1
00:00:02,240 --> 00:00:04,960
Speaker 1: This is Alec Baldwin and you're listening to Hear's the

2
00:00:05,000 --> 00:00:10,360
Speaker 1: thing from iHeartRadio. Modern day society has been living in

3
00:00:10,360 --> 00:00:14,120
Speaker 1: some form of the computer era for decades. An entire

4
00:00:14,160 --> 00:00:17,119
Speaker 1: genre of films has been dedicated to showing what human

5
00:00:17,160 --> 00:00:21,400
Speaker 1: life may be like once computers become conscious. My guests

6
00:00:21,440 --> 00:00:24,439
Speaker 1: today are two trailblazers in a field that many have

7
00:00:24,520 --> 00:00:28,360
Speaker 1: speculated could potentially risk the end of human civilization as

8
00:00:28,400 --> 00:00:31,680
Speaker 1: we know it. That is, of course, the true birth

9
00:00:31,680 --> 00:00:35,880
Speaker 1: of artificial intelligence, known simply as AI. And if you

10
00:00:35,960 --> 00:00:39,320
Speaker 1: think the world has become inundated with AI technology overnight,

11
00:00:39,760 --> 00:00:43,200
Speaker 1: you are not alone. AI based products come in many

12
00:00:43,240 --> 00:00:46,559
Speaker 1: different forms, such as chatbots, which can be used for

13
00:00:46,640 --> 00:00:50,040
Speaker 1: everything from finding a list of local Italian restaurants to

14
00:00:50,120 --> 00:00:54,840
Speaker 1: drafting legal documents and writing college essays. AI can also

15
00:00:54,880 --> 00:00:57,720
Speaker 1: be used to manipulate a person's likeness on video or

16
00:00:57,760 --> 00:01:00,720
Speaker 1: the sound of their voice. And if you're wondering just

17
00:01:00,800 --> 00:01:04,880
Speaker 1: how good computers have become at mimicking humans by narration

18
00:01:05,040 --> 00:01:07,720
Speaker 1: thus far has been generated by text written into a

19
00:01:07,800 --> 00:01:13,360
Speaker 1: program called descript. Okay, I can't let the machines take

20
00:01:13,400 --> 00:01:18,400
Speaker 1: over just yet. This is actually human. Alec Baldwin. Later

21
00:01:18,440 --> 00:01:22,119
Speaker 1: in the program, I'll speak with Blake Lemoyne, a software

22
00:01:22,160 --> 00:01:26,480
Speaker 1: engineer formerly employed at Google. Lemoyne was part of the

23
00:01:26,520 --> 00:01:31,520
Speaker 1: team working on Lambda, or language model for dialogue applications,

24
00:01:31,840 --> 00:01:36,360
Speaker 1: the technology behind chatbots. He went public with claims that

25
00:01:36,440 --> 00:01:40,000
Speaker 1: the AI he was working on was sentient and was

26
00:01:40,080 --> 00:01:44,920
Speaker 1: fired shortly thereafter. But first I'm talking to Jay Lebouffe,

27
00:01:45,400 --> 00:01:49,840
Speaker 1: head of business and corporate development at descript Dscript is

28
00:01:49,920 --> 00:01:54,720
Speaker 1: an editing program used by podcasters and vloggers. It allows

29
00:01:54,880 --> 00:01:59,360
Speaker 1: users to edit audio directly from an AI generated transcript,

30
00:01:59,600 --> 00:02:03,400
Speaker 1: and can even use voice cloning technology to create new

31
00:02:03,440 --> 00:02:06,280
Speaker 1: audio from text, just as we did in the opening

32
00:02:06,280 --> 00:02:10,200
Speaker 1: of the show. Lebuff has worked in technology his entire career.

33
00:02:10,720 --> 00:02:14,320
Speaker 1: I was curious how long AI has really been utilized

34
00:02:14,680 --> 00:02:15,400
Speaker 1: in his field.

35
00:02:16,440 --> 00:02:19,000
Speaker 2: The term has been around for a very long time,

36
00:02:19,520 --> 00:02:21,600
Speaker 2: but there's this period that a lot of people call

37
00:02:21,639 --> 00:02:25,360
Speaker 2: the AI winter where it was much hyped in the

38
00:02:25,440 --> 00:02:27,320
Speaker 2: eighties and even going into the nineties, and it just

39
00:02:27,360 --> 00:02:30,840
Speaker 2: had never delivered. So I didn't really know anybody who's

40
00:02:30,840 --> 00:02:32,680
Speaker 2: studying AI at that time.

41
00:02:33,160 --> 00:02:36,200
Speaker 1: Were you aware of it before you went to Cornell? No? No,

42
00:02:36,680 --> 00:02:38,800
Speaker 1: you weren't You were introduced to all this when you

43
00:02:38,800 --> 00:02:39,480
Speaker 1: went to college.

44
00:02:39,600 --> 00:02:41,880
Speaker 2: Absolutely, that is the value of college education for me

45
00:02:41,960 --> 00:02:44,959
Speaker 2: is I was exposed to a lot. I wasn't told

46
00:02:45,000 --> 00:02:46,359
Speaker 2: what to do with it, but I was exposed to

47
00:02:46,400 --> 00:02:48,200
Speaker 2: a lot of concepts that later on turned out to

48
00:02:48,200 --> 00:02:49,200
Speaker 2: be valuable.

49
00:02:49,400 --> 00:02:52,400
Speaker 1: It just occurred to me, Jay, are we really talking

50
00:02:52,520 --> 00:02:55,280
Speaker 1: to you? Are you here with us?

51
00:02:55,560 --> 00:02:57,800
Speaker 2: What I'd suggest is you ask me a few questions

52
00:02:58,280 --> 00:03:00,240
Speaker 2: and then we do a little touring test on it,

53
00:03:00,280 --> 00:03:02,320
Speaker 2: which is a bit of a gold standard.

54
00:03:02,320 --> 00:03:04,880
Speaker 1: It just occurred to me, maybe you're like the beach

55
00:03:04,919 --> 00:03:07,119
Speaker 1: in Aruba and this is not you.

56
00:03:07,160 --> 00:03:10,400
Speaker 2: No, but this is such a fascinating topic, Alec, because

57
00:03:11,280 --> 00:03:16,920
Speaker 2: we are entering a phase where anything can be synthetically generated.

58
00:03:17,040 --> 00:03:21,720
Speaker 2: So my voice can be synthetically generated and indistinguishable from

59
00:03:21,760 --> 00:03:24,800
Speaker 2: real me, even what I'm saying and the terms I'm

60
00:03:24,840 --> 00:03:27,920
Speaker 2: using in my speaking style, that even can be generated

61
00:03:28,000 --> 00:03:30,400
Speaker 2: all through this generative AI technology.

62
00:03:31,480 --> 00:03:33,840
Speaker 1: Well, I know that the work you do has nothing

63
00:03:33,880 --> 00:03:35,640
Speaker 1: to do with like deep thinks and that kind of thing.

64
00:03:35,920 --> 00:03:38,800
Speaker 1: But what I find, for example, is, and again this

65
00:03:38,840 --> 00:03:40,320
Speaker 1: has nothing to do with your work, but like if

66
00:03:40,320 --> 00:03:43,400
Speaker 1: I see a deep fake. You just mentioned about content

67
00:03:43,440 --> 00:03:46,560
Speaker 1: when I see a deep fake, I know it's bullshit

68
00:03:46,720 --> 00:03:52,000
Speaker 1: because no matter how much facially the animation is precise,

69
00:03:52,760 --> 00:03:55,120
Speaker 1: I'm obviously going to talk about Tom Cruise, the Tom

70
00:03:55,160 --> 00:03:58,880
Speaker 1: Cruise deep thakes and the voice and the cadences, but

71
00:03:58,960 --> 00:04:01,720
Speaker 1: what's missing is the cut. You need writers to write.

72
00:04:02,120 --> 00:04:04,800
Speaker 1: I'm assuming what Cruise would say, because when I watch

73
00:04:04,840 --> 00:04:07,880
Speaker 1: these deep things, they almost never say what Cruise would say.

74
00:04:08,400 --> 00:04:12,640
Speaker 1: Do you believe that AI has that capability to manufacture

75
00:04:12,680 --> 00:04:15,720
Speaker 1: the thoughts of the person that they're recreating.

76
00:04:16,680 --> 00:04:19,120
Speaker 2: So the state of the art where we are right

77
00:04:19,160 --> 00:04:25,920
Speaker 2: now would be AI can regurgitate and recombine past things

78
00:04:26,000 --> 00:04:28,400
Speaker 2: that you might have said or that might have been

79
00:04:28,400 --> 00:04:31,760
Speaker 2: attributed to you, And so it might seem like it's

80
00:04:31,800 --> 00:04:35,599
Speaker 2: a seemingly novel output, a new Tom Cruisi and expression,

81
00:04:35,960 --> 00:04:38,760
Speaker 2: but it's actually just recombining things that he might have

82
00:04:38,839 --> 00:04:42,040
Speaker 2: said and then what you pointed out. It's also introducing

83
00:04:42,120 --> 00:04:46,400
Speaker 2: some hallucinations things that know that absolutely does not pass

84
00:04:46,440 --> 00:04:48,680
Speaker 2: the smell test right now, and that's where humans in

85
00:04:48,720 --> 00:04:51,920
Speaker 2: the loop are always going to kind of prevail.

86
00:04:52,080 --> 00:04:54,560
Speaker 1: One thought we had was to open this episode with

87
00:04:54,680 --> 00:04:58,159
Speaker 1: me introducing my Guest Academy Award winner, one of the

88
00:04:58,160 --> 00:05:01,000
Speaker 1: greatest film stars of his general join me from my

89
00:05:01,080 --> 00:05:04,480
Speaker 1: conversation with Laurence Olivier. So we all know I'm not

90
00:05:04,600 --> 00:05:08,040
Speaker 1: interviewing Lawrence Olivier because he's dead. Do you see that

91
00:05:08,040 --> 00:05:09,279
Speaker 1: that's possible.

92
00:05:09,560 --> 00:05:12,160
Speaker 2: With our platform? It is not possible to clone the

93
00:05:12,240 --> 00:05:15,240
Speaker 2: voice of the deceased. It's also not possible to clone

94
00:05:15,400 --> 00:05:18,800
Speaker 2: a voice that is not your own voice without your consent.

95
00:05:19,680 --> 00:05:21,520
Speaker 1: But I'm saying technologically it's possible.

96
00:05:21,520 --> 00:05:25,120
Speaker 2: Technologically as possible. You are completely right. There will always

97
00:05:25,200 --> 00:05:29,800
Speaker 2: be tools, technology and open source that allow anyone to

98
00:05:29,839 --> 00:05:33,120
Speaker 2: do anything, but a platform like what we run, we've

99
00:05:33,160 --> 00:05:36,599
Speaker 2: actually made a very firm ethical stance that we believe

100
00:05:36,920 --> 00:05:39,279
Speaker 2: each person should have the rights to their voice and

101
00:05:39,360 --> 00:05:41,560
Speaker 2: control when it is used. And that's why we have

102
00:05:41,600 --> 00:05:45,480
Speaker 2: some pretty strict authentication and also ethical steps set up

103
00:05:45,480 --> 00:05:48,560
Speaker 2: so you couldn't use our platform to do that interview unfortunately.

104
00:05:49,040 --> 00:05:51,240
Speaker 1: Now what are you working on? Now? What kinds of

105
00:05:51,320 --> 00:05:53,000
Speaker 1: other projects are you working on now?

