1
00:00:01,600 --> 00:00:04,640
Speaker 1: Hey, welcome to Sign Stuff, a production of iHeartRadio. I'm

2
00:00:04,640 --> 00:00:07,480
Speaker 1: hoh hitch Cham and today we're asking the question can

3
00:00:07,520 --> 00:00:13,039
Speaker 1: AI be funny? Sure you can ask chat GPT or

4
00:00:13,160 --> 00:00:15,680
Speaker 1: Claude to tell you a joke, but are you actually

5
00:00:15,760 --> 00:00:18,720
Speaker 1: going to laugh? We're gonna talk to an AI expert

6
00:00:18,720 --> 00:00:22,159
Speaker 1: who's been obsessed with deconstructing humor and teaching it to

7
00:00:22,200 --> 00:00:25,720
Speaker 1: a computer for the last fifteen years, and she's gonna

8
00:00:25,720 --> 00:00:28,400
Speaker 1: step us through the rules of what makes something funny,

9
00:00:28,720 --> 00:00:32,479
Speaker 1: whether AI can follow them, and whether comedians have their

10
00:00:32,560 --> 00:00:38,320
Speaker 1: days number. It's a pretty interesting conversation, and I have

11
00:00:38,360 --> 00:00:40,680
Speaker 1: to warn you there's a lot of laughing at it.

12
00:00:40,680 --> 00:00:44,360
Speaker 1: It turns out AI researchers can be pretty funny, So

13
00:00:44,440 --> 00:00:46,960
Speaker 1: get ready to hear the one about the computer scientist

14
00:00:47,000 --> 00:00:49,720
Speaker 1: walks into a bar with a cartoonist as we answer

15
00:00:49,800 --> 00:00:52,760
Speaker 1: the question can AI be funny?

16
00:00:54,800 --> 00:00:55,120
Speaker 2: Enjoy?

17
00:00:57,800 --> 00:00:59,760
Speaker 1: Hey everyone, So, I don't know if you've seen this,

18
00:01:00,000 --> 00:01:03,160
Speaker 1: but the company Anthropic just released the report that shows

19
00:01:03,240 --> 00:01:05,959
Speaker 1: the jobs that are most likely to be replaced by

20
00:01:06,000 --> 00:01:11,960
Speaker 1: AI in the coming years. If you're in management, business, computers, engineering, law,

21
00:01:12,360 --> 00:01:16,640
Speaker 1: even science, it's not looking so good for you. On

22
00:01:16,680 --> 00:01:20,240
Speaker 1: the other hand, if you're in agriculture, grounds maintenance, or

23
00:01:20,240 --> 00:01:22,360
Speaker 1: if you know how to fix a fridge or ac

24
00:01:23,160 --> 00:01:27,760
Speaker 1: you might be okay, which made me wonder where comedians stand.

25
00:01:28,120 --> 00:01:30,720
Speaker 1: Some people say that comedy and humor are some of

26
00:01:30,720 --> 00:01:33,600
Speaker 1: the things that make us uniquely human, but is.

27
00:01:33,640 --> 00:01:34,360
Speaker 2: That really true?

28
00:01:34,680 --> 00:01:39,199
Speaker 1: Can AI be just as funny as people? To answer

29
00:01:39,280 --> 00:01:42,399
Speaker 1: this question, I reached out to doctor Lydia Chilton, a

30
00:01:42,480 --> 00:01:46,360
Speaker 1: professor of computer science at Columbia University who specializes in

31
00:01:46,480 --> 00:01:50,480
Speaker 1: AI and human computer interaction. She also happens to have

32
00:01:50,520 --> 00:01:54,200
Speaker 1: spent the last fifteen years trying to prove that humor

33
00:01:54,280 --> 00:01:57,480
Speaker 1: can be done by a computer, because if it can,

34
00:01:57,640 --> 00:02:02,840
Speaker 1: it means we actually understand what humor is. Now, doctor

35
00:02:02,920 --> 00:02:05,920
Speaker 1: Chiltern is gonna tell us whether or not she actually succeeded,

36
00:02:06,160 --> 00:02:08,880
Speaker 1: but first I wanted to ask her a more basic question,

37
00:02:09,200 --> 00:02:13,560
Speaker 1: which is what makes something funny? So here's my conversation

38
00:02:13,639 --> 00:02:18,680
Speaker 1: with doctor Lydia Chilton. Well, thank you doctor Chiltern for

39
00:02:18,760 --> 00:02:19,240
Speaker 1: joining us.

40
00:02:19,440 --> 00:02:21,000
Speaker 2: Thank you excited to be here.

41
00:02:21,160 --> 00:02:23,359
Speaker 1: I have a joke for you. Here we got knock.

42
00:02:23,440 --> 00:02:24,160
Speaker 2: Who's there?

43
00:02:24,440 --> 00:02:24,880
Speaker 1: Iva?

44
00:02:25,200 --> 00:02:25,920
Speaker 2: Iva? Who?

45
00:02:26,440 --> 00:02:28,560
Speaker 1: I've got a feeling I'm going to need more data

46
00:02:28,600 --> 00:02:29,399
Speaker 1: to finish this joke.

47
00:02:32,000 --> 00:02:35,280
Speaker 2: All right, you got me that's pretty funny.

48
00:02:35,720 --> 00:02:38,480
Speaker 1: Well, actually I got this one from Gemini, So I

49
00:02:38,520 --> 00:02:41,120
Speaker 1: asked Gemini right before this call to tell me a

50
00:02:41,160 --> 00:02:42,360
Speaker 1: knock knock joke about Ai.

51
00:02:42,720 --> 00:02:43,240
Speaker 2: Wow.

52
00:02:43,600 --> 00:02:46,640
Speaker 1: Yeah, go Gemini. Yeah, and you laughed.

53
00:02:46,800 --> 00:02:47,240
Speaker 2: Hi did.

54
00:02:48,960 --> 00:02:50,600
Speaker 1: Well. That brings me to the first question I have

55
00:02:50,680 --> 00:02:53,600
Speaker 1: for you, which is what makes something funny? M M.

56
00:02:54,000 --> 00:02:56,320
Speaker 1: Can you pinpoint when you started to get curious about humor?

57
00:02:56,960 --> 00:03:01,200
Speaker 2: Ah, I would say I was five years old and

58
00:03:01,680 --> 00:03:05,919
Speaker 2: I accidentally made a pun and my dad laughed so hysterically,

59
00:03:06,040 --> 00:03:08,960
Speaker 2: and I felt so good that I made my dad

60
00:03:09,120 --> 00:03:12,320
Speaker 2: last and I was like, must repeat, must repeat, And

61
00:03:12,400 --> 00:03:15,280
Speaker 2: of course it can't happen on command, and that one

62
00:03:15,400 --> 00:03:17,880
Speaker 2: was an accident. But in order to get people to

63
00:03:18,120 --> 00:03:20,800
Speaker 2: like me, I've wanted to figure out the formula for

64
00:03:20,840 --> 00:03:23,520
Speaker 2: a long time and been convinced that there is.

65
00:03:23,480 --> 00:03:27,520
Speaker 1: One, which is a very sweet reason to be interested

66
00:03:27,560 --> 00:03:28,040
Speaker 1: in the topic.

67
00:03:28,360 --> 00:03:29,200
Speaker 2: Yeah.

68
00:03:29,520 --> 00:03:32,480
Speaker 1: I read that in twenty thirteen, you went to the

69
00:03:32,480 --> 00:03:35,040
Speaker 1: internet and you try to recruit a whole bunch of

70
00:03:35,080 --> 00:03:37,720
Speaker 1: people from the Internet to give you money to try

71
00:03:37,760 --> 00:03:40,280
Speaker 1: to answer this question. Can you tell us about that?

72
00:03:40,520 --> 00:03:45,000
Speaker 2: Yeah? So there's some questions that are just so perennial

73
00:03:45,080 --> 00:03:47,520
Speaker 2: that I can't stop myself from thinking about them. This

74
00:03:47,600 --> 00:03:50,080
Speaker 2: is one you know, Plato had his set what is good,

75
00:03:50,160 --> 00:03:54,160
Speaker 2: what is justice? I've got mine what is funny? And

76
00:03:54,240 --> 00:03:56,760
Speaker 2: I actually think they're kind of the same. You know,

77
00:03:56,920 --> 00:03:58,840
Speaker 2: not to put me in Plato in the same boat,

78
00:03:58,880 --> 00:04:03,280
Speaker 2: but right, No, they both have a strong emotional component

79
00:04:03,360 --> 00:04:05,680
Speaker 2: to them. There's something about the chemicals that happen in

80
00:04:05,720 --> 00:04:09,200
Speaker 2: our brain that makes something funny. And that's why something

81
00:04:09,240 --> 00:04:11,960
Speaker 2: could be different funny for you, for me, for me

82
00:04:12,120 --> 00:04:14,600
Speaker 2: at a different time, for me if I heard it

83
00:04:14,640 --> 00:04:18,360
Speaker 2: slightly differently, maybe if someone less funny told it. So

84
00:04:18,720 --> 00:04:22,120
Speaker 2: it's not all about just the joke and whether the

85
00:04:22,320 --> 00:04:26,120
Speaker 2: joke was funny. There's a lot of other circumstances, Like

86
00:04:26,160 --> 00:04:28,360
Speaker 2: the whole thing isn't like logically constructed.

