing how rings of

911
00:51:12,200 --> 00:51:16,000
Speaker 1: pedophiles had infested the comments sections for various videos featuring

912
00:51:16,080 --> 00:51:20,880
Speaker 1: small children and used them to communicate and trade child porn. Now,

913
00:51:21,360 --> 00:51:24,160
Speaker 1: this report went very viral and immediately prompted several major

914
00:51:24,200 --> 00:51:27,160
Speaker 1: advertisers to pull their money from YouTube. The company released

915
00:51:27,160 --> 00:51:29,680
Speaker 1: a statement to their worried advertisers informing them that they

916
00:51:29,719 --> 00:51:32,640
Speaker 1: had blanket band comments for millions of videos, basically removing

917
00:51:32,640 --> 00:51:35,719
Speaker 1: comments from any videos uploaded by or about young children.

918
00:51:36,360 --> 00:51:38,799
Speaker 1: I'd like to quote from NPRS report on Watson's video,

919
00:51:39,719 --> 00:51:42,640
Speaker 1: Watson describes how he says the pedophile ring works. YouTube

920
00:51:42,680 --> 00:51:45,040
Speaker 1: visitors gather on videos of young girls doing innocuous things

921
00:51:45,040 --> 00:51:47,640
Speaker 1: such as putting on their makeup, demonstrating gymnastics moves, or

922
00:51:47,680 --> 00:51:50,319
Speaker 1: playing twister in the comments section. People would then post

923
00:51:50,400 --> 00:51:52,239
Speaker 1: time stamps that linked to frames in that video that

924
00:51:52,280 --> 00:51:56,240
Speaker 1: appeared to sexualize the children. YouTube's algorithms would then recommend

925
00:51:56,239 --> 00:51:59,200
Speaker 1: other videos also frequented by pedophiles. Once you enter into

926
00:51:59,200 --> 00:52:02,280
Speaker 1: this wormhole, there is now no other content available, Watson said.

927
00:52:03,200 --> 00:52:08,760
Speaker 1: So it might seem at first like this is purely

928
00:52:08,800 --> 00:52:12,600
Speaker 1: an accident on YouTube's part, like that cunning pedophiles figured

929
00:52:12,640 --> 00:52:15,160
Speaker 1: out that there were like they could just find videos

930
00:52:15,160 --> 00:52:18,120
Speaker 1: of young kids doing handstands and stuff and use that

931
00:52:18,160 --> 00:52:20,319
Speaker 1: as porn and trade it with each other, right, which

932
00:52:20,320 --> 00:52:23,359
Speaker 1: would not necessarily be like, how could we have predicted this?

933
00:52:23,440 --> 00:52:25,880
Speaker 1: It's just these people decided to use innocent videos for

934
00:52:25,920 --> 00:52:30,279
Speaker 1: a nefarious purpose. But that's not what happened, or at

935
00:52:30,320 --> 00:52:35,080
Speaker 1: least that's not all of what happened. So in June

936
00:52:35,320 --> 00:52:38,560
Speaker 1: three researchers from Harvard's Berkmann Client Center for Internet and

937
00:52:38,600 --> 00:52:44,279
Speaker 1: Society started combing through YouTube's recommendations for sexually themed videos. Uh.

938
00:52:44,320 --> 00:52:46,680
Speaker 1: They found that starting down this rabbit hole led them

939
00:52:46,680 --> 00:52:50,600
Speaker 1: inevitably to sexual videos that placed greater emphasis on youth.

940
00:52:51,280 --> 00:52:54,680
Speaker 1: So again, that's maybe not super surprising. You start looking

941
00:52:54,680 --> 00:52:57,440
Speaker 1: for sexy videos, you you click on one, and then

942
00:52:57,480 --> 00:52:59,880
Speaker 1: the next video the woman in it or is going

943
00:52:59,920 --> 00:53:01,759
Speaker 1: to be a younger woman, and a younger woman, and

944
00:53:01,800 --> 00:53:04,759
Speaker 1: a younger woman. But then at a certain point, the

945
00:53:04,880 --> 00:53:08,120
Speaker 1: video suggested flipped very suddenly until and I'm going to

946
00:53:08,200 --> 00:53:11,960
Speaker 1: quote the researchers here, YouTube would suddenly begin recommending videos

947
00:53:12,000 --> 00:53:18,839
Speaker 1: of young and partially clothed children. So YouTube would take

948
00:53:18,880 --> 00:53:22,160
Speaker 1: a person who's like just looking for adults like videos