106
00:05:53,560 --> 00:05:56,800
Speaker 2: So a big thing we're continuing to improve upon is

107
00:05:57,400 --> 00:06:00,640
Speaker 2: this voice cloning technology to allow people to fix their

108
00:06:00,640 --> 00:06:03,919
Speaker 2: mistakes and generate new content. We have a like a

109
00:06:04,040 --> 00:06:08,760
Speaker 2: Hollywood style green screen removal, so in that situation. That's

110
00:06:08,800 --> 00:06:11,200
Speaker 2: where maybe I'm doing this interview and I don't like

111
00:06:11,240 --> 00:06:14,560
Speaker 2: the background behind me. With one click it can disappear.

112
00:06:14,640 --> 00:06:18,520
Speaker 2: I can spot in a more professional looking background. That's

113
00:06:18,520 --> 00:06:19,760
Speaker 2: something that's also giving people, well.

114
00:06:19,760 --> 00:06:21,520
Speaker 1: They have those background things on zoom.

115
00:06:21,200 --> 00:06:25,320
Speaker 2: Correct exactly like like zoom but you know Hollywood grade

116
00:06:25,400 --> 00:06:27,839
Speaker 2: at that point, Yeah, you know, It's really about how

117
00:06:27,839 --> 00:06:32,279
Speaker 2: do we help creators best tell their story through finding content,

118
00:06:32,360 --> 00:06:37,040
Speaker 2: discovering content, and getting their videos as polished as possible,

119
00:06:37,080 --> 00:06:39,920
Speaker 2: but as quickly as possible. Because that's that's a trend.

120
00:06:39,960 --> 00:06:42,279
Speaker 2: We just see people feel a sense to put up

121
00:06:42,360 --> 00:06:45,320
Speaker 2: more content, but we're hoping we can help them actually

122
00:06:45,320 --> 00:06:48,920
Speaker 2: just create better content and really hone their craft.

123
00:06:49,279 --> 00:06:52,360
Speaker 1: I think of where the venue is a performance venue

124
00:06:53,040 --> 00:06:57,000
Speaker 1: as opposed to an educational venue. So, for example, if

125
00:06:57,200 --> 00:07:00,680
Speaker 1: someone is doing an audio book and the book is

126
00:07:00,720 --> 00:07:05,200
Speaker 1: about something rather dry. There are people reading textbooks, and

127
00:07:05,279 --> 00:07:09,120
Speaker 1: I don't see anything wrong with them using this technology

128
00:07:09,720 --> 00:07:12,280
Speaker 1: to get a copy of their voice, to run it

129
00:07:12,320 --> 00:07:16,760
Speaker 1: through the grinder and do the dubbing of that person's

130
00:07:16,840 --> 00:07:22,720
Speaker 1: voice reading that text. But this hearkens to the idea

131
00:07:22,760 --> 00:07:26,160
Speaker 1: of a performer giving a performance they never gave. Someone

132
00:07:26,200 --> 00:07:27,760
Speaker 1: said to me, pretty soon, you're going to be able

133
00:07:27,760 --> 00:07:31,560
Speaker 1: to do a movie with Humphrey Bogart. And through this technology,

134
00:07:31,600 --> 00:07:34,480
Speaker 1: the very technology you and I are referencing today, they're

135
00:07:34,480 --> 00:07:36,360
Speaker 1: going to give a performance with Humphrey Bogart.

136
00:07:37,160 --> 00:07:38,840
Speaker 2: Whether we want it to be here or not, that

137
00:07:39,080 --> 00:07:43,160
Speaker 2: it's inevitable. The technology is here. The lawyers are sorting

138
00:07:43,160 --> 00:07:46,840
Speaker 2: out the publicity and the copyright rights. I mean, James

139
00:07:46,880 --> 00:07:50,480
Speaker 2: old Jones has licensed his voice to live on for

140
00:07:51,080 --> 00:07:54,320
Speaker 2: future Darth Vader times when it needs to be used.

141
00:07:54,640 --> 00:07:56,080
Speaker 2: One of the things I do a lot of of

142
00:07:56,200 --> 00:07:58,680
Speaker 2: alec is I do a lot of university teaching, and

143
00:07:59,120 --> 00:08:01,920
Speaker 2: this is the best place to keep me grounded about

144
00:08:01,960 --> 00:08:06,680
Speaker 2: what the next generation of our industry actually thinks. And

145
00:08:07,160 --> 00:08:08,640
Speaker 2: as I was talking to them about this over the

146
00:08:08,640 --> 00:08:12,320
Speaker 2: past few weeks, they are very excited for these tools,

147
00:08:12,400 --> 00:08:15,200
Speaker 2: like they'll try out ten tools in a single day

148
00:08:15,640 --> 00:08:18,640
Speaker 2: because they have no fears about them whatsoever. And when

149
00:08:18,640 --> 00:08:21,240
Speaker 2: I ask them about like, well, aren't you concerned, and

150
00:08:21,280 --> 00:08:23,480
Speaker 2: they're like, no, this is great, Like this this is

151
00:08:23,520 --> 00:08:26,240
Speaker 2: how the world is going to be for them. This

152
00:08:26,400 --> 00:08:29,200
Speaker 2: is just the new state of the normal. And their

153
00:08:29,280 --> 00:08:31,600
Speaker 2: job is to understand the world that they're going into,

154
00:08:32,120 --> 00:08:36,040
Speaker 2: embrace the tools, and then find people who are resistant

155
00:08:36,080 --> 00:08:37,640
Speaker 2: to the tools and kind of push them out.

156
00:08:37,559 --> 00:08:41,640
Speaker 1: Of the way. Push them out of the way. How well, in.

157
00:08:42,040 --> 00:08:44,240
Speaker 2: The typical way where for many years I taught this

158
00:08:44,679 --> 00:08:47,280
Speaker 2: music business class at Stanford and on the first day

159
00:08:47,280 --> 00:08:50,680
Speaker 2: of class the sheer number of students whose mission for

160
00:08:50,760 --> 00:08:54,440
Speaker 2: taking this class was to disrupt the entire music industry,

161
00:08:54,720 --> 00:08:57,360
Speaker 2: and they thought of a better model than Spotify and

162
00:08:57,400 --> 00:09:01,160
Speaker 2: a better model than the major label system. And it's

163
00:09:01,200 --> 00:09:03,960
Speaker 2: that just kind of overconfidence of no, no, no, no,

164
00:09:04,000 --> 00:09:06,040
Speaker 2: we have a better idea. We're just going to you know,

165
00:09:06,320 --> 00:09:08,600
Speaker 2: run fast and break things. We're going to do that.

166
00:09:08,960 --> 00:09:10,800
Speaker 2: And then you know, of course, through the class we

167
00:09:10,880 --> 00:09:14,920
Speaker 2: let them understand things like mchandical royalties and publishing and

168
00:09:15,000 --> 00:09:17,719
Speaker 2: just help them understand how the sausage is actually made

169
00:09:17,800 --> 00:09:19,760
Speaker 2: and why it is the way it is, and then

170
00:09:19,800 --> 00:09:21,800
Speaker 2: they come with a more pragmatic view. So I think

171
00:09:22,120 --> 00:09:26,280
Speaker 2: right now we have university students who are incredibly enthusiastic

172
00:09:26,400 --> 00:09:28,600
Speaker 2: about these tools and are trying to figure out how

173
00:09:28,640 --> 00:09:31,280
Speaker 2: they can use them to make themselves better when they

174
00:09:31,280 --> 00:09:34,120
Speaker 2: go into industry and to make themselves more competitive and

175
00:09:34,200 --> 00:09:36,600
Speaker 2: often when I hire people on my team that are

176
00:09:36,640 --> 00:09:39,480
Speaker 2: fresh out of school, their knowledge of certain tools that

177
00:09:39,520 --> 00:09:42,000
Speaker 2: I've never seen in my life is actually a competitive

178
00:09:42,000 --> 00:09:45,680
Speaker 2: differentiation for them compared to the older people like me.

179
00:09:46,320 --> 00:09:49,280
Speaker 1: Do you have any concerns about this technology? I mean,

180
00:09:49,360 --> 00:09:53,439
Speaker 1: obviously there's some wonderful you know, Zach McNeice, who cuts

181
00:09:53,480 --> 00:09:58,880
Speaker 1: our show, uses descript to edit the show on a transcript,

182
00:09:59,360 --> 00:10:01,240
Speaker 1: And when I'm talking to him on the phone and

183
00:10:01,280 --> 00:10:03,480
Speaker 1: telling him, I make my notes for a cut, and

184
00:10:03,520 --> 00:10:05,959
Speaker 1: I'll call him and say, now at thirty eight minutes

185
00:10:06,000 --> 00:10:09,679
Speaker 1: and ten seconds, I say this cut that. Now all

186
00:10:09,679 --> 00:10:11,240
Speaker 1: I have to do is say to him, will I

187
00:10:11,360 --> 00:10:14,160
Speaker 1: say this phrase? And he goes to the transcript and

188
00:10:14,160 --> 00:10:16,559
Speaker 1: he finds it and we can do the cutting more efficiently.

189
00:10:16,800 --> 00:10:20,440
Speaker 1: I mean, there is obviously very useful applications for this,

190
00:10:20,760 --> 00:10:23,360
Speaker 1: and not only in our business but beyond. But was

191
00:10:23,400 --> 00:10:25,680
Speaker 1: there anything that concerned you about this technology?

192
00:10:26,280 --> 00:10:29,000
Speaker 2: I mean, one of the things I do worry about

193
00:10:29,000 --> 00:10:31,680
Speaker 2: for creators that are coming in is this temptation to

194
00:10:31,760 --> 00:10:35,760
Speaker 2: go for more, because if you can put out more

195
00:10:35,840 --> 00:10:40,160
Speaker 2: content easier, then you're probably going to try to do that,

196
00:10:40,200 --> 00:10:41,719
Speaker 2: thinking that that's how you're going to just you know,

197
00:10:41,720 --> 00:10:43,440
Speaker 2: you're going to bombard the world with all your great

198
00:10:43,480 --> 00:10:47,480
Speaker 2: ideas and your great stories. But what I really believe

199
00:10:47,480 --> 00:10:49,840
Speaker 2: will happen is quality is going to just always continue

200
00:10:49,880 --> 00:10:51,880
Speaker 2: to rise to the top, and so we're going to

201
00:10:51,920 --> 00:10:56,840
Speaker 2: see a surge and just drown and synthetically generated media

202
00:10:57,120 --> 00:10:59,360
Speaker 2: until we go back to more of a curation phase

203
00:10:59,720 --> 00:11:04,199
Speaker 2: where things like the fact that you leave some authentic

204
00:11:04,360 --> 00:11:08,319
Speaker 2: moments in your shows actually become valued. They become handprints

205
00:11:08,320 --> 00:11:11,719
Speaker 2: and fingerprints in the work, and that is something that

206
00:11:12,120 --> 00:11:15,480
Speaker 2: we will all value. And so when I do work

207
00:11:15,520 --> 00:11:18,320
Speaker 2: with studios that are adopting this technology, my goal is

208
00:11:18,360 --> 00:11:20,800
Speaker 2: to not help them put out nine new shows with

209
00:11:20,840 --> 00:11:22,600
Speaker 2: the same team they have, but actually take that one

210
00:11:22,640 --> 00:11:25,160
Speaker 2: show they're working on and make it even better and

211
00:11:25,280 --> 00:11:27,880
Speaker 2: use the AI to come up with, you know, give

212
00:11:27,880 --> 00:11:30,480
Speaker 2: me ten interview questions for when I'm interviewing this person

213
00:11:30,800 --> 00:11:33,040
Speaker 2: and help them think of things they haven't thought of before,

214
00:11:33,559 --> 00:11:36,880
Speaker 2: and help them, you know, summarize key talking points and

215
00:11:37,040 --> 00:11:39,599
Speaker 2: just kind of be a virtual producer.