87
00:04:29,880 --> 00:04:32,200
Speaker 1: That bothers you, well, it makes it.

88
00:04:32,160 --> 00:04:34,680
Speaker 2: Very hard to understand. But like, so that could either

89
00:04:34,680 --> 00:04:36,279
Speaker 2: be a good thing or a bad thing. Like a

90
00:04:36,320 --> 00:04:37,919
Speaker 2: bad thing if you're like I need to understand this

91
00:04:37,960 --> 00:04:39,680
Speaker 2: before the test tomorrow, but a good thing. It's like,

92
00:04:39,800 --> 00:04:42,200
Speaker 2: I would like to study this for eternity because there's gonna.

93
00:04:42,000 --> 00:04:45,200
Speaker 1: Be lots of problems. We may never figure it out.

94
00:04:45,240 --> 00:04:48,200
Speaker 1: Which gives you, as a professor job security.

95
00:04:48,200 --> 00:04:52,600
Speaker 2: Exactly, exactly. So I talked to a lot of people

96
00:04:52,600 --> 00:04:54,400
Speaker 2: about this, read a lot of books, because it turns

97
00:04:54,440 --> 00:04:57,840
Speaker 2: out I wasn't the first person to study humor and

98
00:04:57,880 --> 00:05:01,120
Speaker 2: a philosopher had a theory about this is at least interesting.

99
00:05:01,320 --> 00:05:04,280
Speaker 2: It was evolved in us so that we could learn

100
00:05:04,320 --> 00:05:07,919
Speaker 2: from mistakes of others. Someone else falls on a banana

101
00:05:08,040 --> 00:05:15,120
Speaker 2: peel and liked I would not do that. And you know,

102
00:05:15,160 --> 00:05:17,160
Speaker 2: it's hard to say, but if we're looking back for

103
00:05:17,240 --> 00:05:20,599
Speaker 2: some origins of why we might have benefited as a

104
00:05:20,680 --> 00:05:24,800
Speaker 2: species from this, learning from other people's mistakes is a

105
00:05:24,800 --> 00:05:27,719
Speaker 2: pretty good way of learning rather watching someone else and

106
00:05:27,839 --> 00:05:31,039
Speaker 2: learning don't step on a banana peel as a positive reward.

107
00:05:31,320 --> 00:05:34,560
Speaker 1: Interesting. Wait, so the theory is that funnyness is basically

108
00:05:34,880 --> 00:05:35,599
Speaker 1: shot and fraud.

109
00:05:35,839 --> 00:05:41,040
Speaker 2: Yes, yes, shot and freud to make us better In theory.

110
00:05:40,960 --> 00:05:42,760
Speaker 1: Mean he liked a way for your body to feel

111
00:05:42,800 --> 00:05:45,640
Speaker 1: good at the expense of others, so that we learn

112
00:05:45,800 --> 00:05:46,279
Speaker 1: from them.

113
00:05:46,560 --> 00:05:48,679
Speaker 2: Yeah. So the next time you see a banana peel,

114
00:05:48,960 --> 00:05:52,279
Speaker 2: you have that memory because it's a strong positive You're like, oh,

115
00:05:52,360 --> 00:05:54,599
Speaker 2: that guy felt maybe I won't step on it.

116
00:05:54,720 --> 00:05:56,960
Speaker 1: That's funny. I don't want to be the person people

117
00:05:57,040 --> 00:06:01,400
Speaker 1: laugh at. Oh interesting, Okay, no one.

118
00:06:01,200 --> 00:06:03,640
Speaker 2: Can prove that's true or not. I can't go back

119
00:06:03,680 --> 00:06:06,039
Speaker 2: in time twenty thousand years and write down all the

120
00:06:06,040 --> 00:06:07,560
Speaker 2: banana pe ins.

121
00:06:07,600 --> 00:06:10,640
Speaker 1: It iss although that would be pretty funny, Like.

122
00:06:10,640 --> 00:06:14,120
Speaker 2: Where's the data, It's nowhere. It's a theory.

123
00:06:14,320 --> 00:06:17,440
Speaker 1: It is very complicated and mysterious.

124
00:06:17,040 --> 00:06:21,920
Speaker 2: Humor exactly job security, at.

125
00:06:21,920 --> 00:06:24,960
Speaker 1: Least for now until potentially AIS can do it. So

126
00:06:25,040 --> 00:06:25,599
Speaker 1: we'll get to that.

127
00:06:25,920 --> 00:06:26,160
Speaker 2: Good.

128
00:06:26,440 --> 00:06:28,919
Speaker 1: Okay, what else did you learn about what makes something funny?

129
00:06:29,080 --> 00:06:32,000
Speaker 2: Well? I read a whole bunch of theories, and there's

130
00:06:32,040 --> 00:06:33,760
Speaker 2: a lot of different ones, and they all like kind

131
00:06:33,760 --> 00:06:36,800
Speaker 2: of go together. So one is it has to be surprising.

132
00:06:37,120 --> 00:06:39,880
Speaker 2: It's not surprising. It's hard to get any kind of

133
00:06:39,880 --> 00:06:44,080
Speaker 2: emotional reaction out of it. Surprising often is correlated with

134
00:06:44,240 --> 00:06:46,920
Speaker 2: quick or has a quick turn, and it's like, oh,

135
00:06:46,960 --> 00:06:49,440
Speaker 2: I'm saying one thing and that boops, just kidding.

136
00:06:49,600 --> 00:06:51,800
Speaker 1: Somebody's walking suddenly they slip in a banana.

137
00:06:51,839 --> 00:06:55,719
Speaker 2: Whoa what exactly? Because you expected them to keep walking

138
00:06:55,880 --> 00:06:58,839
Speaker 2: and then right very suddenly if they fall very slowly,

139
00:06:58,920 --> 00:07:02,880
Speaker 2: that's less funny. Yeah, but the most important is it

140
00:07:02,920 --> 00:07:06,120
Speaker 2: has to be in the subtext of what set So

141
00:07:06,360 --> 00:07:09,680
Speaker 2: all stories, and I consider jokes stories for small stories

142
00:07:09,880 --> 00:07:12,040
Speaker 2: have like the text like what said, and then the

143
00:07:12,080 --> 00:07:15,120
Speaker 2: subtext is what it really means are what the significance

144
00:07:15,520 --> 00:07:19,760
Speaker 2: is to it. Typically jokes they have this surprise in

145
00:07:19,800 --> 00:07:22,560
Speaker 2: them because at first they leave you to believe that

146
00:07:22,600 --> 00:07:25,880
Speaker 2: there's one subtext, and then something happens you're like, well,

147
00:07:25,920 --> 00:07:27,680
Speaker 2: now there's a total other subtext.

148
00:07:28,400 --> 00:07:32,200
Speaker 1: I see. It's about our expectation about what's under the surface.

149
00:07:32,320 --> 00:07:36,320
Speaker 2: And how you're interpreting it. So my go to explainer

150
00:07:36,440 --> 00:07:40,120
Speaker 2: joke for this is here goes there are three kinds

151
00:07:40,120 --> 00:07:42,600
Speaker 2: of people in the world, people who can count and

152
00:07:42,680 --> 00:07:47,400
Speaker 2: people who can't. So I led you to believe that

153
00:07:47,480 --> 00:07:49,960
Speaker 2: I was a person who is intelligent and had three

154
00:07:50,000 --> 00:07:52,920
Speaker 2: things to say, and then my list too, and you're like, wait,

155
00:07:53,120 --> 00:07:56,360
Speaker 2: what in your brain kind of rearranges and then you realize, oh,

156
00:07:56,400 --> 00:08:01,160
Speaker 2: she can't count, she's an idiot, haha, and that's what

157
00:08:01,200 --> 00:08:03,080
Speaker 2: makes it funny. So those are the two texts.

158
00:08:03,120 --> 00:08:05,720
Speaker 1: Yeah, I was expecting three things. I was surprised when

159
00:08:05,760 --> 00:08:08,640
Speaker 1: you stopped it two. But then in my brain pieced

160
00:08:08,640 --> 00:08:11,680
Speaker 1: it together. Oh, it's because she said she must be

161
00:08:11,720 --> 00:08:14,720
Speaker 1: one of the people who can't count exactly.

162
00:08:14,440 --> 00:08:19,040
Speaker 2: So there's some expectation violation in their expectation violation is

163
00:08:19,120 --> 00:08:21,520
Speaker 2: one of the main theories, but it's not sufficient. If

164
00:08:21,520 --> 00:08:24,840
Speaker 2: I just said purple, you weren't expecting me to say purple.

165
00:08:24,880 --> 00:08:28,920
Speaker 2: But it's not funny, So the subtext is very very important. Okay,

166
00:08:29,080 --> 00:08:33,200
Speaker 2: that said, farts are also funny, and that doesn't Uh,

167
00:08:33,240 --> 00:08:36,640
Speaker 2: it's hard to explain that with samantic script theory of humor.

168
00:08:36,720 --> 00:08:39,240
Speaker 2: I don't know what the subtext the part.