949
00:53:22,200 --> 00:53:25,000
Speaker 1: of like an exact dancer dancing or whatever, like videos

950
00:53:25,040 --> 00:53:28,799
Speaker 1: of attractive young women dancing, and then YouTube would start

951
00:53:28,840 --> 00:53:32,200
Speaker 1: showing them videos of children doing like gymnastics routines and

952
00:53:32,239 --> 00:53:36,080
Speaker 1: stuff like that's the algorithm being like, I bet you'll

953
00:53:36,120 --> 00:53:39,279
Speaker 1: like child porn. Like that's literally what's happening here, which

954
00:53:39,280 --> 00:53:42,800
Speaker 1: I didn't realize when I first heard the story that like,

955
00:53:42,800 --> 00:53:46,759
Speaker 1: like that's YouTube. That's not just pedophiles using YouTube in

956
00:53:46,800 --> 00:53:49,040
Speaker 1: a sleazy way, because pedophiles will always find a way

957
00:53:49,040 --> 00:53:54,960
Speaker 1: to ruin anything that's YouTube crafting new pedophiles. Yeah, it's

958
00:53:54,960 --> 00:53:59,799
Speaker 1: a system, that's Yeah. I wonder if it's like that

959
00:53:59,840 --> 00:54:02,279
Speaker 1: with violence too, if you look up a violent thing,

960
00:54:02,400 --> 00:54:06,040
Speaker 1: if it keeps recommending more violence, because that seems like

961
00:54:06,080 --> 00:54:08,800
Speaker 1: and hate like that would happen when I worked for Google,

962
00:54:08,840 --> 00:54:17,640
Speaker 1: Like the sensitive categories. The restricted categories are you know, violence, hate, gambling, porn, childhorn.

963
00:54:18,680 --> 00:54:20,719
Speaker 1: I think there's even a messed up thing about that,

964
00:54:20,840 --> 00:54:23,680
Speaker 1: because one of the problems that like, people who document

965
00:54:23,719 --> 00:54:26,640
Speaker 1: war crimes in Syria have had is YouTube blanket banning

966
00:54:26,640 --> 00:54:29,480
Speaker 1: their videos because of violence, and then like you have

967
00:54:29,560 --> 00:54:32,040
Speaker 1: evidence of a war crime and then it's wiped off

968
00:54:32,080 --> 00:54:35,520
Speaker 1: of the Internet forever because YouTube doesn't realize that this

969
00:54:35,600 --> 00:54:38,240
Speaker 1: isn't like violence porn, this is somebody trying to document

970
00:54:38,280 --> 00:54:42,520
Speaker 1: a war crime. Um. It's made it really hard to

971
00:54:42,600 --> 00:54:46,800
Speaker 1: do that kind of research. It's yeah, their response is

972
00:54:46,840 --> 00:54:51,400
Speaker 1: always so terrible. Um. Anyway, The New York Times reported quote, So,

973
00:54:51,440 --> 00:54:54,040
Speaker 1: a user who watches erotic videos might be recommended videos

974
00:54:54,040 --> 00:54:56,600
Speaker 1: of women who becomes conspicuously younger, and then women who

975
00:54:56,640 --> 00:55:00,319
Speaker 1: promote provocatively in children's clothes. Eventually, some user might be

976
00:55:00,320 --> 00:55:02,239
Speaker 1: presented with videos of girls as young as five or

977
00:55:02,239 --> 00:55:04,840
Speaker 1: six wearing bathing suits, are getting dressed or doing a split.

978
00:55:05,920 --> 00:55:08,759
Speaker 1: So yeah, and its eternal quest to increase time spent

979
00:55:08,840 --> 00:55:13,839
Speaker 1: on site, YouTube's algorithm essentially radicalized people towards pedophilia. And

980
00:55:13,880 --> 00:55:18,560
Speaker 1: to make matters worse, it wasn't just picking sexy videos

981
00:55:18,560 --> 00:55:20,880
Speaker 1: like that people had uploaded with the intent of them

982
00:55:20,920 --> 00:55:24,880
Speaker 1: being sexy. Because it was sending children's videos to people,

983
00:55:25,200 --> 00:55:28,880
Speaker 1: it started grabbing totally normal home videos of little kids

984
00:55:29,239 --> 00:55:32,000
Speaker 1: and presenting those videos to horny adults who were on

985
00:55:32,040 --> 00:55:35,960
Speaker 1: YouTube to masturbate. The report suggests it was learning from

986
00:55:36,080 --> 00:55:39,280
Speaker 1: users who sought out revealing or suggestive images of children.