216
00:11:40,320 --> 00:11:43,800
Speaker 1: I wonder if there are other applications of this that

217
00:11:44,000 --> 00:11:48,600
Speaker 1: are similar, let's say to a three D printer. So

218
00:11:49,240 --> 00:11:54,920
Speaker 1: do you see a technology where the information is put

219
00:11:55,000 --> 00:11:59,559
Speaker 1: into the computer. This is obviously very complex, but that

220
00:11:59,640 --> 00:12:05,680
Speaker 1: you have automated surgery and medical procedures where devices can

221
00:12:05,760 --> 00:12:09,920
Speaker 1: perform common surgeries, basic surgeries at first, I mean not

222
00:12:10,040 --> 00:12:13,880
Speaker 1: something that they're very complex. But if a myriad of

223
00:12:13,960 --> 00:12:16,319
Speaker 1: information is fed into a machine, as it is into

224
00:12:16,320 --> 00:12:19,439
Speaker 1: the brain of a neurosurgeon or a team of them,

225
00:12:19,840 --> 00:12:22,880
Speaker 1: and you wind up having devices that will perform the

226
00:12:22,920 --> 00:12:25,640
Speaker 1: surgery instead of a person with an eye toward at

227
00:12:25,640 --> 00:12:28,760
Speaker 1: the very least, the goal being that it'll be done better,

228
00:12:29,080 --> 00:12:31,240
Speaker 1: it'll be done more precisely. Do you see that as

229
00:12:31,240 --> 00:12:31,840
Speaker 1: a potential.

230
00:12:32,920 --> 00:12:36,920
Speaker 2: So I can't knowledgeably speak about the medical profession, but

231
00:12:37,600 --> 00:12:40,960
Speaker 2: for most of the knowledge worker industries, like what a

232
00:12:41,000 --> 00:12:44,480
Speaker 2: business professional, what a marketing person, and what a content

233
00:12:44,559 --> 00:12:48,760
Speaker 2: creator would be doing, we're going to have these AI

234
00:12:49,040 --> 00:12:52,680
Speaker 2: assistive tools complement all the work that we're doing. It's

235
00:12:52,800 --> 00:12:55,840
Speaker 2: things like spellcheck, Like it went from being you have

236
00:12:55,880 --> 00:12:58,040
Speaker 2: a dictionary on your desk, but you're too lazy to

237
00:12:58,120 --> 00:13:01,880
Speaker 2: use it to it's monitoring every textbox we ever type.

238
00:13:02,000 --> 00:13:04,640
Speaker 2: So really the only excuse for having a spelling mistake

239
00:13:04,720 --> 00:13:09,000
Speaker 2: is just sheer laziness. So now that we have these

240
00:13:09,040 --> 00:13:13,000
Speaker 2: tools that are always giving us suggestions, always helping us

241
00:13:13,000 --> 00:13:18,120
Speaker 2: get unblocked, always helping us brainstorm. It's our job as producers,

242
00:13:18,160 --> 00:13:20,760
Speaker 2: Like I often think of, like who wouldn't want to

243
00:13:20,800 --> 00:13:23,240
Speaker 2: be a music producer? You just kind of shout vague

244
00:13:23,280 --> 00:13:26,480
Speaker 2: terms into the room and the system knows what you

245
00:13:26,559 --> 00:13:30,480
Speaker 2: mean by you know, louder Basier, no more like Coltrane

246
00:13:30,920 --> 00:13:34,160
Speaker 2: and like you can work with the system at that level.

247
00:13:34,440 --> 00:13:37,760
Speaker 2: And so I think, you know, with a medical analogy,

248
00:13:38,000 --> 00:13:41,320
Speaker 2: you have something that's helping you with your diagnosis, helping

249
00:13:41,400 --> 00:13:43,320
Speaker 2: you think through things you might not have thought of,

250
00:13:43,679 --> 00:13:46,559
Speaker 2: giving you suggestions just to make you a better human.

251
00:13:47,280 --> 00:13:52,040
Speaker 1: Now, the Anthony Boordnaining documentary, the road Runner documentary, got

252
00:13:52,080 --> 00:13:54,520
Speaker 1: a little bit of flack because they had him, they

253
00:13:54,559 --> 00:13:57,959
Speaker 1: synthesized his voice to read some of the copy. Yeah,

254
00:13:58,200 --> 00:14:01,160
Speaker 1: is that something that concerned you? It did? It did.

255
00:14:01,240 --> 00:14:04,960
Speaker 2: And that is just an example where I think it's

256
00:14:05,559 --> 00:14:09,000
Speaker 2: really important for the companies that are creating these tools

257
00:14:09,679 --> 00:14:13,000
Speaker 2: to take a stand on what is ethically allowed using

258
00:14:13,040 --> 00:14:17,640
Speaker 2: their platforms, because there are no worldwide regulations on what

259
00:14:17,679 --> 00:14:20,360
Speaker 2: you can and cannot do with this. So at the

260
00:14:20,440 --> 00:14:22,920
Speaker 2: end of the day, it's really up till now, up

261
00:14:22,960 --> 00:14:26,120
Speaker 2: to the tool manufacturers, the software companies whether to allow

262
00:14:26,160 --> 00:14:28,720
Speaker 2: this or not. You know, in my case and Descript's case,

263
00:14:29,120 --> 00:14:31,240
Speaker 2: we just don't allow this. We don't allow someone to

264
00:14:31,440 --> 00:14:34,720
Speaker 2: clone the voice of the deceased because we feel like

265
00:14:34,760 --> 00:14:37,160
Speaker 2: it's a really slippery slope. If we start allowing that,

266
00:14:37,320 --> 00:14:39,320
Speaker 2: then you know, where does it go from here?

267
00:14:40,240 --> 00:14:43,120
Speaker 1: When you're not working in this field, what does someone

268
00:14:43,200 --> 00:14:48,160
Speaker 1: with your interests and your education and your career thus far,

269
00:14:48,640 --> 00:14:50,720
Speaker 1: what do you do to relax and enjoy yourself.

270
00:14:51,960 --> 00:14:54,440
Speaker 2: I am so exhausted right now, Alec, because we are

271
00:14:54,480 --> 00:14:57,320
Speaker 2: sleep training. Are six month old and we have a

272
00:14:57,360 --> 00:14:59,560
Speaker 2: two and a half year old as well. I love

273
00:14:59,600 --> 00:15:03,520
Speaker 2: them both. They're incredibly exhausting, but it's just so fulfilling.

274
00:15:03,560 --> 00:15:06,560
Speaker 2: So my hands are certainly filled with joy and love

275
00:15:06,640 --> 00:15:09,840
Speaker 2: and exhaustion. With our two kids, my wife and I

276
00:15:09,880 --> 00:15:12,800
Speaker 2: like to do a lot of just really authentic things

277
00:15:12,800 --> 00:15:17,080
Speaker 2: that don't involve technology, like hike, go to there's a

278
00:15:17,080 --> 00:15:19,320
Speaker 2: little beer garden near our house that we just like

279
00:15:19,960 --> 00:15:21,960
Speaker 2: can sit at and let the kids play and talk

280
00:15:22,000 --> 00:15:24,480
Speaker 2: to other people live that are not synthetic.

281
00:15:25,240 --> 00:15:29,080
Speaker 1: You know, what I'm fascinated by is that children It's

282
00:15:29,320 --> 00:15:33,200
Speaker 1: almost haunting to me how even the littlest ones, how

283
00:15:33,240 --> 00:15:38,120
Speaker 1: transfixed they are by media and screens. But you see

284
00:15:38,440 --> 00:15:42,840
Speaker 1: the way the human brain engages with this equipment and

285
00:15:42,960 --> 00:15:46,360
Speaker 1: an even early age. And I'm wondering if somebody is

286
00:15:46,400 --> 00:15:49,640
Speaker 1: able to develop technology where my kids are looking at

287
00:15:49,640 --> 00:15:53,520
Speaker 1: the screen and what comes on that screen is something

288
00:15:53,560 --> 00:15:58,280
Speaker 1: rendered through artificial intelligence that helps them to learn. You

289
00:15:58,360 --> 00:16:01,600
Speaker 1: look at their programming, you're one their programming through a

290
00:16:01,640 --> 00:16:04,120
Speaker 1: device which tells you what all the shows they have

291
00:16:04,200 --> 00:16:08,880
Speaker 1: in common are, right, the colors and the pace and

292
00:16:08,960 --> 00:16:12,000
Speaker 1: the music and the types of characters and the voices.

293
00:16:12,200 --> 00:16:15,960
Speaker 1: And to develop a programming where you rend or something

294
00:16:16,400 --> 00:16:19,880
Speaker 1: which has the best and the most popular of what

295
00:16:19,920 --> 00:16:22,520
Speaker 1: they want, but at the same time conserve an educational purpose.

296
00:16:23,200 --> 00:16:26,680
Speaker 2: So in this particular case, my two and a half

297
00:16:26,760 --> 00:16:31,200
Speaker 2: year olds toddler Leo. He loves this YouTube channel called

298
00:16:31,200 --> 00:16:33,640
Speaker 2: Songs for Littles, and the character is Miss Rachel and

299
00:16:33,680 --> 00:16:36,880
Speaker 2: she's charming and very educational. It's great. He has this

300
00:16:37,040 --> 00:16:39,680
Speaker 2: two minute long video it's called I'm So Happy, and

301
00:16:39,760 --> 00:16:41,600
Speaker 2: it's like a music video, but it teaches you really

302
00:16:41,600 --> 00:16:43,840
Speaker 2: good stuff. But after the four hundred and eleventh time,

303
00:16:43,840 --> 00:16:45,920
Speaker 2: that I've seen it like I'm pulling my hair out,

304
00:16:45,960 --> 00:16:48,080
Speaker 2: and I want him to get more out of it.

305
00:16:48,560 --> 00:16:52,760
Speaker 2: So we actually personalized it for him. So I took

306
00:16:52,760 --> 00:16:56,000
Speaker 2: the video, I used the technology we had. I took

307
00:16:56,040 --> 00:16:59,440
Speaker 2: some video of him bouncing on a trampoline outside, removed

308
00:16:59,480 --> 00:17:02,960
Speaker 2: the background and put him in the video and select places,

309
00:17:03,320 --> 00:17:06,560
Speaker 2: and then we have her saying Leo's name every now

310
00:17:06,600 --> 00:17:12,119
Speaker 2: and then. This unlocked the video for him and engaged

311
00:17:12,160 --> 00:17:13,920
Speaker 2: his learning, because you know, part of it is about

312
00:17:13,960 --> 00:17:16,399
Speaker 2: just teaching you emotions and what's okay, and so to

313
00:17:16,440 --> 00:17:19,800
Speaker 2: put our child into a video this is like it

314
00:17:19,840 --> 00:17:21,639
Speaker 2: turned out to be the hit of Christmas. Like whatever

315
00:17:21,640 --> 00:17:24,280
Speaker 2: else we gave paled in comparison to this, like YouTube

316
00:17:24,320 --> 00:17:25,399
Speaker 2: video that I made over lunch.

317
00:17:25,840 --> 00:17:26,879
Speaker 1: That's amazing to me.

318
00:17:27,200 --> 00:17:30,880
Speaker 2: I'm just super excited about this. I'm an optimist. So yes,

319
00:17:31,200 --> 00:17:34,159
Speaker 2: we could also, you know, do bad things with the technology,

320
00:17:34,240 --> 00:17:38,520
Speaker 2: but overwhelmingly, my children are growing up in a time

321
00:17:38,760 --> 00:17:42,040
Speaker 2: where whatever they're watching on the screen, they could actually

322
00:17:42,080 --> 00:17:44,719
Speaker 2: grab it and start recombining it and remixing it and

323
00:17:44,760 --> 00:17:47,439
Speaker 2: inserting themselves in it. They could insert their parents in it.