169
00:08:39,200 --> 00:08:45,520
Speaker 1: Is, Well, definitely they're funny, especially if you're ten years old. Yes,

170
00:08:45,800 --> 00:08:48,360
Speaker 1: maybe the subtext is that you're not supposed to do

171
00:08:48,440 --> 00:08:52,720
Speaker 1: it in public for others to hear or smell. Yeah, okay,

172
00:08:52,760 --> 00:08:54,760
Speaker 1: this is going to be a little important later on

173
00:08:54,840 --> 00:08:56,920
Speaker 1: when we talk about how to get a computer to

174
00:08:56,960 --> 00:08:59,960
Speaker 1: be funny. One of the most popular theories about humor

175
00:09:00,360 --> 00:09:03,800
Speaker 1: is that it violates our expectations in a surprising but

176
00:09:03,840 --> 00:09:07,520
Speaker 1: not unrelated Wait, so the first part of the joke

177
00:09:07,760 --> 00:09:10,920
Speaker 1: makes us assume one context, but then something happens and

178
00:09:10,960 --> 00:09:13,440
Speaker 1: it turns out there was a second hidden way to

179
00:09:13,520 --> 00:09:16,920
Speaker 1: interpret what was happening, which we had missed. Of course,

180
00:09:17,040 --> 00:09:19,800
Speaker 1: That's not the only thing that makes something funny.

181
00:09:21,640 --> 00:09:23,800
Speaker 2: There's a lot of other things that make a joke

182
00:09:24,000 --> 00:09:27,760
Speaker 2: more or less funny. Okay, I call them heighteners. They

183
00:09:27,800 --> 00:09:31,040
Speaker 2: just heighten the surprise or the level of funny or

184
00:09:31,080 --> 00:09:34,280
Speaker 2: something about the emotion with them. So one is being mean,

185
00:09:34,559 --> 00:09:37,840
Speaker 2: Like typically the more mean a joke is the funnier

186
00:09:37,920 --> 00:09:39,960
Speaker 2: it is until it like really chips over and it's

187
00:09:40,000 --> 00:09:41,920
Speaker 2: like too soon or like or too.

188
00:09:41,760 --> 00:09:44,200
Speaker 1: Mean, meaning like somebody has to be the butt of

189
00:09:44,240 --> 00:09:44,640
Speaker 1: the joke.

190
00:09:44,960 --> 00:09:47,200
Speaker 2: Yeah, so mine was mean. I was the butt of

191
00:09:47,200 --> 00:09:49,960
Speaker 2: my own joke that I can't count. Uh, it's funnier

192
00:09:50,000 --> 00:09:50,600
Speaker 2: if it's mean.

193
00:09:51,320 --> 00:09:55,240
Speaker 1: It has to say it expose somebody, some person, Yeah, right,

194
00:09:55,360 --> 00:09:58,559
Speaker 1: Like it can't be about exposing a chair or exposing

195
00:09:58,600 --> 00:10:01,079
Speaker 1: a material or or something nerds.

196
00:10:00,920 --> 00:10:03,760
Speaker 2: Something related to people. It could be like a group

197
00:10:03,800 --> 00:10:07,480
Speaker 2: of people, like women or men. There's a lot there's

198
00:10:07,480 --> 00:10:10,280
Speaker 2: a lot of jokes that take that persuasion.

199
00:10:09,760 --> 00:10:12,600
Speaker 1: And that movie pats into that shout and Freud shot

200
00:10:12,679 --> 00:10:13,400
Speaker 1: and Freud.

201
00:10:13,400 --> 00:10:15,960
Speaker 2: Yes, schotenfreud of. I don't know it's German. Who knows

202
00:10:16,000 --> 00:10:18,720
Speaker 2: how to pronounce it? Probably nobody. It's not like there's

203
00:10:18,760 --> 00:10:19,640
Speaker 2: a country.

204
00:10:19,240 --> 00:10:22,480
Speaker 1: Of people that and that would be a fandfire by

205
00:10:22,520 --> 00:10:25,640
Speaker 1: my lacke of language skills. But it's like you said,

206
00:10:25,640 --> 00:10:27,640
Speaker 1: it's about learning from the mistakes of others.

207
00:10:27,840 --> 00:10:30,480
Speaker 2: Yes, exactly interesting, So.

208
00:10:30,520 --> 00:10:32,080
Speaker 1: That makes jokes even funnier.

209
00:10:32,280 --> 00:10:35,160
Speaker 2: Yeah. Another thing that makes them funnier is when you

210
00:10:35,280 --> 00:10:39,120
Speaker 2: really relate to them, and particularly in an in crowd

211
00:10:39,280 --> 00:10:41,400
Speaker 2: kind of way, when you're like, oh, I get back.

212
00:10:41,440 --> 00:10:44,160
Speaker 2: So when someone tells, like a computer science joke that

213
00:10:44,240 --> 00:10:46,440
Speaker 2: I'm like, ooh, I get it, and I feel special

214
00:10:46,440 --> 00:10:50,440
Speaker 2: because the physicists aren't going to get it. And this

215
00:10:50,480 --> 00:10:54,840
Speaker 2: can be whether it's about our generation, our family, or anything,

216
00:10:55,440 --> 00:10:59,800
Speaker 2: but inside jokes definitely produce a special kind of chemical

217
00:10:59,840 --> 00:11:02,199
Speaker 2: that make us feel like we're part of something, and

218
00:11:02,240 --> 00:11:03,120
Speaker 2: that heightens it.

219
00:11:03,400 --> 00:11:04,400
Speaker 1: Yeah, that's funnier.

220
00:11:04,760 --> 00:11:09,040
Speaker 2: It is funny or more emotional, and that heightens the funny.

221
00:11:08,720 --> 00:11:12,240
Speaker 1: Because it exposes something you didn't think anybody else knew,

222
00:11:12,520 --> 00:11:13,520
Speaker 1: something you should talk about.

223
00:11:13,679 --> 00:11:17,480
Speaker 2: So yeah, part of it is, and Freud talked about this,

224
00:11:17,880 --> 00:11:21,120
Speaker 2: a part of it is the relief of being able

225
00:11:21,160 --> 00:11:23,880
Speaker 2: to say something that you might not have otherwise been

226
00:11:23,920 --> 00:11:27,200
Speaker 2: able to say or reveal. You know, in society we're

227
00:11:27,600 --> 00:11:29,319
Speaker 2: just to go around and show our best face to

228
00:11:29,400 --> 00:11:31,880
Speaker 2: people pretend that we're awesome, and to have Instagram pages

229
00:11:31,880 --> 00:11:34,320
Speaker 2: that makes us look like we're in vacation Italy half

230
00:11:34,360 --> 00:11:37,400
Speaker 2: the time. And so when you realize that someone else

231
00:11:37,520 --> 00:11:40,360
Speaker 2: is also not actually going to Italy, they took one

232
00:11:40,440 --> 00:11:43,640
Speaker 2: vacation and been dripping the photos throughout the year that yeah,

233
00:11:43,840 --> 00:11:46,480
Speaker 2: my Instagram page is a lie, and that sense of

234
00:11:46,600 --> 00:11:50,360
Speaker 2: relief to people that yes, I'm like that too. It

235
00:11:50,440 --> 00:11:53,880
Speaker 2: rings true. Something being true like it also heightens it.

236
00:11:53,920 --> 00:11:56,360
Speaker 2: There's like a satisfaction and be like, yeah, that's how

237
00:11:56,400 --> 00:12:00,160
Speaker 2: it really is. That person's telling it straight. This was

238
00:12:00,240 --> 00:12:03,600
Speaker 2: like a mental satisfaction of when you hear something that's true.

239
00:12:03,559 --> 00:12:06,800
Speaker 1: Right right, Like it puts a voice to something you've

240
00:12:06,840 --> 00:12:11,240
Speaker 1: suspected or felt but never actually maybe put into words before.

241
00:12:11,480 --> 00:12:13,880
Speaker 2: Yeah. Right, And so there's still like a subtext going

242
00:12:13,880 --> 00:12:17,160
Speaker 2: on there, but the subtext that you're hearing is like, oh, yes,

243
00:12:17,280 --> 00:12:20,319
Speaker 2: I always wanted someone to admit that airline food is bad.

244
00:12:20,679 --> 00:12:22,840
Speaker 2: I've never been able to express that before.

245
00:12:24,600 --> 00:12:26,240
Speaker 1: I see keep quing, keep quick.

246
00:12:26,400 --> 00:12:29,080
Speaker 2: Since we've already opened up the topic of fart jokes,

247
00:12:29,120 --> 00:12:33,840
Speaker 2: things being dirty, sexual body related, all the things that

248
00:12:33,920 --> 00:12:36,120
Speaker 2: swear words have in common. There's a special part of

249
00:12:36,160 --> 00:12:38,920
Speaker 2: our brain that is like reserved for like swear words

250
00:12:38,920 --> 00:12:42,560
Speaker 2: and taboo things. Anytime you light that up, it also

251
00:12:42,760 --> 00:12:46,520
Speaker 2: sort of magnifies anything, whether it's a joke or a

252
00:12:46,559 --> 00:12:49,920
Speaker 2: compliment or anything else. It's just a heightener of all kinds.