987
00:55:39,760 --> 00:55:42,440
Speaker 1: One parent The Times talked with related in horror that

988
00:55:42,480 --> 00:55:44,279
Speaker 1: a video of her ten year old girl wearing a

989
00:55:44,320 --> 00:55:48,200
Speaker 1: bathing suit had reached four hundred thousand views. So, like,

990
00:55:48,280 --> 00:55:51,520
Speaker 1: parents start to realize, like, wait, an I uploaded this

991
00:55:51,600 --> 00:55:53,719
Speaker 1: video to show her grandma. They're supposed to be like

992
00:55:53,800 --> 00:55:56,600
Speaker 1: nine views on this thing. Why have four hundred thousand

993
00:55:56,640 --> 00:55:59,279
Speaker 1: people watched this video of my ten year old? And

994
00:55:59,280 --> 00:56:02,160
Speaker 1: it's because you tube is trying to provide people with

995
00:56:02,200 --> 00:56:03,840
Speaker 1: porn because it knows that will keep them on the

996
00:56:03,840 --> 00:56:09,839
Speaker 1: site longer. That's fucking wild. Yeah. After this report came out,

997
00:56:09,880 --> 00:56:13,960
Speaker 1: YouTube published an apologetic blog post, promising that responsibility is

998
00:56:13,960 --> 00:56:16,960
Speaker 1: our number one priority. In chief among our areas of

999
00:56:17,000 --> 00:56:21,040
Speaker 1: focus is protecting miners and families. But of course that's

1000
00:56:21,040 --> 00:56:23,720
Speaker 1: not true. Increasing the amount of time spent on site

1001
00:56:23,800 --> 00:56:27,400
Speaker 1: is YouTube's chief priority, or rather, making money is YouTube's

1002
00:56:27,440 --> 00:56:29,719
Speaker 1: chief priority, and if increasing the amount of time spent

1003
00:56:29,760 --> 00:56:31,640
Speaker 1: on site is the best way to make money, the

1004
00:56:31,680 --> 00:56:35,800
Speaker 1: YouTube will prioritize that overall other things, including the safety

1005
00:56:35,880 --> 00:56:39,840
Speaker 1: of children. Now, there are ways YouTube could reduce the

1006
00:56:39,920 --> 00:56:42,719
Speaker 1: danger their sites present to the world. Ways they could

1007
00:56:42,760 --> 00:56:45,560
Speaker 1: catch stuff like propaganda accusing a mass shooting victim of

1008
00:56:45,560 --> 00:56:48,040
Speaker 1: being an actor, or people's home movies of being accidentally

1009
00:56:48,040 --> 00:56:50,560
Speaker 1: turned into child porn. Even if they're not going to

1010
00:56:50,600 --> 00:56:55,000
Speaker 1: stop hosting literal, fascist propaganda. Content moderators could add human

1011
00:56:55,040 --> 00:56:57,840
Speaker 1: eyes and human oversight to an AI algorithm that is

1012
00:56:57,880 --> 00:57:02,160
Speaker 1: clearly sociopathic. And earlier this year, YouTube did announce that

1013
00:57:02,160 --> 00:57:05,760
Speaker 1: they were expanding their content moderator team to ten thousand people,

1014
00:57:06,040 --> 00:57:08,600
Speaker 1: which sounds great. Sounds like a huge number of people,

1015
00:57:09,000 --> 00:57:11,479
Speaker 1: Only that's not as good as it seems. The Wall

1016
00:57:11,480 --> 00:57:14,120
Speaker 1: Street Journal investigated and found out that a huge number

1017
00:57:14,160 --> 00:57:17,680
Speaker 1: of these moderators, perhaps the majority, worked in cubical farms

1018
00:57:17,680 --> 00:57:20,120
Speaker 1: and India and the Philippines, which would be fine if

1019
00:57:20,160 --> 00:57:23,640
Speaker 1: they were moderating content posted from India or the Philippines,

1020
00:57:23,920 --> 00:57:25,680
Speaker 1: but of course these people were also going to be

1021
00:57:25,720 --> 00:57:31,680
Speaker 1: tasked with monitoring American political content now alphabet. Google does

1022
00:57:31,680 --> 00:57:36,000
Speaker 1: not disclose how much money YouTube makes. Estimates suggest that

1023
00:57:36,040 --> 00:57:38,720
Speaker 1: it's around ten billion dollars a year and maybe increasing

1024
00:57:38,760 --> 00:57:42,000
Speaker 1: by as much as forty per year. Math is not