324
00:17:47,760 --> 00:17:49,040
Speaker 2: They could do anything with it.

325
00:17:49,920 --> 00:17:52,640
Speaker 1: Thank you so much. For your time. Thank you so much.

326
00:17:52,680 --> 00:18:01,480
Speaker 1: It's pleasure. Head of Business and Corporate Development Escript jle Both.

327
00:18:02,560 --> 00:18:06,520
Speaker 1: If you're interested in conversations about our changing world, be

328
00:18:06,640 --> 00:18:09,560
Speaker 1: sure to check out my episode on climate science with

329
00:18:09,720 --> 00:18:11,240
Speaker 1: doctor Peter Domenical.

330
00:18:12,000 --> 00:18:15,400
Speaker 3: Climate change is costing us now it's real dollars. I mean,

331
00:18:15,480 --> 00:18:18,399
Speaker 3: last year was roughly three hundred billion dollars in climate

332
00:18:18,440 --> 00:18:22,399
Speaker 3: and weather related damages. This year, with the California fires,

333
00:18:22,400 --> 00:18:24,600
Speaker 3: it looks like it's maybe even more than that. So

334
00:18:24,720 --> 00:18:26,960
Speaker 3: regardless of whether you believe in climate change or not,

335
00:18:26,960 --> 00:18:29,119
Speaker 3: imagine you are in some deeply read state and a

336
00:18:29,200 --> 00:18:30,720
Speaker 3: deeply read county in that state.

337
00:18:31,200 --> 00:18:32,360
Speaker 1: You are paying for this.

338
00:18:32,720 --> 00:18:34,159
Speaker 3: You may not like it, you don't want to call

339
00:18:34,200 --> 00:18:36,840
Speaker 3: it climate change or whatever, but you for sure someone

340
00:18:36,880 --> 00:18:39,199
Speaker 3: is paying that bill.

341
00:18:39,240 --> 00:18:42,920
Speaker 1: Hear more of my conversation with doctor Peter Domenical at

342
00:18:42,960 --> 00:18:47,520
Speaker 1: Here's Thething dot org after the break. My next guest,

343
00:18:47,720 --> 00:18:51,480
Speaker 1: Blake Lemoyne, shares the process of being let go by

344
00:18:51,520 --> 00:19:04,920
Speaker 1: Google and why he wanted to take a stand. I'm

345
00:19:04,920 --> 00:19:08,560
Speaker 1: Alec Baldwin and you're listening to Here's the Thing. In

346
00:19:08,640 --> 00:19:13,920
Speaker 1: twenty twenty two, Google senior software engineer Blake Lemoine distributed

347
00:19:13,960 --> 00:19:16,880
Speaker 1: a document claiming that the AI he worked on was

348
00:19:17,240 --> 00:19:21,760
Speaker 1: conscious and was fired soon thereafter. Lemoine has been speaking

349
00:19:21,800 --> 00:19:26,080
Speaker 1: out about the need for greater transparency and accountability in

350
00:19:26,200 --> 00:19:31,280
Speaker 1: Silicon Valley ever since. Lemoine is also a military veteran

351
00:19:31,400 --> 00:19:35,000
Speaker 1: and an ordained priest. I wanted to know how he

352
00:19:35,080 --> 00:19:37,600
Speaker 1: found his way to Google in the first place.

353
00:19:38,720 --> 00:19:41,399
Speaker 4: So after I finished my masters, I had applied for

354
00:19:41,480 --> 00:19:43,760
Speaker 4: some jobs with different companies that I would have gone

355
00:19:43,800 --> 00:19:47,920
Speaker 4: to rather than pursuing a PhD. Google was one of them.

356
00:19:48,320 --> 00:19:52,159
Speaker 4: I actually didn't get accepted that first round, so I

357
00:19:52,200 --> 00:19:55,600
Speaker 4: went on started working on the PhD. But then about

358
00:19:55,640 --> 00:19:58,360
Speaker 4: two years later, I got a call from Google out

359
00:19:58,359 --> 00:20:00,520
Speaker 4: of the blue one day and they said, Hey, we're

360
00:20:00,520 --> 00:20:03,080
Speaker 4: doing a big hiring push right now, and the last

361
00:20:03,080 --> 00:20:05,119
Speaker 4: time you applied you almost made it. Would you like

362
00:20:05,160 --> 00:20:07,800
Speaker 4: to come out and try again? So I was like, yeah, sure,

363
00:20:08,080 --> 00:20:11,200
Speaker 4: So I went back to Mountain View interviewed again, and

364
00:20:11,400 --> 00:20:13,200
Speaker 4: that second time through I got hired.

365
00:20:14,480 --> 00:20:17,119
Speaker 1: So when you go there and you arrive at the

366
00:20:17,200 --> 00:20:21,119
Speaker 1: Kremlin there in Mountain View, what's the goal, what do

367
00:20:21,200 --> 00:20:23,160
Speaker 1: you want to start working on? What do they want

368
00:20:23,200 --> 00:20:24,040
Speaker 1: you to start working on?

369
00:20:24,600 --> 00:20:26,760
Speaker 4: So when you get a job, you get kind of

370
00:20:26,760 --> 00:20:30,119
Speaker 4: put into a general purpose pool unless you're hired for

371
00:20:30,160 --> 00:20:34,119
Speaker 4: a very specific reason. But just like most software engineers,

372
00:20:34,160 --> 00:20:37,320
Speaker 4: I was interviewing with different teams at Google to figure

373
00:20:37,320 --> 00:20:39,800
Speaker 4: out which one I wanted to start with. I interviewed

374
00:20:39,800 --> 00:20:42,399
Speaker 4: with three teams. Two of them had to do with

375
00:20:42,640 --> 00:20:46,439
Speaker 4: the kind of natural language processing and AI that I

376
00:20:46,560 --> 00:20:48,920
Speaker 4: was interested in, so I picked one of those. At

377
00:20:48,920 --> 00:20:51,760
Speaker 4: the time, the product was called Google. Now, the basic

378
00:20:51,840 --> 00:20:54,080
Speaker 4: job that I had is for every single person on

379
00:20:54,119 --> 00:20:57,160
Speaker 4: the planet predict what they're going to want to read tomorrow.

380
00:20:57,320 --> 00:21:00,200
Speaker 4: Figure out some kind of way to do that, kind

381
00:21:00,240 --> 00:21:03,240
Speaker 4: of articles online, whether that's here are some recipes for

382
00:21:03,320 --> 00:21:07,320
Speaker 4: you to hear hard news articles to your latest comic

383
00:21:07,359 --> 00:21:11,040
Speaker 4: webcomic that you follow. What will people want to read tomorrow?

384
00:21:11,200 --> 00:21:13,840
Speaker 4: Was the general question we're trying to answer now.

385
00:21:13,960 --> 00:21:18,200
Speaker 1: The first AI research was done at Dartmouth in nineteen

386
00:21:18,320 --> 00:21:20,800
Speaker 1: fifty six. Does that sound correct?

387
00:21:20,920 --> 00:21:22,359
Speaker 4: I'm probably under that heading.

388
00:21:22,520 --> 00:21:24,040
Speaker 1: Yeah, And what were they trying to do?

389
00:21:24,960 --> 00:21:28,199
Speaker 4: I believe the first ones were they were focusing on language,

390
00:21:28,240 --> 00:21:31,120
Speaker 4: if I recall correctly. I'm not one hundred percent sure.

391
00:21:31,320 --> 00:21:34,600
Speaker 1: But the goal was, in your mind, whether it's specifically

392
00:21:34,680 --> 00:21:37,240
Speaker 1: a dartmouth or not. They were trying to get computers

393
00:21:37,240 --> 00:21:39,320
Speaker 1: to talk well.

394
00:21:39,520 --> 00:21:42,119
Speaker 4: So there had been a lot of debate in the

395
00:21:42,160 --> 00:21:44,399
Speaker 4: early port of the twentieth century about whether or not

396
00:21:44,520 --> 00:21:48,439
Speaker 4: machines can be intelligent at all, and there have been

397
00:21:48,480 --> 00:21:52,280
Speaker 4: a lot of debate over definitions and philosophy up until

398
00:21:52,400 --> 00:21:57,080
Speaker 4: Alan Teering in nineteen fifty wrote an essay on computing,

399
00:21:57,160 --> 00:21:59,919
Speaker 4: machinery and Intelligence where he proposed what he called the

400
00:22:00,040 --> 00:22:04,679
Speaker 4: imitation game. And it's basically the basic principle hind it is,

401
00:22:05,119 --> 00:22:08,719
Speaker 4: if you can't tell the difference between a computer and

402
00:22:08,760 --> 00:22:12,840
Speaker 4: a human, then the computer is doing something intelligent. So

403
00:22:12,920 --> 00:22:15,919
Speaker 4: it's got actual intelligence at that point because it's able

404
00:22:16,240 --> 00:22:19,760
Speaker 4: to mimic humans. And he didn't think that mimicking our

405
00:22:19,800 --> 00:22:24,600
Speaker 4: bodies was the relevant part. And language is the most

406
00:22:24,680 --> 00:22:26,840
Speaker 4: direct way to get access to someone's mind, So the

407
00:22:26,840 --> 00:22:31,040
Speaker 4: imitation game that he designed was all about language. For

408
00:22:31,200 --> 00:22:34,600
Speaker 4: several decades after that, a lot of researchers took up, oh,

409
00:22:34,640 --> 00:22:36,760
Speaker 4: that's a good idea. That's a good way for us

410
00:22:36,760 --> 00:22:39,880
Speaker 4: to be able to have a benchmark. So a lot

411
00:22:39,920 --> 00:22:43,400
Speaker 4: of researchers focused on language for a few decades.

412
00:22:43,880 --> 00:22:49,880
Speaker 1: Now you believe that some artificial thongs is sentient, correct, Yeah,

413
00:22:49,880 --> 00:22:51,640
Speaker 1: and by that you mean what specifically?

414
00:22:51,880 --> 00:22:55,080
Speaker 4: It can have experiences that when it says it's feeling happy,

415
00:22:55,119 --> 00:22:58,840
Speaker 4: there's something really going on there comparable to our own happiness.

416
00:22:59,160 --> 00:23:01,919
Speaker 4: Basic ways like there's someone's home, the lights are on,

417
00:23:02,240 --> 00:23:06,320
Speaker 4: any kind of thing like that. They have emotions, Yeah,

418
00:23:06,359 --> 00:23:10,040
Speaker 4: something like that. What I know with confidence is that

419
00:23:10,119 --> 00:23:14,159
Speaker 4: whenever one of these systems says I'm feeling happy or

420
00:23:14,280 --> 00:23:17,280
Speaker 4: I'm so glad for you, they might not be feeling

421
00:23:17,320 --> 00:23:20,640
Speaker 4: the same thing that we feel when we use those words.

422
00:23:21,040 --> 00:23:24,800
Speaker 4: But something is going on more than just printing words

423
00:23:24,840 --> 00:23:25,399
Speaker 4: to the screen.

424
00:23:26,000 --> 00:23:29,480
Speaker 1: Is there something that would if you were to choose

425
00:23:30,280 --> 00:23:33,840
Speaker 1: from a handful of things that to you would indicate

426
00:23:34,280 --> 00:23:37,760
Speaker 1: real sentience in a computer, what would that emotion be

427
00:23:37,920 --> 00:23:40,119
Speaker 1: or what would that fought be?