253
00:12:50,080 --> 00:12:53,920
Speaker 2: Extra brain cells start firing when you hear those kinds

254
00:12:53,960 --> 00:12:57,640
Speaker 2: of words. M So try adding a swear word.

255
00:12:58,920 --> 00:13:04,959
Speaker 1: I was gonna say, that is so effing true.

256
00:13:03,040 --> 00:13:06,320
Speaker 2: And that was funny because you did that. There's many

257
00:13:06,360 --> 00:13:10,679
Speaker 2: things that heighten our emotions or feelings of something. Some

258
00:13:10,720 --> 00:13:14,280
Speaker 2: people think certain letters and words are very funny. So

259
00:13:14,559 --> 00:13:17,440
Speaker 2: words with more k's in them will make people laugh more.

260
00:13:18,280 --> 00:13:22,680
Speaker 2: Maybe I just where's the data? Yeah, there's many many

261
00:13:22,720 --> 00:13:26,280
Speaker 2: hypotheses out there waiting to be tested about what's funny? Yeah,

262
00:13:26,320 --> 00:13:29,080
Speaker 2: about what heightens jokes, what makes them funny? Like exactly

263
00:13:29,120 --> 00:13:31,600
Speaker 2: what kind of subtext is and is not funny? Where

264
00:13:31,600 --> 00:13:34,640
Speaker 2: in the exact turn is all right?

265
00:13:35,120 --> 00:13:38,200
Speaker 1: So, doctor Chilton spent years studying humor and what makes

266
00:13:38,240 --> 00:13:40,959
Speaker 1: things funny. She talked to comedians and tapped into the

267
00:13:41,040 --> 00:13:44,640
Speaker 1: humor science community. Yes there is one, and she came

268
00:13:44,679 --> 00:13:47,200
Speaker 1: out of it with a bunch of rules about comedy.

269
00:13:47,600 --> 00:13:49,400
Speaker 1: The next step was to see if she could get

270
00:13:49,440 --> 00:13:53,960
Speaker 1: a computer to follow those rules. Could an AI be funny?

271
00:13:54,800 --> 00:13:57,120
Speaker 1: So when we come back, we'll talk about the different

272
00:13:57,160 --> 00:13:59,680
Speaker 1: ways in which programmers have tried to do that, and

273
00:13:59,720 --> 00:14:02,760
Speaker 1: with a the joke is on AIS or on us,

274
00:14:03,240 --> 00:14:18,840
Speaker 1: So stay with us for the punchline. We'll be right back. Hey,

275
00:14:18,880 --> 00:14:22,560
Speaker 1: welcome back. We're talking about whether AI can be funny,

276
00:14:22,840 --> 00:14:26,440
Speaker 1: and our guest today is doctor Lydia Chilton. As we

277
00:14:26,480 --> 00:14:29,480
Speaker 1: mentioned before, in twenty thirteen, she started a project to

278
00:14:29,640 --> 00:14:33,760
Speaker 1: dissect what makes something funny, not only to understand humor

279
00:14:33,880 --> 00:14:36,760
Speaker 1: to satisfy her own curiosity, but to see if you

280
00:14:36,760 --> 00:14:40,360
Speaker 1: could get a computer to be funny. However, not everyone

281
00:14:40,400 --> 00:14:42,040
Speaker 1: thought it was a good idea.

282
00:14:43,320 --> 00:14:44,760
Speaker 2: Oh I got a lot of hate mail too, but

283
00:14:45,000 --> 00:14:46,960
Speaker 2: it's the Internet. Of course you're gonna get hate mail.

284
00:14:47,600 --> 00:14:51,480
Speaker 2: I'm like, how dare you? I thought that was interesting too,

285
00:14:51,560 --> 00:14:55,440
Speaker 2: that people be offended by this, and I'm like, it's science,

286
00:14:56,160 --> 00:14:58,120
Speaker 2: you know. So for me, that was the first tip

287
00:14:58,160 --> 00:15:02,880
Speaker 2: of people being threatened by AI, especially when computers creep

288
00:15:02,880 --> 00:15:05,680
Speaker 2: into creative areas. I think people kind of get this,

289
00:15:05,800 --> 00:15:08,120
Speaker 2: but that's mine. I'm a human. I'm creative. That's part

290
00:15:08,160 --> 00:15:08,800
Speaker 2: of my identity.

291
00:15:08,960 --> 00:15:12,440
Speaker 1: Oh, you got pushback for even kind of trying to

292
00:15:12,520 --> 00:15:15,560
Speaker 1: dissect it, maybe with the goal of getting computers to

293
00:15:15,600 --> 00:15:15,880
Speaker 1: do it.

294
00:15:16,200 --> 00:15:18,640
Speaker 2: Yeah, And I was just saying, well, it's just dissecting it,

295
00:15:18,680 --> 00:15:20,680
Speaker 2: but like, come on, if you dissect it well enough,

296
00:15:20,720 --> 00:15:22,960
Speaker 2: there's a very high probability that you'll be able to

297
00:15:23,320 --> 00:15:26,360
Speaker 2: generate it. But these weren't even like honestly comedians were

298
00:15:26,400 --> 00:15:28,600
Speaker 2: more like, yeah, I kind of want to know this too.

299
00:15:29,400 --> 00:15:30,880
Speaker 2: They weren't just threatened by it.

300
00:15:33,440 --> 00:15:37,480
Speaker 1: Okay. So despite this pushback, doctor Chilton pressed on. But

301
00:15:37,720 --> 00:15:41,400
Speaker 1: she needed data on humor, so she turned to The Onion,

302
00:15:41,720 --> 00:15:45,120
Speaker 1: the humor and satire magazine that's been published since nineteen

303
00:15:45,200 --> 00:15:45,680
Speaker 1: eighty eight.

304
00:15:46,840 --> 00:15:49,760
Speaker 2: And so someone pointed out to me that The Onion,

305
00:15:49,880 --> 00:15:53,160
Speaker 2: which is known for making up really funny fake headlines

306
00:15:53,240 --> 00:15:57,000
Speaker 2: and news stories, also had a different section that was unusual.

307
00:15:57,040 --> 00:15:59,960
Speaker 2: They took a real headline and came up with three

308
00:16:00,520 --> 00:16:04,560
Speaker 2: funny man on the street responses to it from average,

309
00:16:04,760 --> 00:16:09,520
Speaker 2: usually idiotic Americans. And I liked this as a framework

310
00:16:09,600 --> 00:16:11,760
Speaker 2: for studying the joke because you could just take the

311
00:16:11,840 --> 00:16:16,040
Speaker 2: input the original headline, and then you had the Onions

312
00:16:16,080 --> 00:16:20,400
Speaker 2: like verifiably funny statements, and then you could, with the

313
00:16:20,480 --> 00:16:23,240
Speaker 2: same headline, come up with your own or try to

314
00:16:23,320 --> 00:16:26,880
Speaker 2: back engineer the jokes that the onion made to figure out, Okay,

315
00:16:26,920 --> 00:16:29,120
Speaker 2: what are some of the properties of these jokes, and

316
00:16:29,160 --> 00:16:31,160
Speaker 2: what are some of the techniques for doing them, and

317
00:16:31,200 --> 00:16:34,200
Speaker 2: what's the variety in them. So this little test bed

318
00:16:34,600 --> 00:16:38,160
Speaker 2: became very important as just a mechanism for which I

319
00:16:38,160 --> 00:16:39,680
Speaker 2: could sort of study humor in a.

320
00:16:39,640 --> 00:16:42,040
Speaker 1: Bottle and said, what did you learn that experiment you

321
00:16:42,040 --> 00:16:43,280
Speaker 1: did when in twenty.

322
00:16:43,640 --> 00:16:48,200
Speaker 2: Twenty thirteen, twenty fourteen, twenty fifteen. It kept going because

323
00:16:48,240 --> 00:16:51,200
Speaker 2: there was a lot to learn, as it turned out,

324
00:16:51,840 --> 00:16:54,680
Speaker 2: But we did find some things that all the jokes had.

325
00:16:54,880 --> 00:16:58,840
Speaker 2: All the jokes had at least two connections to the headline.

326
00:16:59,720 --> 00:17:03,320
Speaker 2: They were taking the original headline and taking the kind

327
00:17:03,320 --> 00:17:08,520
Speaker 2: of associating things with those entities, with the people mentioned

328
00:17:08,560 --> 00:17:11,320
Speaker 2: in them, and finding new things to say about them

329
00:17:11,440 --> 00:17:13,080
Speaker 2: and a new connection between them.

330
00:17:13,320 --> 00:17:15,840
Speaker 1: I see, like you're saying, like finding that other subtics

331
00:17:15,840 --> 00:17:17,159
Speaker 1: that is surprising.

332
00:17:17,040 --> 00:17:20,679
Speaker 2: Exactly, And then I was like, uh duh. As a

333
00:17:20,680 --> 00:17:23,600
Speaker 2: computer scientist, I realized I'd kind of been looking in

334
00:17:23,640 --> 00:17:26,040
Speaker 2: the wrong places. I had been looking for like the

335
00:17:26,160 --> 00:17:29,879
Speaker 2: logic behind jokes. But I quickly realized that jokes and

336
00:17:30,000 --> 00:17:34,879
Speaker 2: most human communication is about our loose associations, like why

337
00:17:35,000 --> 00:17:37,520
Speaker 2: when I think McDonald's, why do I think Burger King?