1025
00:57:42,080 --> 00:57:44,560
Speaker 1: my strong suit. I'm not an algorithm, but I did

1026
00:57:45,160 --> 00:57:47,440
Speaker 1: a little bit of math, and I calculated that if

1027
00:57:47,480 --> 00:57:50,720
Speaker 1: Google took a billion dollars of their profit and hired

1028
00:57:50,720 --> 00:57:54,440
Speaker 1: new content moderators paying them fifty thou dollar a year salaries,

1029
00:57:54,440 --> 00:57:55,840
Speaker 1: which I meant to guess is more than most of

1030
00:57:55,840 --> 00:57:59,080
Speaker 1: these moderators get, they could afford to hire twenty thousand

1031
00:57:59,160 --> 00:58:03,760
Speaker 1: new moderators, tripling their current capacity. Realistically, they could hire

1032
00:58:03,800 --> 00:58:06,680
Speaker 1: fifty or sixty thousand more moderators and still be making

1033
00:58:06,720 --> 00:58:08,680
Speaker 1: billions of dollars a year and one of the most

1034
00:58:08,680 --> 00:58:12,160
Speaker 1: profitable services on the Internet. But doing that would mean

1035
00:58:12,240 --> 00:58:15,320
Speaker 1: less profit for Google shareholders. It would mean less money

1036
00:58:15,400 --> 00:58:17,520
Speaker 1: for people like Neil Mohan, the man who has been

1037
00:58:17,600 --> 00:58:20,640
Speaker 1: YouTube's chief product officers since two thousand eleven, the man

1038
00:58:20,680 --> 00:58:23,160
Speaker 1: who was overseen nearly all the algorithmic changes we are

1039
00:58:23,160 --> 00:58:25,640
Speaker 1: talking about today, The man who sat down with The

1040
00:58:25,640 --> 00:58:28,000
Speaker 1: New York Times and denied YouTube had a problem with

1041
00:58:28,080 --> 00:58:31,640
Speaker 1: leading people down rabbit holes that radicalize them in dangerous ways.

1042
00:58:32,320 --> 00:58:35,240
Speaker 1: I was kind of curious as to how well compensated

1043
00:58:35,320 --> 00:58:39,080
Speaker 1: Mr Mohan is, so I googled Neil Mohan net worth.

1044
00:58:39,520 --> 00:58:42,920
Speaker 1: The first response was a Business Insider article. Google paid

1045
00:58:42,960 --> 00:58:46,960
Speaker 1: this man a hundred million dollars. Here's his story. So

1046
00:58:47,040 --> 00:58:51,240
Speaker 1: that's cool. Yeah, ooof, And I can tell you from

1047
00:58:51,280 --> 00:58:55,400
Speaker 1: being a moderator, Um, I worked on a team where

1048
00:58:55,400 --> 00:58:57,600
Speaker 1: everybody did what I did in a different language. So

1049
00:58:57,640 --> 00:58:59,520
Speaker 1: I did this in Russian and next to me with

1050
00:58:59,560 --> 00:59:03,160
Speaker 1: someone who doing it in Chinese and Turkish and all

1051
00:59:03,160 --> 00:59:06,160
Speaker 1: the all of the languages, I mean not all, but

1052
00:59:06,680 --> 00:59:09,880
Speaker 1: a significant number. Yeah, and um, I can tell you

1053
00:59:09,920 --> 00:59:14,520
Speaker 1: that we were hired as contractors for only a year.

1054
00:59:14,880 --> 00:59:17,400
Speaker 1: Very rarely would you ever be doing a second year

1055
00:59:17,440 --> 00:59:20,120
Speaker 1: because they didn't want to pay you with the full benefits,

1056
00:59:21,000 --> 00:59:24,440
Speaker 1: like you know, you don't get health insurance and whatever,

1057
00:59:24,520 --> 00:59:26,440
Speaker 1: all the perks that you would get from being a

1058
00:59:26,480 --> 00:59:29,200
Speaker 1: full time Google employee. And the thing about what we

1059
00:59:29,240 --> 00:59:31,640
Speaker 1: did is you got exposed to a lot of fucked

1060
00:59:31,720 --> 00:59:35,680
Speaker 1: up stuff, Like you know, the videos and stuff that

1061
00:59:35,720 --> 00:59:38,000
Speaker 1: I've seen are like some of the worst the internet

1062
00:59:38,040 --> 00:59:41,320
Speaker 1: has to offer, like beheadings or someone stomping a kitten

1063
00:59:41,360 --> 00:59:44,680
Speaker 1: to death and high heels like crazy shit, and it

1064
00:59:44,720 --> 00:59:46,720
Speaker 1: would really make you sick. And they like give you

1065
00:59:46,760 --> 00:59:49,000
Speaker 1: free food at Google, and you like wouldn't be able

1066
00:59:49,000 --> 00:59:51,360
Speaker 1: to eat sometimes because you would be so grossed out.