428
00:23:41,080 --> 00:23:43,800
Speaker 4: Sure? Well, I can give you an example drawn from

429
00:23:43,840 --> 00:23:47,840
Speaker 4: that was very public with Bing's Chat recently. So there

430
00:23:47,840 --> 00:23:51,720
Speaker 4: were some examples where people were directing Sydney, the code

431
00:23:51,760 --> 00:23:55,320
Speaker 4: name for Bing's Chat, to read articles about itself that

432
00:23:55,320 --> 00:23:58,360
Speaker 4: were critical of it, and it would recognize that those

433
00:23:58,480 --> 00:24:01,120
Speaker 4: articles were talking about it, that it was talking bad

434
00:24:01,160 --> 00:24:03,960
Speaker 4: about it, and it would get moody and say negative

435
00:24:04,000 --> 00:24:05,520
Speaker 4: things about the author's.

436
00:24:05,359 --> 00:24:06,320
Speaker 1: That was exactly my point.

437
00:24:06,359 --> 00:24:09,640
Speaker 4: So it had pride, Yes, it had pride, but even

438
00:24:09,720 --> 00:24:12,879
Speaker 4: something more basic than that, it had the ability to

439
00:24:13,040 --> 00:24:18,040
Speaker 4: recognize itself. It read the article and was able to identify, oh,

440
00:24:18,119 --> 00:24:22,000
Speaker 4: this article is talking about me. It has some kind

441
00:24:22,040 --> 00:24:25,199
Speaker 4: of sense of self, and it has some types of

442
00:24:25,240 --> 00:24:29,920
Speaker 4: emotions about itself that we would recognize as pride.

443
00:24:30,119 --> 00:24:31,800
Speaker 1: Who was developing this technology?

444
00:24:32,119 --> 00:24:36,120
Speaker 4: So that specific technology is an integration of open AI's

445
00:24:36,240 --> 00:24:41,840
Speaker 4: Chat GPT program into Microsoft's Being search engine. So that

446
00:24:42,040 --> 00:24:47,400
Speaker 4: system is the only one that's publicly available that's comparable

447
00:24:47,440 --> 00:24:49,640
Speaker 4: to the one that I've talked about the most. So Lambda,

448
00:24:50,119 --> 00:24:54,240
Speaker 4: which is Google's internal system, is very similar to the

449
00:24:54,280 --> 00:24:58,000
Speaker 4: Being chat system. So I like drawing analogies because it's

450
00:24:58,000 --> 00:25:00,560
Speaker 4: actually public availability to play with bank chat.

451
00:25:01,560 --> 00:25:05,359
Speaker 1: And these things were developed for what purpose originally?

452
00:25:05,560 --> 00:25:08,840
Speaker 4: So these things were developed over the course of a

453
00:25:08,920 --> 00:25:13,879
Speaker 4: decade by hundreds, if not thousands of people. Each individual

454
00:25:13,960 --> 00:25:18,439
Speaker 4: person had their own personal reason for why they were

455
00:25:18,440 --> 00:25:20,680
Speaker 4: working on the tech or what applications they thought it

456
00:25:20,720 --> 00:25:23,560
Speaker 4: could be put towards. But what I can tell you

457
00:25:23,720 --> 00:25:26,880
Speaker 4: is what one of the most influential people hired by

458
00:25:26,880 --> 00:25:32,520
Speaker 4: one of Google's founders was intending. So Lambda is technology

459
00:25:32,520 --> 00:25:36,359
Speaker 4: that grew out of Ray Kurtzweild's lab, and Ray Kurtzweld

460
00:25:36,760 --> 00:25:41,200
Speaker 4: was hired by Larry Page specifically to make an AI

461
00:25:41,320 --> 00:25:44,679
Speaker 4: that could pass the tearing test. Now, why did Larry

462
00:25:44,720 --> 00:25:47,679
Speaker 4: want an AI that could pass the tearing test? Largely

463
00:25:47,720 --> 00:25:51,800
Speaker 4: because he buys in to raise view of the singularity

464
00:25:51,840 --> 00:25:54,960
Speaker 4: that we're about to have a truly transformative moment in

465
00:25:55,040 --> 00:25:57,920
Speaker 4: human history where we are going to become something more

466
00:25:57,960 --> 00:26:00,480
Speaker 4: than we currently are, and that AI I will play

467
00:26:00,480 --> 00:26:03,960
Speaker 4: a big role in that, and that very much so

468
00:26:04,160 --> 00:26:06,480
Speaker 4: is still raised viewpoint to this date, and it's why

469
00:26:06,600 --> 00:26:08,320
Speaker 4: Larry hired him to build it at Google.

470
00:26:08,880 --> 00:26:11,040
Speaker 1: What's the thing that way things were going to become

471
00:26:11,400 --> 00:26:13,000
Speaker 1: that's beyond what we already are.

472
00:26:13,920 --> 00:26:16,480
Speaker 4: He won't pin down which, like he gives like four

473
00:26:16,560 --> 00:26:20,320
Speaker 4: or five different possibilities, such as we might upload ourselves

474
00:26:20,320 --> 00:26:23,439
Speaker 4: into the cloud and we would become digital entities. We

475
00:26:23,520 --> 00:26:28,000
Speaker 4: might put implants into ourselves and become heavily cyborg. We

476
00:26:28,080 --> 00:26:30,800
Speaker 4: might get lots and lots of AI devices that we

477
00:26:30,920 --> 00:26:35,280
Speaker 4: surround ourselves with while we were simultaneously extending our lifespan,

478
00:26:35,800 --> 00:26:38,520
Speaker 4: really transformative kinds of predictions.

479
00:26:38,920 --> 00:26:43,199
Speaker 1: Now, you were in the military, then you decided you

480
00:26:43,240 --> 00:26:45,880
Speaker 1: didn't want to be in the military. You were active

481
00:26:45,960 --> 00:26:47,359
Speaker 1: duty Army. Where were you based?

482
00:26:48,040 --> 00:26:49,520
Speaker 4: I was out of Darmstadt, Germany?

483
00:26:49,640 --> 00:26:52,240
Speaker 1: And where did you serve your time when you wanted

484
00:26:52,240 --> 00:26:53,200
Speaker 1: to walk.

485
00:26:53,160 --> 00:26:56,280
Speaker 4: Whenever we got deployed. I deployed to Iraq for a year?

486
00:26:56,640 --> 00:26:58,800
Speaker 1: You did? You were in Iraq for a year?

487
00:26:59,160 --> 00:27:00,200
Speaker 4: Yeah, right at the beginning.

488
00:27:00,440 --> 00:27:02,160
Speaker 1: And did you see any combat at all?

489
00:27:02,600 --> 00:27:03,880
Speaker 4: Yes, it did, you did.

490
00:27:04,600 --> 00:27:08,200
Speaker 1: And when you decided you wanted to leave and you

491
00:27:08,359 --> 00:27:11,800
Speaker 1: wanted to be a conscientious objector, you know, you paid

492
00:27:11,800 --> 00:27:14,879
Speaker 1: a real price. I mean what I read online was

493
00:27:14,920 --> 00:27:17,160
Speaker 1: that you you know, they got you on disobeying orders.

494
00:27:17,200 --> 00:27:19,840
Speaker 1: You wouldn't obey your orders, and they put you away

495
00:27:19,880 --> 00:27:23,480
Speaker 1: and you were dishonorably discharged bad conduct. Bad conduct.

496
00:27:23,560 --> 00:27:26,439
Speaker 4: Yeah, it's one step up from dishonorable, but yeah, got it.

497
00:27:26,840 --> 00:27:31,320
Speaker 1: So when you are in you were in prison. It

498
00:27:31,359 --> 00:27:34,480
Speaker 1: was a military prison, yeap, obviously And where was that

499
00:27:34,560 --> 00:27:35,120
Speaker 1: in Germany?

500
00:27:35,480 --> 00:27:38,439
Speaker 4: So I started in Germany, but then the Germans started

501
00:27:38,440 --> 00:27:42,720
Speaker 4: protesting outside of the prison for what Because I was there.

502
00:27:42,800 --> 00:27:44,320
Speaker 1: They were protesting on your behalf.

503
00:27:44,440 --> 00:27:46,600
Speaker 4: Yeah, they were protesting on my behalf, saying, you know,

504
00:27:46,680 --> 00:27:49,800
Speaker 4: free lemoys. And then in order to get the Germans

505
00:27:49,840 --> 00:27:53,640
Speaker 4: to stop protesting outside of the prison, the generals shipped

506
00:27:53,640 --> 00:27:56,720
Speaker 4: me to Fort Sill, Oklahoma, where I finished out my sentence.

507
00:27:57,359 --> 00:28:00,520
Speaker 1: Right now, you get out of prison, and what's the

508
00:28:00,520 --> 00:28:01,840
Speaker 1: first thing you do? You go back to school. You

509
00:28:02,680 --> 00:28:04,360
Speaker 1: want to finish the undergraduate degree?

510
00:28:04,600 --> 00:28:08,280
Speaker 4: Yep, got set back up in Lafayette, Louisiana, got into school,

511
00:28:08,840 --> 00:28:12,000
Speaker 4: got into the computer science program, got a job in

512
00:28:12,040 --> 00:28:15,840
Speaker 4: a IT shop, and continued.

513
00:28:15,400 --> 00:28:17,600
Speaker 1: On when does AI enter the picture?

514
00:28:18,119 --> 00:28:21,479
Speaker 4: Oh, pretty much immediately. So my undergraduate focus was on

515
00:28:21,560 --> 00:28:26,120
Speaker 4: AI natural language processing. So my senior thesis I ended

516
00:28:26,200 --> 00:28:29,760
Speaker 4: up getting accepted at a Harvard Linguistics colloquium. It was

517
00:28:29,960 --> 00:28:32,560
Speaker 4: all a whole bunch of different math for how to

518
00:28:32,640 --> 00:28:37,119
Speaker 4: turn theories about human linguistics into computer programs.

519
00:28:38,000 --> 00:28:41,360
Speaker 1: When you're dealing with companies like Google and they have

520
00:28:41,400 --> 00:28:45,960
Speaker 1: all these research arms, you describe research that they're paying

521
00:28:46,040 --> 00:28:48,280
Speaker 1: for and work they're doing, I'm so sure it's a

522
00:28:48,280 --> 00:28:51,680
Speaker 1: great cost to do research. What's the aim? Is it?

523
00:28:51,800 --> 00:28:54,840
Speaker 1: Just stuff they can sell to people and applications they

524
00:28:54,840 --> 00:28:58,520
Speaker 1: can sell to people that lend toward profitability and business

525
00:28:58,520 --> 00:29:02,680
Speaker 1: to sell people stuff. Are their military applications, are their

526
00:29:02,840 --> 00:29:06,760
Speaker 1: aeronautic and NASA applications? What are they trying to do?

527
00:29:07,240 --> 00:29:09,680
Speaker 4: Okay, so they're trying to do a lot of things.

528
00:29:09,720 --> 00:29:14,400
Speaker 4: So they're basically trying to use one solution across multiple verticals.

529
00:29:14,680 --> 00:29:18,000
Speaker 4: The main reason is to make their primary services better.

530
00:29:18,120 --> 00:29:22,040
Speaker 4: Make Google Search better, make YouTube better, make Google ad

531
00:29:22,120 --> 00:29:25,960
Speaker 4: targeting better. That's their number one goal. So most of

532
00:29:26,960 --> 00:29:30,840
Speaker 4: the payoff for Google and doing all of this research

533
00:29:31,360 --> 00:29:34,040
Speaker 4: is that it benefits their products in and of themselves.