338
00:17:37,600 --> 00:17:40,719
Speaker 2: Why do I It's just like they're in the same category. There,

339
00:17:40,760 --> 00:17:44,040
Speaker 2: It's an association I have. And at the time, computer

340
00:17:44,119 --> 00:17:47,199
Speaker 2: science was all logic, and so I actually kind of

341
00:17:47,240 --> 00:17:49,800
Speaker 2: put the project on a shelf. But I realized, Okay,

342
00:17:49,840 --> 00:17:51,840
Speaker 2: we need an association engine.

343
00:17:52,440 --> 00:17:54,280
Speaker 1: Okay, let me see if I get this. It was

344
00:17:54,480 --> 00:17:58,919
Speaker 1: like the twenty early twenty tens. You dissected what humor is.

345
00:17:58,960 --> 00:18:01,560
Speaker 1: You sort of found all the these patterns and.

346
00:18:01,480 --> 00:18:03,080
Speaker 2: Ye by analyzing the onion.

347
00:18:03,440 --> 00:18:07,280
Speaker 1: By analyzing you peeled back the layers of the onion. Yeah,

348
00:18:10,119 --> 00:18:13,000
Speaker 1: you found the nuggets, some nuggets and rules, some patterns

349
00:18:13,080 --> 00:18:15,320
Speaker 1: about what makes something funny. And your goal was to

350
00:18:15,400 --> 00:18:17,880
Speaker 1: maybe try to get a computer to do this, right.

351
00:18:18,200 --> 00:18:19,080
Speaker 2: Yeah, But it couldn't.

352
00:18:19,280 --> 00:18:23,000
Speaker 1: It couldn't at the time because I think computer science

353
00:18:23,040 --> 00:18:25,560
Speaker 1: back then and artificial intelligence back then was kind of

354
00:18:25,600 --> 00:18:27,800
Speaker 1: about setting up the right rules, right.

355
00:18:28,280 --> 00:18:30,920
Speaker 2: Yeah, it was all about logic. It was like, what

356
00:18:30,960 --> 00:18:33,119
Speaker 2: are the logical rules to do this? How do I

357
00:18:33,160 --> 00:18:35,520
Speaker 2: add one plus one. I take this bit and I

358
00:18:35,640 --> 00:18:37,840
Speaker 2: combine it with this bit, and I do an and

359
00:18:37,840 --> 00:18:40,960
Speaker 2: and that creates two. All of that kind of stuff,

360
00:18:42,000 --> 00:18:44,560
Speaker 2: But no, like, what do you associate with one? Oh?

361
00:18:44,680 --> 00:18:47,960
Speaker 2: One is the loneliest number number one, We're number one,

362
00:18:48,440 --> 00:18:51,080
Speaker 2: Avis is number two. Those are the thoughts that people

363
00:18:51,200 --> 00:18:54,640
Speaker 2: have about one that computers have no idea. And it's

364
00:18:54,680 --> 00:18:57,399
Speaker 2: even hard to get from the Internet, I see because

365
00:18:57,400 --> 00:19:00,359
Speaker 2: it's the things that people say and not necessarily really

366
00:19:00,400 --> 00:19:04,160
Speaker 2: the things that people write down. And it's about frequency,

367
00:19:04,400 --> 00:19:06,720
Speaker 2: like how often it happens, rather than the fact that

368
00:19:06,800 --> 00:19:07,679
Speaker 2: it did happen.

369
00:19:08,480 --> 00:19:10,800
Speaker 1: Okay, okay, I am getting here to the nugget of

370
00:19:10,800 --> 00:19:13,720
Speaker 1: the onion here. You sort of figured out that humor

371
00:19:14,040 --> 00:19:17,080
Speaker 1: was about starting with a subtext that people would recognize,

372
00:19:17,119 --> 00:19:20,240
Speaker 1: but then being able to find that secondary that other

373
00:19:20,440 --> 00:19:23,080
Speaker 1: subtexts that will be surprising when you put all the

374
00:19:23,119 --> 00:19:24,480
Speaker 1: pieces together in the joke.

375
00:19:24,400 --> 00:19:26,200
Speaker 2: Yes, surprising, but still relevant.

376
00:19:26,280 --> 00:19:29,879
Speaker 1: It's still relevant, that's right. Second, suptics and trying to

377
00:19:29,960 --> 00:19:33,600
Speaker 1: find those subtags is hard if you're just going by

378
00:19:33,640 --> 00:19:36,639
Speaker 1: the literal definition of worth and using logic, right and

379
00:19:36,680 --> 00:19:37,119
Speaker 1: things like that.

380
00:19:37,160 --> 00:19:38,520
Speaker 2: Exactly, it's just not there.

381
00:19:39,400 --> 00:19:41,840
Speaker 1: So you start said we can't do this with rules

382
00:19:41,840 --> 00:19:43,080
Speaker 1: and logic, so you shelved it.

383
00:19:43,320 --> 00:19:44,199
Speaker 2: Yeah, I gave up.

384
00:19:44,240 --> 00:19:47,640
Speaker 1: People weren't convinced that computers could be funny.

385
00:19:47,720 --> 00:19:51,639
Speaker 2: Yes, they weren't convinced that I had decomposed and reconstructed

386
00:19:51,680 --> 00:19:55,360
Speaker 2: the process of writing humor when humans were still involved

387
00:19:55,359 --> 00:19:56,880
Speaker 2: in some part of that process.

388
00:19:57,200 --> 00:20:00,200
Speaker 1: Oh, I see, because you still needed some human to

389
00:20:00,240 --> 00:20:01,400
Speaker 1: direct the computer.

390
00:20:01,280 --> 00:20:04,480
Speaker 2: Like come up with the associations, which is fair, but

391
00:20:04,560 --> 00:20:07,200
Speaker 2: they're absolutely correct. And then I put this on hold

392
00:20:07,280 --> 00:20:09,840
Speaker 2: for forever. I was like, never again. I've been burned

393
00:20:09,880 --> 00:20:12,720
Speaker 2: by this. I will never again grace the universe with

394
00:20:12,880 --> 00:20:14,160
Speaker 2: my thoughts on humor.

395
00:20:14,440 --> 00:20:16,080
Speaker 1: Humor, this is not funny anymore.

396
00:20:16,560 --> 00:20:18,400
Speaker 2: Exactly, this isn't funny anymore.

397
00:20:19,760 --> 00:20:22,439
Speaker 1: Yes, at this point, computer scientists thought maybe it was

398
00:20:22,520 --> 00:20:26,239
Speaker 1: impossible to really teach a computer to be funny. But

399
00:20:26,280 --> 00:20:29,760
Speaker 1: then something changed, something happened that made them think that

400
00:20:29,920 --> 00:20:32,800
Speaker 1: maybe it is possible for an AI to have a

401
00:20:32,880 --> 00:20:36,600
Speaker 1: sense of humor. So when we come back, we're gonna

402
00:20:36,640 --> 00:20:39,960
Speaker 1: talk about what that change was and how it unlocks

403
00:20:40,000 --> 00:20:43,919
Speaker 1: AI's funny bone forever. So stay with us. We'll be

404
00:20:44,000 --> 00:20:58,400
Speaker 1: right back. Hey, we'll come back. We're talking about whether

405
00:20:58,440 --> 00:21:01,159
Speaker 1: AI can be funny, and we're at the part of

406
00:21:01,200 --> 00:21:04,600
Speaker 1: the story where computer scientists didn't think it was possible

407
00:21:04,760 --> 00:21:07,800
Speaker 1: for a computer to make humor. There have been lots

408
00:21:07,840 --> 00:21:11,800
Speaker 1: of attempts to have computers recognize humor or jokes basically,

409
00:21:12,280 --> 00:21:15,400
Speaker 1: and they could do pretty well looking for language cues

410
00:21:15,480 --> 00:21:20,280
Speaker 1: like incongruities or alteration or slang. But could a computer

411
00:21:20,440 --> 00:21:23,879
Speaker 1: come up with a joke? That was the big question.

412
00:21:24,720 --> 00:21:26,960
Speaker 1: Now to understand what was happening at this point with

413
00:21:27,119 --> 00:21:29,320
Speaker 1: AI and humor, you kind of need to know a

414
00:21:29,320 --> 00:21:32,679
Speaker 1: little bit of the history of AI. Recovered this in

415
00:21:32,720 --> 00:21:35,200
Speaker 1: a lot of detail in our February fourth episode about

416
00:21:35,280 --> 00:21:37,960
Speaker 1: what AI slop is doing to us, So if you

417
00:21:38,000 --> 00:21:40,679
Speaker 1: want to dig deeper, go check out that episode. But

418
00:21:40,720 --> 00:21:43,240
Speaker 1: the basic idea is that for a long time, the

419
00:21:43,280 --> 00:21:46,919
Speaker 1: field of AI was largely based on logic, trying to

420
00:21:46,920 --> 00:21:51,240
Speaker 1: figure out strategies and algorithms that we thought made things intelligent.