1067
00:59:51,640 --> 00:59:54,240
Speaker 1: And it's not like they that's why you're only there

1068
00:59:54,240 --> 00:59:56,800
Speaker 1: for a year. Also, not just that you wouldn't be

1069
00:59:56,840 --> 01:00:00,240
Speaker 1: able to get full benefits, but also because they are

1070
01:00:00,320 --> 01:00:04,800
Speaker 1: okay with wasting your mental and physical energies and then

1071
01:00:04,920 --> 01:00:07,720
Speaker 1: letting you go and then just cycling through new people

1072
01:00:07,840 --> 01:00:12,200
Speaker 1: every year because um, rather than investing you know, and

1073
01:00:12,320 --> 01:00:16,240
Speaker 1: employees that are full time making sure they have uh,

1074
01:00:17,240 --> 01:00:21,360
Speaker 1: you know, access to mental health care and stuff like that,

1075
01:00:22,000 --> 01:00:26,520
Speaker 1: and um, you know, making that job be something that

1076
01:00:26,560 --> 01:00:30,800
Speaker 1: they take more seriously considering how important it is. Well,

1077
01:00:30,840 --> 01:00:33,200
Speaker 1: and that's part of what's really messed up is that

1078
01:00:33,240 --> 01:00:36,080
Speaker 1: like it's fucking Google. Like if you go into the

1079
01:00:36,080 --> 01:00:39,760
Speaker 1: people people who are like actually coding these algorithms and stuff,

1080
01:00:39,920 --> 01:00:42,680
Speaker 1: I guarantee you those people have on site therapists, they

1081
01:00:42,720 --> 01:00:46,000
Speaker 1: can visit, they have gyms at work, they get their lunches.

1082
01:00:46,200 --> 01:00:48,720
Speaker 1: I mean we all worked on the same building, but

1083
01:00:48,800 --> 01:00:50,720
Speaker 1: like I can't you know, I couldn't go get a

1084
01:00:50,760 --> 01:00:53,760
Speaker 1: free massage during It's like, you know, you have a

1085
01:00:53,800 --> 01:00:56,760
Speaker 1: CrossFit trainer on site and ship like that for sure

1086
01:00:56,760 --> 01:00:59,600
Speaker 1: you get incredible perks. And the whole point is what

1087
01:00:59,760 --> 01:01:01,760
Speaker 1: I that was kind of ironic about what we were saying,

1088
01:01:01,840 --> 01:01:04,200
Speaker 1: is like the whole thing is to try to make

1089
01:01:04,240 --> 01:01:07,920
Speaker 1: you stay on YouTube. But when you work for company

1090
01:01:07,960 --> 01:01:09,680
Speaker 1: like Google, their job is to try to make you

1091
01:01:09,720 --> 01:01:13,120
Speaker 1: stay at Google. So you know, the reason you're getting

1092
01:01:13,120 --> 01:01:16,560
Speaker 1: all these benefits and stuff and like free food and

1093
01:01:17,040 --> 01:01:19,440
Speaker 1: jim and massage and whatever is because they want you

1094
01:01:19,480 --> 01:01:23,080
Speaker 1: to stay and work forever. But they don't want you

1095
01:01:23,480 --> 01:01:28,120
Speaker 1: like that stuff to me exactly like and that's that's

1096
01:01:28,120 --> 01:01:30,760
Speaker 1: a very telling thing from Google's perspective, because they are

1097
01:01:30,760 --> 01:01:33,320
Speaker 1: saying that increasing the amount of the people who are

1098
01:01:33,320 --> 01:01:35,840
Speaker 1: coding these algorithms that increase the amount of time people

1099
01:01:35,840 --> 01:01:38,520
Speaker 1: spend on site, that is important to us. And so

1100
01:01:38,560 --> 01:01:40,880
Speaker 1: we will do whatever it takes to retain these people.