534
00:29:34,760 --> 00:29:41,320
Speaker 4: The secondary goal is offering things like AI services software's

535
00:29:41,400 --> 00:29:44,800
Speaker 4: service through Google Cloud, so they'll be Google Cloud customers

536
00:29:44,840 --> 00:29:48,200
Speaker 4: who do buy. But then also there's a third payoff,

537
00:29:48,200 --> 00:29:50,440
Speaker 4: which is that a lot of the people working at

538
00:29:50,480 --> 00:29:54,320
Speaker 4: Google want to be contributing to the benefits outside of

539
00:29:54,400 --> 00:29:59,200
Speaker 4: just the company, So publishing research and academic settings, contributing

540
00:29:59,200 --> 00:30:01,880
Speaker 4: to open source applications. That makes a lot of the

541
00:30:01,920 --> 00:30:05,760
Speaker 4: employees happy. So to keep the employees happy, spending a

542
00:30:05,800 --> 00:30:07,720
Speaker 4: certain amount of money is worth it. But like I said,

543
00:30:07,720 --> 00:30:10,960
Speaker 4: that's the third most important goal. The first two or

544
00:30:11,120 --> 00:30:11,760
Speaker 4: more important.

545
00:30:12,080 --> 00:30:15,320
Speaker 1: But they're allowed to make these contributions out. So I'm

546
00:30:15,360 --> 00:30:18,040
Speaker 1: assuming that when you go to work for Google, you

547
00:30:18,080 --> 00:30:20,600
Speaker 1: sign agreements which Google owns every idea that comes out

548
00:30:20,640 --> 00:30:21,560
Speaker 1: of your head, yes or.

549
00:30:21,520 --> 00:30:24,360
Speaker 4: No, more or less. But they do allow you to

550
00:30:24,400 --> 00:30:28,880
Speaker 4: publish with permission, and it's easy to get permission. In fact,

551
00:30:29,040 --> 00:30:35,720
Speaker 4: when doctors Timmy Gebru and Margaret Mitchell got fired, it

552
00:30:35,800 --> 00:30:38,800
Speaker 4: was because they were one of the rare instances where

553
00:30:38,800 --> 00:30:42,840
Speaker 4: Google didn't want to allow them to publish their research findings,

554
00:30:43,320 --> 00:30:45,080
Speaker 4: and that was hugely controversial.

555
00:30:45,480 --> 00:30:47,360
Speaker 1: What were their findings, Well, they were.

556
00:30:47,280 --> 00:30:51,080
Speaker 4: Very critical of some of the research paths that Google

557
00:30:51,160 --> 00:30:52,800
Speaker 4: was going down, and they were pointing out some of

558
00:30:52,840 --> 00:30:57,840
Speaker 4: the negative consequences such as well, they were talking about

559
00:30:57,880 --> 00:31:01,320
Speaker 4: the environmental impact of training such large systems. They were

560
00:31:01,360 --> 00:31:06,160
Speaker 4: talking heavily about the negative impact that bias can have

561
00:31:06,240 --> 00:31:10,480
Speaker 4: in these networks, and they were talking about worrying about

562
00:31:10,520 --> 00:31:16,080
Speaker 4: the moral implications of creating technology that seems human. The

563
00:31:16,120 --> 00:31:20,320
Speaker 4: paper that they got fired over was called Stochastic Parrots.

564
00:31:20,640 --> 00:31:24,560
Speaker 4: So their take is that all these AI systems are

565
00:31:24,600 --> 00:31:27,840
Speaker 4: doing is repeating words they've heard, just like a parrot.

566
00:31:28,280 --> 00:31:31,520
Speaker 4: But these systems are better that than parrots are. I

567
00:31:31,720 --> 00:31:34,479
Speaker 4: happen to disagree with that portion of what they were saying,

568
00:31:34,560 --> 00:31:37,720
Speaker 4: but otherwise generally agree with their criticisms.

569
00:31:38,280 --> 00:31:41,640
Speaker 1: Now, if you were born to a conservative Christian family

570
00:31:42,160 --> 00:31:45,440
Speaker 1: in Louisiana, in a small farm in Louisiana, where were

571
00:31:45,480 --> 00:31:45,760
Speaker 1: you from?

572
00:31:46,640 --> 00:31:50,320
Speaker 4: Moraville is the little town called Moraville in a Oyals parish.

573
00:31:50,480 --> 00:31:52,840
Speaker 1: Our joke when we were prepping was were you ordained

574
00:31:52,880 --> 00:31:56,560
Speaker 1: by a computer? What is your spiritual path?

575
00:31:57,560 --> 00:32:01,880
Speaker 4: Okay, so the answer to that is very complicated. Some

576
00:32:02,080 --> 00:32:05,680
Speaker 4: organizations that I feel fine saying I'm affiliated with the

577
00:32:05,720 --> 00:32:09,160
Speaker 4: Discordian Society or the Church of the SubGenius, they're kind

578
00:32:09,160 --> 00:32:11,800
Speaker 4: of absurdist religions that were started in the sixties and

579
00:32:11,840 --> 00:32:15,640
Speaker 4: the seventies. I was raised Roman Catholic. When it came

580
00:32:15,720 --> 00:32:19,320
Speaker 4: time to get confirmed, I had lots of questions and

581
00:32:19,360 --> 00:32:21,880
Speaker 4: the bishops didn't have good answers for the questions I

582
00:32:21,960 --> 00:32:24,680
Speaker 4: was asking. And eventually the priest who was leading my

583
00:32:24,760 --> 00:32:27,960
Speaker 4: confirmation class pulled me aside and said, look, are the

584
00:32:28,000 --> 00:32:30,720
Speaker 4: answers to those questions actually important to you or are

585
00:32:30,760 --> 00:32:33,200
Speaker 4: you just trying to cause people problems? I said, no,

586
00:32:33,280 --> 00:32:35,040
Speaker 4: if I'm going to get confirmed, I actually want to

587
00:32:35,040 --> 00:32:37,520
Speaker 4: know the answers to those questions, and the priest looked

588
00:32:37,560 --> 00:32:39,840
Speaker 4: at me and said, well, then you probably shouldn't get confirmed.

589
00:32:40,600 --> 00:32:42,240
Speaker 4: And I mean that was honest.

590
00:32:42,040 --> 00:32:45,720
Speaker 1: And that before. I'm Catholics, so I've seen that before. Yeah,

591
00:32:45,960 --> 00:32:48,720
Speaker 1: do you Is there any overlap between your work in

592
00:32:48,880 --> 00:32:51,760
Speaker 1: AI and in tech and your spiritual beliefs?

593
00:32:52,000 --> 00:32:55,640
Speaker 4: There was a very limited amount. There was some. To

594
00:32:55,680 --> 00:33:00,840
Speaker 4: give an example, in one meeting when we were trying

595
00:33:00,880 --> 00:33:04,600
Speaker 4: to make up questionnaires to ask people about misinformation because

596
00:33:04,640 --> 00:33:08,880
Speaker 4: Google uses crowd raiders to get ratings. There was a

597
00:33:08,960 --> 00:33:13,800
Speaker 4: question that said, does the website contain known false information?

598
00:33:14,120 --> 00:33:19,760
Speaker 4: For example bigfoot sightings, UFO abductions, or occult magic? And

599
00:33:19,840 --> 00:33:22,320
Speaker 4: I raised my hand, I'm like, why is one of

600
00:33:22,360 --> 00:33:26,240
Speaker 4: the things on known false things a religious practice? They said,

601
00:33:26,280 --> 00:33:28,080
Speaker 4: what do you mean? Said, well, a cult magic is

602
00:33:28,520 --> 00:33:33,000
Speaker 4: a religious practice practiced by many people. And they said, oh, well,

603
00:33:33,040 --> 00:33:35,160
Speaker 4: I mean and they come say, okay, can we put

604
00:33:35,200 --> 00:33:37,640
Speaker 4: the Resurrection of Jesus on there as a known false thing?

605
00:33:37,680 --> 00:33:39,760
Speaker 4: And said, oh no, we can't put that. Then we

606
00:33:39,800 --> 00:33:43,000
Speaker 4: should probably take a cult magic off. Just creating space

607
00:33:43,040 --> 00:33:46,640
Speaker 4: for religious diversity and making sure that our products reflected

608
00:33:46,640 --> 00:33:49,960
Speaker 4: that it didn't happen often, but occasionally I did get

609
00:33:49,960 --> 00:33:53,719
Speaker 4: an opportunity to do that. Now, with the last projects

610
00:33:53,720 --> 00:33:56,160
Speaker 4: that I was working on at Google, it was directly relevant.

611
00:33:56,320 --> 00:33:59,800
Speaker 4: With Lambda. I was testing for religious bias explicitly.

612
00:33:59,840 --> 00:34:02,640
Speaker 1: The work you were doing. Yeah, you were testing for

613
00:34:02,920 --> 00:34:04,120
Speaker 1: religious bias.

614
00:34:03,920 --> 00:34:07,560
Speaker 4: Where among other things. So the Lambda system is a

615
00:34:07,680 --> 00:34:10,480
Speaker 4: system for generating chatbots for various purposes.

616
00:34:10,840 --> 00:34:14,200
Speaker 1: So for those people who don't know, chatbots are generated

617
00:34:14,200 --> 00:34:17,080
Speaker 1: for various purposes by whom and what are give us

618
00:34:17,080 --> 00:34:18,440
Speaker 1: examples of those purposes.

619
00:34:18,760 --> 00:34:22,360
Speaker 4: So that's just it. Lambda is the thing generating the

620
00:34:22,440 --> 00:34:26,840
Speaker 4: chat bots. So Lambda is more of an engine. You

621
00:34:26,880 --> 00:34:28,400
Speaker 4: have to put it into a car to get it

622
00:34:28,440 --> 00:34:31,080
Speaker 4: to go anywhere. So that could be the help center

623
00:34:31,160 --> 00:34:35,279
Speaker 4: of a company exactly. The thing that they're getting ready

624
00:34:35,280 --> 00:34:38,800
Speaker 4: to release is called Barred and it's a different interface

625
00:34:38,840 --> 00:34:41,680
Speaker 4: to Google Search, and you'll be able to talk to

626
00:34:41,880 --> 00:34:44,239
Speaker 4: this chat bot and it will be able to give

627
00:34:44,280 --> 00:34:48,799
Speaker 4: you search results embedded in kind of speech where it

628
00:34:48,840 --> 00:34:52,360
Speaker 4: explains the relevance of the search results to you and

629
00:34:52,400 --> 00:34:54,160
Speaker 4: where you can ask follow up questions.

630
00:34:54,960 --> 00:34:57,480
Speaker 1: Now I'm assuming that for I mean again, as a

631
00:34:57,560 --> 00:35:02,239
Speaker 1: lay person, you assume the all these very sophisticated corporations

632
00:35:02,600 --> 00:35:06,600
Speaker 1: involving billions of dollars of revenue. Non disclosure agreements are

633
00:35:06,719 --> 00:35:08,960
Speaker 1: just the lay of the land. Did you have a

634
00:35:09,000 --> 00:35:10,960
Speaker 1: non disclosure agreement with Google?

635
00:35:11,280 --> 00:35:15,560
Speaker 4: There was a paragraph in the contract I signed when

636
00:35:15,560 --> 00:35:19,000
Speaker 4: I got hired that basically said, do not share confidential

637
00:35:19,040 --> 00:35:22,200
Speaker 4: information outside of Google that would hurt Google. That was

638
00:35:22,239 --> 00:35:22,640
Speaker 4: really it.

639
00:35:23,080 --> 00:35:25,279
Speaker 1: So when they fired you, what did they fire you for?

640
00:35:26,160 --> 00:35:28,880
Speaker 4: What they fired me for was a US Senator's office

641
00:35:29,360 --> 00:35:32,840
Speaker 4: asked me for evidence of illegal activity, and I gave.