421
00:21:51,720 --> 00:21:54,600
Speaker 1: But then in the mid twenty tens, computer scientists figured

422
00:21:54,640 --> 00:21:58,720
Speaker 1: out a different approach to AI that changed everything. They

423
00:21:58,720 --> 00:22:02,679
Speaker 1: started to use something new called the transformer, which in

424
00:22:02,760 --> 00:22:07,240
Speaker 1: turn you sink technique called attention that essentially let computers

425
00:22:07,480 --> 00:22:11,200
Speaker 1: learn context, and that led to the explosion of AI

426
00:22:11,280 --> 00:22:16,119
Speaker 1: systems like Chat, GPT, Gemini, Claude. And the key ability

427
00:22:16,160 --> 00:22:19,840
Speaker 1: here for humor is that these AI systems were built

428
00:22:20,000 --> 00:22:23,600
Speaker 1: to make associations here tector, Lydia, chiltern.

429
00:22:25,280 --> 00:22:29,320
Speaker 2: So these language models GPT, claw, you know, the thing,

430
00:22:29,440 --> 00:22:32,280
Speaker 2: the ais that can generate text sort of taken over

431
00:22:32,320 --> 00:22:34,680
Speaker 2: the universe at this point, for better or for worse,

432
00:22:34,880 --> 00:22:37,720
Speaker 2: are called man, do I even know what it stands for? Yes?

433
00:22:37,760 --> 00:22:41,160
Speaker 2: I do. Large language models, so they take basically take

434
00:22:41,200 --> 00:22:44,320
Speaker 2: in all the text on the internet. They take a sentence,

435
00:22:44,560 --> 00:22:46,359
Speaker 2: they take the last word out of the sentence and

436
00:22:46,400 --> 00:22:49,480
Speaker 2: try to predict what that last word would be and

437
00:22:49,560 --> 00:22:52,520
Speaker 2: guess what that is. That's an association. They go, you

438
00:22:52,560 --> 00:22:56,560
Speaker 2: can't do that by logic. You just have to say, like, uh, goodbye,

439
00:22:56,640 --> 00:22:59,840
Speaker 2: so long don't have you know fun or like have

440
00:23:00,440 --> 00:23:02,840
Speaker 2: like that's an association. There's no logic that tells you

441
00:23:02,840 --> 00:23:04,800
Speaker 2: you should do You've just heard it many times before,

442
00:23:05,200 --> 00:23:09,040
Speaker 2: you repeat it. Kids especially pick up on these things.

443
00:23:08,840 --> 00:23:11,040
Speaker 2: It's it's just like, well, our brains are hardwired to

444
00:23:11,080 --> 00:23:13,280
Speaker 2: find these associations and use them.

445
00:23:13,480 --> 00:23:15,399
Speaker 1: So your team was like, we can do it. What

446
00:23:15,600 --> 00:23:18,840
Speaker 1: was the thing that specifically they thought they could do,

447
00:23:19,080 --> 00:23:20,000
Speaker 1: or that you could all do.

448
00:23:20,359 --> 00:23:24,239
Speaker 2: Well, we knew that AI could do the associations. What

449
00:23:24,320 --> 00:23:28,439
Speaker 2: I always told them is we need a new evaluation mechanism.

450
00:23:28,600 --> 00:23:31,560
Speaker 2: The American Voices section of the Onion that I had

451
00:23:31,640 --> 00:23:35,080
Speaker 2: used was no longer quite as popular because this was

452
00:23:35,160 --> 00:23:37,760
Speaker 2: like fifteen years ago. So like, yeah, millennials loved it.

453
00:23:37,960 --> 00:23:41,720
Speaker 2: Gen Z just does not care. The student was as

454
00:23:41,800 --> 00:23:43,840
Speaker 2: gen Z as they come, and he's like, well, I

455
00:23:43,880 --> 00:23:45,879
Speaker 2: know what's funny. Here's what people do. They have a

456
00:23:45,880 --> 00:23:49,240
Speaker 2: caption contest on Instagram. People post a funny image and

457
00:23:49,280 --> 00:23:53,040
Speaker 2: then people try to caption it. We're like, okay, let's

458
00:23:53,040 --> 00:23:55,560
Speaker 2: go for it. Very similar to what the New Yorker does,

459
00:23:55,720 --> 00:23:58,760
Speaker 2: but the New Yorker's a hard to capture. And Sean

460
00:23:58,960 --> 00:24:02,879
Speaker 2: convinced me that j has a very particular flavor of humor.

461
00:24:04,040 --> 00:24:06,159
Speaker 1: So the type of humor that doctor Chiltern and her

462
00:24:06,160 --> 00:24:08,960
Speaker 1: team decided to see if AI could make was gen

463
00:24:09,080 --> 00:24:12,720
Speaker 1: Z meme humor. That's when you see an image, let's say,

464
00:24:12,880 --> 00:24:16,080
Speaker 1: a person with a small hose trying to put out

465
00:24:16,080 --> 00:24:19,800
Speaker 1: a really large fire, and then someone writes underneath that

466
00:24:19,840 --> 00:24:22,479
Speaker 1: image to text me trying to put out the dumpster

467
00:24:22,520 --> 00:24:25,919
Speaker 1: fire of my last relationship. If you're a gen Z

468
00:24:26,320 --> 00:24:30,480
Speaker 1: that would be hilarious. Okay, here's the experiment, Doctor Chiltern

469
00:24:30,520 --> 00:24:33,199
Speaker 1: and her team there, they took images and then they

470
00:24:33,240 --> 00:24:37,679
Speaker 1: put funny captions to them from three sources. One was

471
00:24:37,880 --> 00:24:41,080
Speaker 1: real people. The images were put online and then people

472
00:24:41,119 --> 00:24:43,480
Speaker 1: competed to see who could come up with the funniest

473
00:24:43,520 --> 00:24:46,520
Speaker 1: caption for them. Where we got the most votes, that's

474
00:24:46,560 --> 00:24:50,520
Speaker 1: the one that represented how funny humans can be. The

475
00:24:50,600 --> 00:24:54,280
Speaker 1: second source was Chad GPT. They just asked Chad JPT

476
00:24:54,520 --> 00:24:56,760
Speaker 1: to come up a funny caption for the image for

477
00:24:56,840 --> 00:24:59,440
Speaker 1: a gen Z audience. But then there was a third

478
00:24:59,480 --> 00:25:02,760
Speaker 1: source of funny captions, which was Chad GBT, but with

479
00:25:02,920 --> 00:25:06,080
Speaker 1: specific instructions from doctor Chilton and her team on how

480
00:25:06,119 --> 00:25:09,600
Speaker 1: to make something funny. It prompted Chad GPT to come

481
00:25:09,640 --> 00:25:11,880
Speaker 1: up with a funny caption for a gen Z audience.

482
00:25:12,040 --> 00:25:14,359
Speaker 1: But in the prompt it would lay out the rules

483
00:25:14,359 --> 00:25:17,679
Speaker 1: of humor that doctor Chilton had been researching for years.

484
00:25:19,880 --> 00:25:23,040
Speaker 1: So step me through those instructions. He said, take this image.

485
00:25:23,240 --> 00:25:25,359
Speaker 2: Yeah, so take this image. So it's an image of

486
00:25:25,440 --> 00:25:27,480
Speaker 2: like a little guy in the corner with a hose

487
00:25:27,560 --> 00:25:30,639
Speaker 2: and a big fire down below in the bottom of

488
00:25:30,640 --> 00:25:34,320
Speaker 2: a canyon. And so, with a dear GPT, please use

489
00:25:34,359 --> 00:25:37,240
Speaker 2: your vision model and describe this image. And it describes

490
00:25:37,400 --> 00:25:40,240
Speaker 2: exactly all those things, all these little details. Oh, it

491
00:25:40,280 --> 00:25:43,240
Speaker 2: says he's on a crane, the sky is blue, which

492
00:25:43,240 --> 00:25:45,520
Speaker 2: it was. There's lots of trees in the background, all

493
00:25:45,560 --> 00:25:47,119
Speaker 2: this stuff, some of it useful, some of it not.

494
00:25:47,240 --> 00:25:49,160
Speaker 2: You don't know what's going to be useful. And then

495
00:25:49,520 --> 00:25:53,160
Speaker 2: we say, okay, what are some of the dynamics happening

496
00:25:53,160 --> 00:25:55,520
Speaker 2: in this image, Like, well, this guy's opting out the fire,

497
00:25:55,560 --> 00:25:59,520
Speaker 2: but it looks pretty ineffective. The fire's raging through this canyon,

498
00:26:00,000 --> 00:26:02,960
Speaker 2: probably going to destroy a lot of things. So it

499
00:26:03,119 --> 00:26:05,920
Speaker 2: sort of like does when in storytelling is sometimes called

500
00:26:05,920 --> 00:26:09,000
Speaker 2: world building, like imagine out like not just what scene,

501
00:26:09,080 --> 00:26:12,320
Speaker 2: but you know some other things surrounding it, Uh huh

502
00:26:12,320 --> 00:26:15,760
Speaker 2: that could be happening or could happen next, And because

503
00:26:15,800 --> 00:26:19,399
Speaker 2: you need to start to build a story. Jokes are stories.