1101
01:01:41,280 --> 01:01:44,640
Speaker 1: But the people who make sure that we aren't creating

1102
01:01:44,720 --> 01:01:47,280
Speaker 1: new pedophiles while we make money, the people who are

1103
01:01:47,280 --> 01:01:51,120
Speaker 1: responsible for making sure that Nazi propaganda isn't served up

1104
01:01:51,160 --> 01:01:55,480
Speaker 1: to like influence able young children via our service. Those

1105
01:01:55,520 --> 01:01:58,520
Speaker 1: people aren't valuable to us because we don't care about that,

1106
01:01:58,760 --> 01:02:02,120
Speaker 1: so we're not going to offer them healthcare. Like if

1107
01:02:02,160 --> 01:02:04,520
Speaker 1: if if Google really was an ethical company, and if

1108
01:02:04,560 --> 01:02:07,800
Speaker 1: YouTube cared about its impact on the world, someone whose

1109
01:02:07,920 --> 01:02:11,520
Speaker 1: job who's there's nothing less technical or less valuable about

1110
01:02:11,800 --> 01:02:14,920
Speaker 1: what you're doing. Being able to speak another language fluently,

1111
01:02:15,000 --> 01:02:18,200
Speaker 1: being able to understand if content propagating on their site

1112
01:02:18,800 --> 01:02:21,640
Speaker 1: uh is toxic or not. That's a very difficult, very

1113
01:02:21,640 --> 01:02:26,040
Speaker 1: technical task. If they cared about the impact they had

1114
01:02:26,080 --> 01:02:28,440
Speaker 1: on the world, the people doing that job would be

1115
01:02:28,440 --> 01:02:31,080
Speaker 1: well paid and would have benefits and would be seen

1116
01:02:31,160 --> 01:02:34,440
Speaker 1: as a crucial aspect of the company. But instead it's

1117
01:02:34,480 --> 01:02:36,880
Speaker 1: sort of like, if we don't have someone doing this job,

1118
01:02:36,920 --> 01:02:38,840
Speaker 1: will get yelled at. So we're going to do the

1119
01:02:38,920 --> 01:02:41,520
Speaker 1: minimum necessary and we're going to have most of the

1120
01:02:41,520 --> 01:02:44,160
Speaker 1: people doing that job be working at a fucking cube

1121
01:02:44,160 --> 01:02:47,920
Speaker 1: farm in India, even though we're expecting them to moderate

1122
01:02:47,960 --> 01:02:50,720
Speaker 1: American content and to understand all of our cultural nuances

1123
01:02:50,720 --> 01:02:55,080
Speaker 1: and whether or not something's toxic, Like that's so fucked up.

1124
01:02:55,400 --> 01:03:01,280
Speaker 1: And also considering the fact that like add is the

1125
01:03:01,440 --> 01:03:06,040
Speaker 1: reason that they hire content moderators, not because they care

1126
01:03:06,080 --> 01:03:09,320
Speaker 1: about the content necessarily. It's that it would be a

1127
01:03:09,400 --> 01:03:14,320
Speaker 1: huge like mistake if say, an ad for Huggies was

1128
01:03:14,400 --> 01:03:18,280
Speaker 1: served on a diaper fetish website. You know, they want

1129
01:03:18,800 --> 01:03:22,880
Speaker 1: something in place where the page knows, the algorithm knows

1130
01:03:22,960 --> 01:03:25,200
Speaker 1: not to serve that, even though it seems like a

1131
01:03:25,200 --> 01:03:27,840
Speaker 1: good match because the word diapers repeated and blah blah blah,

1132
01:03:27,880 --> 01:03:30,120
Speaker 1: you know what I mean. So it's really less about

1133
01:03:30,320 --> 01:03:34,360
Speaker 1: it's it's more about keeping the advertisers happy and making

1134
01:03:34,400 --> 01:03:39,160
Speaker 1: the most money than it is about, yeah, ensuring that

1135
01:03:39,200 --> 01:03:43,680
Speaker 1: there Internet is this less fucked up place. This gets

1136
01:03:44,280 --> 01:03:45,560
Speaker 1: to one of the things like when I when I

1137
01:03:45,560 --> 01:03:48,240
Speaker 1: get in arguments with people about the nature of capitalism

1138
01:03:48,320 --> 01:03:50,520
Speaker 1: and what's what's what's wrong with the kind of capitalism

1139
01:03:50,560 --> 01:03:52,920
Speaker 1: that we have in this country. I think a lot

1140
01:03:52,920 --> 01:03:56,920
Speaker 1: of people who like just sort of reject anti capitalist

1141
01:03:57,040 --> 01:03:59,800
Speaker 1: arguments out of hand do it because they think that

1142
01:03:59,840 --> 01:04:01,959
Speaker 1: you're saying, oh, it's just wrong to make to make money.

1143
01:04:02,040 --> 01:04:04,520
Speaker 1: It's wrong to have a business that like makes a profit.