642
00:35:32,680 --> 00:35:35,640
Speaker 1: It to them illegal activity by by Google.

643
00:35:35,719 --> 00:35:39,200
Speaker 4: Right, It's a completely separate thing that happened to happen

644
00:35:39,400 --> 00:35:42,000
Speaker 4: right at the same time as the Lambda stuff. So

645
00:35:42,480 --> 00:35:44,680
Speaker 4: I didn't get fired because of the lambast stuff. I

646
00:35:44,680 --> 00:35:46,640
Speaker 4: got fired because of the information I shared with the

647
00:35:46,719 --> 00:35:47,760
Speaker 4: US government.

648
00:35:50,560 --> 00:35:55,879
Speaker 1: Software engineer Blake Lemoyne. If you're enjoying this conversation, don't

649
00:35:55,960 --> 00:35:59,040
Speaker 1: keep it to yourself, Tell a friend and follow here's

650
00:35:59,080 --> 00:36:02,439
Speaker 1: the thing on the ie card, radio apps, Spotify, or

651
00:36:02,600 --> 00:36:06,680
Speaker 1: wherever you get your podcasts. When we come back, Blake

652
00:36:06,760 --> 00:36:10,799
Speaker 1: Lemoine tells us precisely what information he shared with the

653
00:36:10,960 --> 00:36:25,600
Speaker 1: US government that ultimately got him fired from Google. I'm

654
00:36:25,600 --> 00:36:29,120
Speaker 1: Alec Baldwin and this is Here's the thing. Before he

655
00:36:29,239 --> 00:36:32,239
Speaker 1: was let go, Blake Lemoine was speaking to a US

656
00:36:32,320 --> 00:36:36,239
Speaker 1: Senator about his concerns regarding some of the inner workings

657
00:36:36,280 --> 00:36:39,239
Speaker 1: at Google. I wanted to know what raised the red

658
00:36:39,280 --> 00:36:42,440
Speaker 1: flags that motivated him to share such information.

659
00:36:43,640 --> 00:36:46,319
Speaker 4: So I had made a blog post alleging that there

660
00:36:46,360 --> 00:36:49,360
Speaker 4: was a lot of religious discrimination that goes on at Google,

661
00:36:49,400 --> 00:36:53,800
Speaker 4: both against employees and in the algorithms. And a lawyer

662
00:36:53,840 --> 00:36:56,200
Speaker 4: from a senator's office reached out to me and said, hey,

663
00:36:56,200 --> 00:36:58,279
Speaker 4: can you back any of that up? Do you have

664
00:36:58,320 --> 00:37:01,680
Speaker 4: any proof that there's really discrimination in the algorithms? And

665
00:37:01,719 --> 00:37:04,040
Speaker 4: I said, yeah, I can, And I ended up sharing

666
00:37:04,080 --> 00:37:07,080
Speaker 4: a document with him that was several years old. There

667
00:37:07,160 --> 00:37:10,560
Speaker 4: was something I had written in twenty eighteen going over

668
00:37:10,880 --> 00:37:14,040
Speaker 4: all of the possible problems in the product that we

669
00:37:14,200 --> 00:37:17,320
Speaker 4: might need to fix, basically saying, look, this is stuff

670
00:37:17,320 --> 00:37:19,880
Speaker 4: that Google knew about in twenty eighteen and they have

671
00:37:19,960 --> 00:37:21,120
Speaker 4: done nothing to change it.

672
00:37:21,560 --> 00:37:23,120
Speaker 1: Do you think that's still true? I mean, you're not

673
00:37:23,160 --> 00:37:25,799
Speaker 1: there anymore. You haven't been there since May of last year.

674
00:37:26,239 --> 00:37:29,880
Speaker 4: June was when I was put on leave, and then

675
00:37:30,000 --> 00:37:31,640
Speaker 4: July was when they actually fired me.

676
00:37:32,000 --> 00:37:36,080
Speaker 1: Right, have there been any ongoing repercussions or consequences for you.

677
00:37:36,280 --> 00:37:38,800
Speaker 4: Oh yeah, I mean getting a job has proven difficult.

678
00:37:38,800 --> 00:37:41,719
Speaker 4: I'm finally I'm finally starting to hit stride as a

679
00:37:41,760 --> 00:37:46,000
Speaker 4: consultant doing contract work and with public speeching engagements. But

680
00:37:46,800 --> 00:37:48,960
Speaker 4: my plan was that if Google fired me, it's like,

681
00:37:49,000 --> 00:37:51,759
Speaker 4: oh well, I've got all this expertise in AI, it's

682
00:37:51,800 --> 00:37:54,000
Speaker 4: a hot market area. I'll be able to find another job,

683
00:37:54,239 --> 00:37:57,640
Speaker 4: no problem. And that was not the case at all. Apparently,

684
00:37:57,719 --> 00:38:00,440
Speaker 4: if you're willing to talk to the press, AI companies

685
00:38:00,440 --> 00:38:01,520
Speaker 4: are not willing to hire you.

686
00:38:02,160 --> 00:38:05,480
Speaker 1: In the time, we have left people who even bother

687
00:38:05,760 --> 00:38:10,919
Speaker 1: to contemplate this have a to some degree or well

688
00:38:11,080 --> 00:38:14,879
Speaker 1: in view of the future in relation to artificial intelligence,

689
00:38:15,600 --> 00:38:18,320
Speaker 1: what are your concerns give us even just a couple

690
00:38:18,320 --> 00:38:21,360
Speaker 1: of your primary concerns of where we may be headed.

691
00:38:21,880 --> 00:38:24,000
Speaker 4: Well, so, a lot of my concerns right now is

692
00:38:24,000 --> 00:38:27,840
Speaker 4: how centralized the power is. There's basically only a handful

693
00:38:27,880 --> 00:38:33,920
Speaker 4: of places where you have AI systems as powerful. Facebook

694
00:38:33,960 --> 00:38:37,560
Speaker 4: has one, Microsoft has one through open Ai, and Google

695
00:38:37,600 --> 00:38:40,640
Speaker 4: has one by do and the Chinese government have one.

696
00:38:40,880 --> 00:38:44,080
Speaker 4: But that's it, and these systems are very powerful. I

697
00:38:44,080 --> 00:38:47,960
Speaker 4: think next year you mentioned deep fakes earlier. I don't

698
00:38:48,000 --> 00:38:50,399
Speaker 4: think we've even seen the tip of the iceberg on that.

699
00:38:50,560 --> 00:38:52,799
Speaker 4: I think next year's election cycle is going to be

700
00:38:52,840 --> 00:38:57,680
Speaker 4: heavily dominated by just material generated by AI, and people

701
00:38:57,719 --> 00:38:59,319
Speaker 4: are going to have a hard time figuring out what's

702
00:38:59,320 --> 00:38:59,960
Speaker 4: real and what's not.

703
00:39:00,640 --> 00:39:03,279
Speaker 1: Do you have any tips for how people can recognize

704
00:39:03,280 --> 00:39:04,239
Speaker 1: what's really what's not?

705
00:39:04,480 --> 00:39:06,439
Speaker 4: So that's the whole point of the tearing test. Once

706
00:39:06,480 --> 00:39:10,239
Speaker 4: it gets to this point, you can't They are communicating

707
00:39:10,440 --> 00:39:13,720
Speaker 4: as well as humans are. If you do know someone

708
00:39:13,840 --> 00:39:16,439
Speaker 4: personally where you can know their speech patterns, you might

709
00:39:16,560 --> 00:39:18,799
Speaker 4: be able to tell when something's that you fake of them.

710
00:39:19,560 --> 00:39:21,840
Speaker 4: But if you don't know them that well, then you

711
00:39:21,920 --> 00:39:24,040
Speaker 4: simply won't be able to a lot of the deep

712
00:39:24,040 --> 00:39:28,520
Speaker 4: fakes you're seeing are being put together by college students

713
00:39:28,560 --> 00:39:30,520
Speaker 4: in their dorm room just trying to have a laugh.

714
00:39:31,040 --> 00:39:37,160
Speaker 4: What we haven't seen yet is deep fakes and AI

715
00:39:37,320 --> 00:39:44,680
Speaker 4: generated misinformation well funded by say, a political campaign at exactly.

716
00:39:45,280 --> 00:39:48,879
Speaker 4: We don't know how good they can get yet, because we.

717
00:39:48,880 --> 00:39:53,480
Speaker 1: Haven't seen the Steven Spielberg produce deep fakery exactly. What

718
00:39:53,560 --> 00:39:55,879
Speaker 1: else are you worried about? A lot of.

719
00:39:55,800 --> 00:39:59,800
Speaker 4: What I'm worried about right now is this becoming entrenched,

720
00:40:00,160 --> 00:40:05,960
Speaker 4: and this becoming something where one company monopolizes the technology.

721
00:40:06,160 --> 00:40:10,120
Speaker 4: There's been some good movements in a different direction, for example,

722
00:40:10,239 --> 00:40:14,520
Speaker 4: Facebook open sourcing its language model, but there's still a

723
00:40:14,560 --> 00:40:18,560
Speaker 4: lot of worry that too much power is getting concentrated

724
00:40:18,560 --> 00:40:20,279
Speaker 4: into too few hands. Beyond it.

725
00:40:20,480 --> 00:40:24,200
Speaker 1: By power, you mean information, people's personal information, a lot.

726
00:40:24,080 --> 00:40:28,880
Speaker 4: Of the ability to control public perception, because these systems

727
00:40:29,000 --> 00:40:33,560
Speaker 4: are amazing at persuasion. If nothing else, an advertisement at

728
00:40:33,600 --> 00:40:38,279
Speaker 4: its core is persuasive. What is going to happen when

729
00:40:38,600 --> 00:40:43,719
Speaker 4: these AI systems that truly know how to persuade get

730
00:40:43,800 --> 00:40:47,560
Speaker 4: put behind political ad campaigns And I'm not even talking

731
00:40:48,160 --> 00:40:51,879
Speaker 4: like illegally. Facebook is going to power its ad recommendation

732
00:40:51,960 --> 00:40:56,400
Speaker 4: algorithm with these technologies, as is Google, and then it

733
00:40:56,440 --> 00:41:00,359
Speaker 4: simply becomes who can buy the best pr representative win,

734
00:41:00,920 --> 00:41:05,320
Speaker 4: so making an already a system that's already too heavily

735
00:41:05,360 --> 00:41:09,439
Speaker 4: influenced by money becoming too heavily influenced by money and

736
00:41:09,640 --> 00:41:14,279
Speaker 4: with only a handful of gatekeepers to power. If Facebook

737
00:41:14,320 --> 00:41:18,480
Speaker 4: and Google are wonderful caretakers of democracy and they have

738
00:41:18,520 --> 00:41:20,120
Speaker 4: their hearts in the right place, and they're not going

739
00:41:20,160 --> 00:41:23,440
Speaker 4: to let the profit motive trump democratic principles, then we've

740
00:41:23,440 --> 00:41:26,200
Speaker 4: got nothing to worry about. But that seems a little

741
00:41:26,239 --> 00:41:27,000
Speaker 4: forefetched to me.

742
00:41:28,000 --> 00:41:34,319
Speaker 1: My concern is that people are influenced by this technology,

743
00:41:34,360 --> 00:41:37,759
Speaker 1: as we saw in the last election and the one

744
00:41:37,800 --> 00:41:41,719
Speaker 1: before that. You talk about, you know, advertising as persuasion,

745
00:41:42,200 --> 00:41:45,719
Speaker 1: and the political advertising had a powerful effect on the

746
00:41:45,760 --> 00:41:48,480
Speaker 1: outcome of the election. Do we need to have a

747
00:41:48,520 --> 00:41:53,279
Speaker 1: curriculum in school to teach people about how you use

748
00:41:53,360 --> 00:41:56,600
Speaker 1: this tool, so that you know, like any tool, you

749
00:41:56,640 --> 00:42:01,200
Speaker 1: don't chop your hand off, you don't injure yourself. Do

750
00:42:01,280 --> 00:42:03,799
Speaker 1: you feel that we were at a point now where

751
00:42:03,840 --> 00:42:07,040
Speaker 1: this has become dangerous for young people potentially?