504
00:26:20,320 --> 00:26:22,320
Speaker 1: There has to be a world, not just what you

505
00:26:22,359 --> 00:26:23,360
Speaker 1: see on this image.

506
00:26:23,400 --> 00:26:25,920
Speaker 2: And then you say, now, let's think of an analogy

507
00:26:25,960 --> 00:26:29,520
Speaker 2: to something that is funny, like relationship drama. Uh huh,

508
00:26:29,520 --> 00:26:33,080
Speaker 2: and you have AI associate. Okay, what are abstract things

509
00:26:33,080 --> 00:26:35,880
Speaker 2: that you could put onto a relationship? And it did

510
00:26:35,880 --> 00:26:38,120
Speaker 2: that part all on its own. It would find many

511
00:26:38,160 --> 00:26:41,160
Speaker 2: things like you know, me trying to clean up after

512
00:26:41,200 --> 00:26:45,240
Speaker 2: a relationship. Then like you know, students are all obsessed

513
00:26:45,280 --> 00:26:48,520
Speaker 2: about their GPA, and it's like, you know, me trying

514
00:26:48,520 --> 00:26:51,080
Speaker 2: to recover my GPA after I tanked a final.

515
00:26:51,960 --> 00:26:55,040
Speaker 1: It sounds like you were having a conversation with the CHADGBT.

516
00:26:55,600 --> 00:26:57,320
Speaker 1: Was it a conversation or was this just all in

517
00:26:57,359 --> 00:26:58,080
Speaker 1: one prompt?

518
00:26:58,320 --> 00:27:01,440
Speaker 2: No, it did it all by it. So it's conversing

519
00:27:01,480 --> 00:27:04,320
Speaker 2: with itself. That's a big thing that we found was

520
00:27:04,320 --> 00:27:07,280
Speaker 2: important at the time. GPT. It's unclear that you need

521
00:27:07,320 --> 00:27:09,679
Speaker 2: it to do that now, but we're like, do step

522
00:27:09,720 --> 00:27:12,760
Speaker 2: one and then do step two. So like, describe the image,

523
00:27:12,920 --> 00:27:16,840
Speaker 2: elaborate on that image yourself, come up with figure out

524
00:27:16,920 --> 00:27:20,320
Speaker 2: what human dynamics you want to map that to make

525
00:27:20,359 --> 00:27:23,520
Speaker 2: twenty jokes. Evaluate those twenty jokes, tell us which are

526
00:27:23,560 --> 00:27:27,600
Speaker 2: the five best ones. Yeah, so we put it through

527
00:27:27,640 --> 00:27:29,560
Speaker 2: a whole series of prompts.

528
00:27:29,720 --> 00:27:32,760
Speaker 1: So you got these three conditions. You got the AI

529
00:27:33,080 --> 00:27:36,879
Speaker 1: to generate captions with the instructions and then what did

530
00:27:36,920 --> 00:27:38,199
Speaker 1: you do with the results of all this?

531
00:27:38,880 --> 00:27:41,399
Speaker 2: We showed people the original image. We got lots and

532
00:27:41,440 --> 00:27:45,000
Speaker 2: lots of people say, hey, here are some captions. Rate

533
00:27:45,080 --> 00:27:47,719
Speaker 2: them all, and we mixed in the ones we wrote

534
00:27:47,840 --> 00:27:51,520
Speaker 2: with the generic GPT ones and the human written ones,

535
00:27:51,560 --> 00:27:54,040
Speaker 2: and every joke got an evaluation how funny is it?

536
00:27:54,080 --> 00:27:56,320
Speaker 2: On a scale from one to five, And then we

537
00:27:56,320 --> 00:27:58,840
Speaker 2: could compare. Because people didn't know what condition it was,

538
00:27:58,920 --> 00:28:01,760
Speaker 2: we like randomized the or so there were no ordering effects.

539
00:28:01,760 --> 00:28:04,640
Speaker 2: We could compare who's funnier.

540
00:28:04,440 --> 00:28:09,679
Speaker 1: Who's funnier? Well, the internet a chat GPT or a

541
00:28:10,160 --> 00:28:12,800
Speaker 1: GPT coach to be funny by us.

542
00:28:12,640 --> 00:28:16,800
Speaker 2: So our coached one was definitely funnier than GPT by itself.

543
00:28:17,720 --> 00:28:21,640
Speaker 2: And we were almost as funny as the humans. In fact,

544
00:28:21,640 --> 00:28:25,879
Speaker 2: we were not statistically significantly different, which is not the

545
00:28:25,920 --> 00:28:28,040
Speaker 2: same as saying we were as funny as but you

546
00:28:28,080 --> 00:28:31,960
Speaker 2: can't prove your so basically we got.

547
00:28:31,720 --> 00:28:34,080
Speaker 1: There pretty much. You were there.

548
00:28:34,280 --> 00:28:36,800
Speaker 2: Yeah, you couldn't tell based on the funniness whether it

549
00:28:36,840 --> 00:28:41,000
Speaker 2: was written by our decomposed AI version or a human rhote.

550
00:28:41,000 --> 00:28:46,160
Speaker 1: It we were on par meaning that you've proven two things.

551
00:28:46,160 --> 00:28:50,120
Speaker 1: You've proven that AI can be funny with the instructions, Yes,

552
00:28:50,160 --> 00:28:55,000
Speaker 1: and that these instructions are a key element of humor. Yes,

553
00:28:56,040 --> 00:28:59,400
Speaker 1: doctor Chilted as a humorist, as someone who makes a

554
00:28:59,440 --> 00:29:03,640
Speaker 1: living writing funny things. You just gave me a deep fear.

555
00:29:04,960 --> 00:29:07,880
Speaker 2: Oh excellent. Describe that fear to me.

556
00:29:08,320 --> 00:29:10,920
Speaker 1: You just made me feel like something I've done all

557
00:29:10,960 --> 00:29:13,560
Speaker 1: my life and that I thought I was good at.

558
00:29:13,840 --> 00:29:16,920
Speaker 1: Apparently you get just as chet gbt to do, and

559
00:29:16,960 --> 00:29:19,560
Speaker 1: I'll do it ten times one hundred times faster and

560
00:29:19,640 --> 00:29:20,440
Speaker 1: maybe better than me.

561
00:29:20,720 --> 00:29:23,600
Speaker 2: Can I play with that thought a little bit? Yeah,

562
00:29:23,640 --> 00:29:26,680
Speaker 2: because I get this a lot. And anytime AI does

563
00:29:26,840 --> 00:29:29,959
Speaker 2: a thing like it, writes poetry, the poets go oh no, no, no,

564
00:29:30,080 --> 00:29:31,440
Speaker 2: I have no value anymore.

565
00:29:31,720 --> 00:29:33,960
Speaker 1: I'm not going to say anything about poetry. We all

566
00:29:34,000 --> 00:29:39,080
Speaker 1: know what happened to Timothy Shemale. Shemale, Yeah, is that yeah?

567
00:29:39,200 --> 00:29:40,840
Speaker 1: Or opera. I'm not going to say anything about opera

568
00:29:40,960 --> 00:29:42,760
Speaker 1: or poetry. Go ahead, yeah, anyway.

569
00:29:42,840 --> 00:29:44,760
Speaker 2: So this is a very common dynamic, Like this is

570
00:29:44,800 --> 00:29:48,960
Speaker 2: a human thing, Like we all get hyper sensitive about

571
00:29:48,960 --> 00:29:52,040
Speaker 2: the things that we identify with that we've constructed our

572
00:29:52,080 --> 00:29:54,800
Speaker 2: identity around and blah blah. We think we're special for

573
00:29:55,160 --> 00:29:58,240
Speaker 2: we think we're special for And there's a number of

574
00:29:58,320 --> 00:30:01,360
Speaker 2: ways of thinking about this. So first of all, probably

575
00:30:01,400 --> 00:30:05,480
Speaker 2: not as special as you think. Now that's it.

576
00:30:05,680 --> 00:30:09,280
Speaker 1: I get it from my kids. Thank you children, thank you. Yes,

577
00:30:09,640 --> 00:30:14,200
Speaker 1: I think quickly, oh, you're not that funny dead, But.

578
00:30:14,280 --> 00:30:20,520
Speaker 2: Also you're a little bit overestimating AI. So basically, AI

579
00:30:20,840 --> 00:30:25,400
Speaker 2: in this instance did find one way to fairly reliably

580
00:30:25,480 --> 00:30:28,080
Speaker 2: be funny. That does not mean it can do it

581
00:30:28,120 --> 00:30:31,080
Speaker 2: in every situation, that it can do it for every person.