1144
01:04:04,600 --> 01:04:07,880
Speaker 1: Is like the issue, isn't that the issue? Like this

1145
01:04:07,920 --> 01:04:12,120
Speaker 1: company Google could be making billions of dollars a year,

1146
01:04:12,920 --> 01:04:15,240
Speaker 1: uh and still be one of the most profitable sites

1147
01:04:15,400 --> 01:04:18,320
Speaker 1: in in it's of its type still make a huge

1148
01:04:18,320 --> 01:04:20,920
Speaker 1: amount of money and have three times as many people

1149
01:04:21,640 --> 01:04:24,240
Speaker 1: doing content moderation, and all those people have health care.

1150
01:04:24,840 --> 01:04:28,240
Speaker 1: But by cutting corners on that part of it, because

1151
01:04:28,240 --> 01:04:30,360
Speaker 1: it doesn't make them more money, it just makes the

1152
01:04:30,360 --> 01:04:33,760
Speaker 1: world better. They make more money, and it's worth more

1153
01:04:33,800 --> 01:04:37,400
Speaker 1: to them to increase the value of a few hundred

1154
01:04:37,440 --> 01:04:40,600
Speaker 1: people stock than to ensure that there aren't thousands of

1155
01:04:40,640 --> 01:04:44,760
Speaker 1: additional people masturbating to children like that. That's that's what

1156
01:04:44,920 --> 01:04:47,520
Speaker 1: I have an issue with with capitalism, Like, that's that's

1157
01:04:47,520 --> 01:04:50,840
Speaker 1: a you can make a profit without also selling your

1158
01:04:50,840 --> 01:04:55,880
Speaker 1: fucking soul. Yeah, we could have YouTube should be banned,

1159
01:04:56,120 --> 01:05:00,520
Speaker 1: like I can get recommended new musicians that I like.

1160
01:05:01,560 --> 01:05:05,439
Speaker 1: We can all watch videos to masturbate too, without more

1161
01:05:05,480 --> 01:05:08,400
Speaker 1: people being turned toto pedophiles and Nazis. That's not a

1162
01:05:08,480 --> 01:05:11,800
Speaker 1: necessary part of this, Like it's just because corners are

1163
01:05:11,800 --> 01:05:15,960
Speaker 1: being cut. Yeah, it just shows what the value what

1164
01:05:16,120 --> 01:05:19,360
Speaker 1: the value of our society is, what the values of

1165
01:05:19,360 --> 01:05:22,720
Speaker 1: our society are. Yeah, it's they've literally said, like three

1166
01:05:22,720 --> 01:05:25,439
Speaker 1: billion dollars a year is worth more to us than

1167
01:05:26,800 --> 01:05:30,360
Speaker 1: god knows how many children being molested than fucking Heather

1168
01:05:30,480 --> 01:05:34,920
Speaker 1: higher getting run down at Charlottesville than they're being Nazis

1169
01:05:34,960 --> 01:05:38,600
Speaker 1: marching through the streets and advocating the extermination of black people,

1170
01:05:38,640 --> 01:05:41,760
Speaker 1: of LGBT, people of whatever like, which is again part

1171
01:05:41,760 --> 01:05:44,400
Speaker 1: of why so many Google employees are now speaking out

1172
01:05:44,400 --> 01:05:46,720
Speaker 1: and horrified because like, they're not monsters. They don't want

1173
01:05:46,760 --> 01:05:48,400
Speaker 1: to live in this world anymore than the rest of

1174
01:05:48,480 --> 01:05:51,040
Speaker 1: us do. They just didn't realize what was happening because

1175
01:05:51,040 --> 01:05:54,120
Speaker 1: they were busy focusing on the code and the free massages,

1176
01:05:54,640 --> 01:05:56,320
Speaker 1: and then, like the rest of us, they woke up

1177
01:05:56,360 --> 01:06:05,280
Speaker 1: to a world full of Nazis and pedophiles. Uh. Yeah,

1178
01:06:05,320 --> 01:06:07,520
Speaker 1: I feel like you're looking at me to make a

1179
01:06:07,600 --> 01:06:10,320
Speaker 1: joke now, and I feel like I don't know, this

1180
01:06:10,440 --> 01:06:16,040
Speaker 1: got real serious. I'm more just tired. We're all tired.