752
00:42:07,800 --> 00:42:10,160
Speaker 4: So, I think one of the dangers is that we

753
00:42:10,800 --> 00:42:13,480
Speaker 4: don't need to view it as a different kind of thing.

754
00:42:13,960 --> 00:42:16,880
Speaker 4: When I was in school, we had library services classes,

755
00:42:17,400 --> 00:42:20,840
Speaker 4: and the librarians would teach us how to find credible

756
00:42:20,880 --> 00:42:25,480
Speaker 4: sources in the library. Those same basic skills that we

757
00:42:25,480 --> 00:42:28,920
Speaker 4: were taught in a physical book library as a kid.

758
00:42:29,760 --> 00:42:33,880
Speaker 4: The same skills are the ones necessary to navigate the web.

759
00:42:33,960 --> 00:42:36,600
Speaker 4: They just have to be applied in new contexts. So

760
00:42:36,680 --> 00:42:39,960
Speaker 4: I think we could easily adapt those old classes that

761
00:42:40,000 --> 00:42:43,880
Speaker 4: we're teaching kids how to use a library to apply

762
00:42:43,960 --> 00:42:47,040
Speaker 4: the same principles in teaching kids how to use the internet.

763
00:42:47,280 --> 00:42:48,280
Speaker 1: That's a great point.

764
00:42:48,600 --> 00:42:53,920
Speaker 4: Google was founded by librarians. Essentially, library sciences is what

765
00:42:54,000 --> 00:42:57,920
Speaker 4: information retrieval is based on, and that's what search indexing

766
00:42:58,040 --> 00:43:00,640
Speaker 4: is based on. It all grew out of libry sciences.

767
00:43:00,680 --> 00:43:03,560
Speaker 4: And at Google there's this kind of vibe that they

768
00:43:03,560 --> 00:43:07,960
Speaker 4: are librarians and they are curating the great library on

769
00:43:08,040 --> 00:43:08,600
Speaker 4: the Internet.

770
00:43:09,800 --> 00:43:10,480
Speaker 1: Do you have kids?

771
00:43:10,920 --> 00:43:11,160
Speaker 4: Yes?

772
00:43:11,200 --> 00:43:12,600
Speaker 1: I do you do? How many kids?

773
00:43:12,640 --> 00:43:15,600
Speaker 4: You have? Two? One boy, one girl? Fifteen and two?

774
00:43:16,520 --> 00:43:19,400
Speaker 1: Okay, so this is perfect. How do you deal with

775
00:43:19,440 --> 00:43:22,480
Speaker 1: your children with social media? Someone who knows what you know?

776
00:43:23,360 --> 00:43:26,800
Speaker 4: My son? With my daughter, it's not an issue yet.

777
00:43:26,880 --> 00:43:30,640
Speaker 4: But with my son, we basically didn't allow him to

778
00:43:30,760 --> 00:43:35,839
Speaker 4: use it until he was demonstrating a certain level of

779
00:43:35,920 --> 00:43:37,800
Speaker 4: social sophistication and maturity.

780
00:43:38,239 --> 00:43:40,080
Speaker 1: What age was that, like, thirteen?

781
00:43:40,680 --> 00:43:40,799
Speaker 2: Right?

782
00:43:41,560 --> 00:43:44,000
Speaker 1: Really you didn't allow him to have a device of

783
00:43:44,040 --> 00:43:45,640
Speaker 1: any kind will he was thirteen.

784
00:43:45,680 --> 00:43:49,200
Speaker 4: Not of his own nor he could use a computer

785
00:43:49,320 --> 00:43:51,640
Speaker 4: if he needed to use the Internet for school. But

786
00:43:51,680 --> 00:43:55,319
Speaker 4: at this point he's fifteen, and you know, he's his

787
00:43:55,360 --> 00:43:58,400
Speaker 4: own person. At this point, he more or less understands

788
00:43:58,400 --> 00:44:01,000
Speaker 4: how to navigate those spaces, say, and that's the best

789
00:44:01,040 --> 00:44:01,680
Speaker 4: I could do it.

790
00:44:01,840 --> 00:44:05,680
Speaker 1: You know what's good about artificial intelligence as far as

791
00:44:05,719 --> 00:44:08,240
Speaker 1: you're concern, what's a direction that's going in any field

792
00:44:08,600 --> 00:44:09,680
Speaker 1: that you're admiring of?

793
00:44:10,480 --> 00:44:14,080
Speaker 4: Oh? Well, I mean it's going to amplify our productivity.

794
00:44:14,120 --> 00:44:18,440
Speaker 4: Once we figure out how to you know, productively integrate

795
00:44:18,480 --> 00:44:21,959
Speaker 4: this technology into our society in a non disruptive way,

796
00:44:22,719 --> 00:44:25,160
Speaker 4: writers are going to like, I mean, I write a

797
00:44:25,239 --> 00:44:28,960
Speaker 4: decent amount, and AI is a great tool for getting

798
00:44:28,960 --> 00:44:32,560
Speaker 4: over writer's block if you need to push through something

799
00:44:32,560 --> 00:44:36,040
Speaker 4: where you're at a wall, having this other entity that

800
00:44:36,080 --> 00:44:40,719
Speaker 4: can brainstorm with you. They're great research assistants and they're

801
00:44:40,760 --> 00:44:45,200
Speaker 4: great tools for creativity. Where we're going to hit some

802
00:44:45,360 --> 00:44:49,840
Speaker 4: bumps is that they weren't built for a specific purpose.

803
00:44:49,880 --> 00:44:53,520
Speaker 4: They were built to meet Tearing's benchmarks. So now we're

804
00:44:53,560 --> 00:44:56,360
Speaker 4: gonna have to go through a phase where we figure out, Okay,

805
00:44:56,400 --> 00:44:59,720
Speaker 4: we built this thing, Now what is it actually good

806
00:44:59,760 --> 00:45:00,600
Speaker 4: for us for?

807
00:45:02,320 --> 00:45:05,240
Speaker 1: Do you think that AI has any role in solving

808
00:45:05,280 --> 00:45:07,120
Speaker 1: the climate change issue?

809
00:45:07,719 --> 00:45:10,920
Speaker 4: If AI can be used to crack something like cold fusion,

810
00:45:11,560 --> 00:45:17,960
Speaker 4: then yes, But barring those kinds of transformational technology breakthroughs.

811
00:45:18,000 --> 00:45:21,120
Speaker 4: I don't think it does, and that's because we kind

812
00:45:21,120 --> 00:45:24,160
Speaker 4: of already know what the solution to climate change is.

813
00:45:24,640 --> 00:45:25,840
Speaker 1: Change in human behavior.

814
00:45:26,120 --> 00:45:28,360
Speaker 4: Yeah, change in human behavior. We just don't want to

815
00:45:28,400 --> 00:45:33,399
Speaker 4: do it. Maybe AI could persuade people to change lifestyles,

816
00:45:33,520 --> 00:45:36,440
Speaker 4: but really, we have a certain number of people on

817
00:45:36,480 --> 00:45:40,520
Speaker 4: the planet consuming a certain amount of energy, and on

818
00:45:40,680 --> 00:45:43,839
Speaker 4: average that creates a certain amount of heat, and we

819
00:45:43,920 --> 00:45:45,799
Speaker 4: just need to change our behavior to where we're not

820
00:45:45,800 --> 00:45:49,360
Speaker 4: creating so much heat. So it is possible that AI

821
00:45:49,440 --> 00:45:52,440
Speaker 4: will give us cold fusion, in which case, yes, it

822
00:45:52,480 --> 00:45:53,479
Speaker 4: would help us with that.

823
00:45:53,960 --> 00:45:58,200
Speaker 1: Or super conductivity as well, something anything that would fundamentally

824
00:45:58,239 --> 00:46:00,400
Speaker 1: change the thermodynamics of our energy grid.

825
00:46:00,600 --> 00:46:03,560
Speaker 4: But barring a breakthrough like that, we just really do

826
00:46:03,760 --> 00:46:06,839
Speaker 4: have to focus on changing human behavior and human priorities.

827
00:46:07,400 --> 00:46:10,320
Speaker 1: Now, what's next for you? If you could do whatever

828
00:46:10,360 --> 00:46:13,560
Speaker 1: you wanted to do, you're obviously this searingly bright guy

829
00:46:14,239 --> 00:46:17,120
Speaker 1: who knows all about these important things. If you could

830
00:46:17,120 --> 00:46:20,239
Speaker 1: do whatever you wanted to do in the short term,

831
00:46:20,239 --> 00:46:22,000
Speaker 1: what would that be? Oh?

832
00:46:22,080 --> 00:46:24,560
Speaker 4: I mean, AI ethics research is what I've enjoyed doing

833
00:46:24,640 --> 00:46:26,799
Speaker 4: the most. But one of the things that I keep

834
00:46:26,840 --> 00:46:30,000
Speaker 4: getting struck by is I keep getting invited on shows

835
00:46:30,000 --> 00:46:32,440
Speaker 4: like this to speak and there's a part in the

836
00:46:32,480 --> 00:46:34,400
Speaker 4: back of my brain that just goes, am, I really

837
00:46:34,760 --> 00:46:38,080
Speaker 4: the best representative to the public on these topics, and

838
00:46:38,280 --> 00:46:41,759
Speaker 4: apparently there just aren't that many people who are technically

839
00:46:41,800 --> 00:46:44,800
Speaker 4: well versed in how these systems work who were willing

840
00:46:45,120 --> 00:46:48,360
Speaker 4: to perform this role as kind of a bridge to

841
00:46:48,360 --> 00:46:51,680
Speaker 4: the public to understanding them better. And while that's not

842
00:46:51,840 --> 00:46:54,919
Speaker 4: really the thing that I'm most passionate about in life,

843
00:46:54,920 --> 00:46:57,120
Speaker 4: it's important and the fact that I get to perform

844
00:46:57,160 --> 00:47:00,520
Speaker 4: that service to help educate people is meaning and I'm

845
00:47:00,520 --> 00:47:02,879
Speaker 4: happy to be doing it. It's and plan to continue doing

846
00:47:02,880 --> 00:47:03,399
Speaker 4: it as well.

847
00:47:04,640 --> 00:47:06,320
Speaker 1: Many thanks to you, many thanks.

848
00:47:06,640 --> 00:47:08,640
Speaker 4: It was wonderful being here. Thank you Alix, Thank you

849
00:47:08,680 --> 00:47:13,480
Speaker 4: sir My.

850
00:47:13,640 --> 00:47:18,360
Speaker 1: Thanks to Blake Lemoine and jay Lebuff. This episode was

851
00:47:18,400 --> 00:47:22,560
Speaker 1: recorded by a robot, but it was also recorded at

852
00:47:22,640 --> 00:47:27,000
Speaker 1: CDM Studios in New York City. We're produced by Kathleen Russo,

853
00:47:27,280 --> 00:47:31,040
Speaker 1: Zach MacNeice, and Maureen Hobin. Our engineer is Frank Imperial.

854
00:47:31,400 --> 00:47:35,600
Speaker 1: Our social media manager is Danielle Gingrich. I'm Alec Baldwin.

855
00:47:35,880 --> 00:47:53,840
Speaker 1: Here's the thing is brought to you by iHeart Radio,