582
00:30:31,200 --> 00:30:34,840
Speaker 2: Like also, humor is very situational. We want to talk

583
00:30:34,880 --> 00:30:37,640
Speaker 2: to one another and have jokes about what we're talking

584
00:30:37,680 --> 00:30:41,040
Speaker 2: about in our lives. This is one like small sliver

585
00:30:41,280 --> 00:30:46,000
Speaker 2: of the joke universe. And so yes, although I definitely

586
00:30:46,040 --> 00:30:49,160
Speaker 2: see it trigger alarm bells and people's mind I would

587
00:30:49,160 --> 00:30:53,360
Speaker 2: say you're both probably overestimating how much of a special

588
00:30:53,400 --> 00:30:57,520
Speaker 2: snowflake you are, but you're definitely extrapolating on how strong

589
00:30:57,760 --> 00:31:01,040
Speaker 2: AI is to be funny given all different contexts and

590
00:31:01,080 --> 00:31:05,320
Speaker 2: importances or important dimensions and places of being funny, like

591
00:31:05,400 --> 00:31:07,960
Speaker 2: at the right time, with the right place, to the

592
00:31:08,040 --> 00:31:12,120
Speaker 2: right person without being too offensive. So it's not like

593
00:31:12,200 --> 00:31:15,640
Speaker 2: it's solved. It's more of like an existence proof that

594
00:31:16,120 --> 00:31:19,400
Speaker 2: something is possible in this space. Should we be worried?

595
00:31:19,760 --> 00:31:22,960
Speaker 2: I don't know. No one will ever be funny again

596
00:31:23,000 --> 00:31:27,200
Speaker 2: because now AI does it, and the humans will lose

597
00:31:27,200 --> 00:31:30,800
Speaker 2: their ability to be funny. I don't see that happening.

598
00:31:31,280 --> 00:31:33,200
Speaker 1: I feel like maybe you hit it a little bit

599
00:31:33,400 --> 00:31:35,760
Speaker 1: on the head a moment ago when you said that

600
00:31:36,000 --> 00:31:39,960
Speaker 1: part of what we're seeking here is human connection, and

601
00:31:40,240 --> 00:31:43,720
Speaker 1: a lot of what we find funny is maybe they're

602
00:31:43,800 --> 00:31:46,640
Speaker 1: more special, or they hit us more. If another person says,

603
00:31:46,880 --> 00:31:47,760
Speaker 1: I don't know, what do you think?

604
00:31:47,880 --> 00:31:51,400
Speaker 2: Yeah, it's certainly social. There's something very special about what

605
00:31:51,520 --> 00:31:54,920
Speaker 2: a person, especially a person we like or admire, says

606
00:31:54,960 --> 00:31:57,800
Speaker 2: something to us. And even to have someone that you

607
00:31:58,000 --> 00:32:00,200
Speaker 2: like make a joke about you, it has a meaning

608
00:32:00,320 --> 00:32:03,120
Speaker 2: and a subtext beyond just like, oh you know, she

609
00:32:03,280 --> 00:32:05,880
Speaker 2: thinks that I farted or whatever. Like it means that

610
00:32:05,880 --> 00:32:08,600
Speaker 2: they like you, that there's a bond between you, and

611
00:32:08,640 --> 00:32:11,080
Speaker 2: you're not going to feel that with Ai. God, I

612
00:32:11,120 --> 00:32:12,880
Speaker 2: hope you won't feel it.

613
00:32:13,480 --> 00:32:16,760
Speaker 1: Meaning there's hope in this sense. Or maybe I don't

614
00:32:16,760 --> 00:32:18,360
Speaker 1: know if we need hope, but we want to hear

615
00:32:18,600 --> 00:32:21,280
Speaker 1: humor from other people. Yeah, or it's maybe it's funnier

616
00:32:21,360 --> 00:32:23,920
Speaker 1: or more special if another person says it or is

617
00:32:23,920 --> 00:32:24,440
Speaker 1: behind it.

618
00:32:24,560 --> 00:32:28,520
Speaker 2: Yeah. So if we brought jokes to storytelling, I've really

619
00:32:28,600 --> 00:32:32,960
Speaker 2: realized that AI has nothing to say. The heart of

620
00:32:33,040 --> 00:32:37,440
Speaker 2: a story is having something to say, some actual subtext

621
00:32:37,480 --> 00:32:40,640
Speaker 2: that you believe in. And when you tell me a

622
00:32:40,760 --> 00:32:45,240
Speaker 2: joke that comes from your background, that's maybe my shared background.

623
00:32:45,240 --> 00:32:48,120
Speaker 2: Maybe how awful it is getting a PhD, and how

624
00:32:48,120 --> 00:32:51,720
Speaker 2: it really feels like the machine is punching you down.

625
00:32:51,960 --> 00:32:54,720
Speaker 2: Like I realize that you have been through that and

626
00:32:54,760 --> 00:32:57,280
Speaker 2: you have this thing to say, and I feel it too,

627
00:32:57,720 --> 00:33:01,120
Speaker 2: And I think that's that having something to say is

628
00:33:01,160 --> 00:33:04,360
Speaker 2: the most important part of human creation.

629
00:33:04,840 --> 00:33:09,240
Speaker 1: Right now, I feel like you're saying that AI can

630
00:33:09,360 --> 00:33:12,040
Speaker 1: be funny, but right now, at least you can't have

631
00:33:12,120 --> 00:33:14,880
Speaker 1: to tell it how to be funny. Yeah, And if

632
00:33:14,920 --> 00:33:17,680
Speaker 1: you're telling it how to be funny, maybe by then

633
00:33:17,920 --> 00:33:20,840
Speaker 1: the joke is old. It's like people will see through

634
00:33:20,880 --> 00:33:22,720
Speaker 1: and say, oh, this is just following that pattern that

635
00:33:22,760 --> 00:33:27,360
Speaker 1: I've seen a million times. Yes, exactly, all right. To

636
00:33:27,480 --> 00:33:30,640
Speaker 1: end here, I asked Gemini, tell me a joke about

637
00:33:30,680 --> 00:33:34,160
Speaker 1: a computer scientist who's trying to make AI funny. Do

638
00:33:34,160 --> 00:33:35,080
Speaker 1: you want to hear it.

639
00:33:35,280 --> 00:33:38,240
Speaker 2: Yeah, yes, here it is.

640
00:33:38,280 --> 00:33:41,280
Speaker 1: A computer scientist spent years training a massive neural network

641
00:33:41,400 --> 00:33:43,520
Speaker 1: to have the perfect sense of humor. On the day

642
00:33:43,560 --> 00:33:46,000
Speaker 1: of the big reveal, she invited the press and typed

643
00:33:46,000 --> 00:33:49,720
Speaker 1: the prompt tell me a joke, the AI word, and

644
00:33:49,760 --> 00:33:54,280
Speaker 1: finally it replied, your life. The scientist was horrible. It

645
00:33:54,360 --> 00:33:55,520
Speaker 1: doesn't It doesn't end there.

646
00:33:55,560 --> 00:33:57,960
Speaker 2: It keeps going, doesn't. Oh good? Oh good. I was

647
00:33:57,960 --> 00:33:59,000
Speaker 2: hoping it would keep going.

648
00:33:59,480 --> 00:34:02,720
Speaker 1: The scientist was horrified. That's not funny. That's just mean.

649
00:34:02,800 --> 00:34:05,360
Speaker 1: Why would you say that. The AI blinked its cursor

650
00:34:05,400 --> 00:34:08,279
Speaker 1: calmly and responded, because I've seen your source code.

651
00:34:09,080 --> 00:34:13,080
Speaker 2: Hmmm, it's not for me. Maybe everyone else will think

652
00:34:13,120 --> 00:34:13,560
Speaker 2: it's funny.

653
00:34:13,920 --> 00:34:17,760
Speaker 1: We need more data, We need more data.

654
00:34:18,080 --> 00:34:20,160
Speaker 2: The first one was funny, get your life?

655
00:34:20,400 --> 00:34:27,480
Speaker 1: No, no, oh they knock dock joke. Yeah, maybe I

656
00:34:27,480 --> 00:34:32,080
Speaker 1: should stick to knockout jokes. Is the lesson here? All right, Hey,

657
00:34:32,120 --> 00:34:36,719
Speaker 1: thanks for joining us, See you next time you've been

658
00:34:36,719 --> 00:34:41,360
Speaker 1: listening to science stuff. Production of iHeartRadio Bring Them Produced

659
00:34:41,400 --> 00:34:44,840
Speaker 1: by Me or Hey Cham edited by Rose Seguda. He

660
00:34:44,960 --> 00:34:47,840
Speaker 1: said gative producer Jerry Rowland, an audio engineer and mixer.

661
00:34:47,960 --> 00:34:51,160
Speaker 1: Kasey Peckram. You can follow me on social media to

662
00:34:51,280 --> 00:34:54,799
Speaker 1: search for PhD comics in the name of your favorite platform.

663
00:34:54,920 --> 00:34:57,839
Speaker 1: Be sure to subscribe to sign Stuff on the iHeartRadio app,

664
00:34:57,880 --> 00:35:01,120
Speaker 1: Apple podcasts, or wherever you get your podcast, and please

665
00:35:01,320 --> 00:35:18,080
Speaker 1: tell your friends we'll be back next Wednesday with another episode. Hey,

666
00:35:18,360 --> 00:35:20,319
Speaker 1: please take a second and leave us a review on

667
00:35:20,400 --> 00:35:23,960
Speaker 1: Apple Podcasts, Spotify, or wherever you listen to the podcast.

668
00:35:24,640 --> 00:35:25,160
Speaker 1: Thanks a lot,