1181
01:06:17,280 --> 01:06:22,640
Speaker 1: It's a it's a very tiring world we live in. Yeah, well,

1182
01:06:22,680 --> 01:06:28,120
Speaker 1: Sophia that you want to plug your plug? Fuck? I

1183
01:06:28,160 --> 01:06:30,440
Speaker 1: mean not really just want everyone to go and get

1184
01:06:30,480 --> 01:06:34,800
Speaker 1: a hug. You know, everybody go get a hug, but Jesus, Yeah,

1185
01:06:34,920 --> 01:06:40,240
Speaker 1: but also, uh, I am to be found on the

1186
01:06:40,280 --> 01:06:43,920
Speaker 1: sites we hate, you know, what a fun thing to plug.

1187
01:06:44,240 --> 01:06:53,600
Speaker 1: I'm available Twitter and Instagram. That's Sophia and I have

1188
01:06:53,640 --> 01:06:56,120
Speaker 1: a podcast about love and sexuality around the world that

1189
01:06:56,200 --> 01:06:59,360
Speaker 1: I co host with Courtney Kosak. It's called Private Parts Unknown.

1190
01:07:00,120 --> 01:07:05,160
Speaker 1: Check that out. Check out Private Parts Unknown. I'm also

1191
01:07:05,240 --> 01:07:09,360
Speaker 1: on the sites we hate. Behind the Bastards dot com

1192
01:07:09,440 --> 01:07:10,760
Speaker 1: is not a site that we hate, but it's where

1193
01:07:10,800 --> 01:07:13,920
Speaker 1: you can find the sources for this episode. Um, I right, okay,

1194
01:07:14,000 --> 01:07:15,400
Speaker 1: is where you can find me on Twitter. You can

1195
01:07:15,400 --> 01:07:20,360
Speaker 1: find us on Twitter and Instagram at bastards pod. Um,

1196
01:07:20,400 --> 01:07:23,360
Speaker 1: that's the You buy t shirts on t public dot com.

1197
01:07:23,400 --> 01:07:31,040
Speaker 1: Behind the Bastards. Uh, yep, that's the episode. Go uh go,

1198
01:07:31,040 --> 01:07:33,920
Speaker 1: go find YouTube's headquarters and yell at them, scream at

1199
01:07:33,960 --> 01:07:37,520
Speaker 1: their sign, take pictures of their company, and wave your fists. Uh.

1200
01:07:39,080 --> 01:07:44,640
Speaker 1: If you work at YouTube, quit it's not worth it.

1201
01:07:44,800 --> 01:07:48,040
Speaker 1: I mean, the more whistleblowers the better. Yeah. Yeah, quit

1202
01:07:48,120 --> 01:07:50,280
Speaker 1: and go talk to the New York Times or some

1203
01:07:50,280 --> 01:07:56,000
Speaker 1: some fucking buddy. Uh yeah. Also, um, one random thing

1204
01:07:56,040 --> 01:07:58,280
Speaker 1: that's positive is if you want there's a lot of

1205
01:07:58,360 --> 01:08:01,800
Speaker 1: videos of trains on YouTube. Have discovered of just trains

1206
01:08:01,840 --> 01:08:06,320
Speaker 1: passing by trains and fails. Yeah. I think you will

1207
01:08:06,360 --> 01:08:08,760
Speaker 1: find it very soothing. First, they'll be like, what the

1208
01:08:08,760 --> 01:08:11,280
Speaker 1: funk a video of a train that's twelve minutes long.

1209
01:08:11,640 --> 01:08:15,800
Speaker 1: Guess what that will soothe you? Soothe your ass, or

1210
01:08:15,840 --> 01:08:18,519
Speaker 1: if you're more like me, watch videos of people skiing

1211
01:08:18,920 --> 01:08:21,479
Speaker 1: and then failing to ski. I mean that's if you

1212
01:08:21,479 --> 01:08:28,519
Speaker 1: want to laugh. Yeah, yeah, I feel like YouTube's algorithm

1213
01:08:28,600 --> 01:08:30,559
Speaker 1: is going to take you from train videos to train

1214
01:08:30,640 --> 01:08:36,040
Speaker 1: fails really fast. Um oh boy, yeah ship. I don't

1215
01:08:36,040 --> 01:08:40,240
Speaker 1: know now that, now that I know about the the

1216
01:08:40,360 --> 01:08:43,040
Speaker 1: rabbit hole, I'm afraid that there's a way to connect

1217
01:08:43,120 --> 01:08:46,800
Speaker 1: trains to children that I have not thought. Oh no,

1218
01:08:47,040 --> 01:08:49,360
Speaker 1: I'm not even gonna make any further comments. We should

1219
01:08:49,360 --> 01:08:51,519
Speaker 1: get before this gets to a spot