1
00:00:04,240 --> 00:00:10,640
Speaker 1: Get in test with technology with text stuff from dot com.

2
00:00:12,039 --> 00:00:15,600
Speaker 1: Say everyone, and welcome to text Uff. I'm Jonathan Strickland,

3
00:00:16,960 --> 00:00:20,439
Speaker 1: and uh, you know, something hit the news recently. You

4
00:00:20,480 --> 00:00:22,160
Speaker 1: may have seen it, in fact, you may have seen

5
00:00:22,200 --> 00:00:26,880
Speaker 1: it in multiple places about a program, a computer program

6
00:00:26,960 --> 00:00:30,920
Speaker 1: passing the Turing test, possibly the first such program to

7
00:00:31,000 --> 00:00:34,080
Speaker 1: ever pass the Turing test ever. And we had planned

8
00:00:34,120 --> 00:00:37,000
Speaker 1: on doing an episode on this, and then on top

9
00:00:37,040 --> 00:00:40,720
Speaker 1: of that, one of our listeners, Nick on Twitter said,

10
00:00:40,880 --> 00:00:43,280
Speaker 1: just saw this headline on Google News. A computer just

11
00:00:43,400 --> 00:00:46,840
Speaker 1: passed the Turing test in landmark trial. So we knew

12
00:00:46,840 --> 00:00:49,879
Speaker 1: that the timing was perfect. Thank you Nick for writing in.

13
00:00:50,640 --> 00:00:52,800
Speaker 1: We are really ready to talk about this, but in

14
00:00:52,880 --> 00:00:55,880
Speaker 1: order to do that, it's important that we, you know,

15
00:00:56,000 --> 00:00:59,120
Speaker 1: kind of lay some groundwork. So first of all, the

16
00:00:59,120 --> 00:01:03,040
Speaker 1: Touring test is named after Alan Turing, and we've done

17
00:01:03,040 --> 00:01:06,160
Speaker 1: a full episode on Alan Turing way back when, back

18
00:01:06,160 --> 00:01:12,479
Speaker 1: in November of Yeah. Phenomenal person, amazing thinker. Yeah, one

19
00:01:12,520 --> 00:01:17,440
Speaker 1: of the like the grandfather of computer science. Also tragic

20
00:01:17,520 --> 00:01:21,160
Speaker 1: life story which we went into detail back in that story.

21
00:01:21,240 --> 00:01:22,800
Speaker 1: And so you might wonder why, right, So what does

22
00:01:22,840 --> 00:01:25,679
Speaker 1: that have to do? Was the guy who was the

23
00:01:25,760 --> 00:01:28,720
Speaker 1: essentially the father of computer science or grandfather of computer science.

24
00:01:29,080 --> 00:01:30,720
Speaker 1: What does he have to do with the story about

25
00:01:31,160 --> 00:01:35,080
Speaker 1: a computer program in June two thousand fourteen, a computer

26
00:01:35,120 --> 00:01:40,360
Speaker 1: program with an interesting name, Eugene Goostman passing the Turing test.

27
00:01:41,080 --> 00:01:42,920
Speaker 1: What does that all have to do with each other? Well?

28
00:01:43,280 --> 00:01:45,800
Speaker 1: To answer that, we have to ask what is a

29
00:01:45,880 --> 00:01:49,760
Speaker 1: Turing test? Well, as it turns out Away, back in

30
00:01:49,880 --> 00:01:55,800
Speaker 1: the nineteen fifties, he started envisioning uh I thought experiment. Yeah,

31
00:01:55,840 --> 00:01:59,600
Speaker 1: he published a paper in a journal called Mind in

32
00:01:59,680 --> 00:02:03,520
Speaker 1: ninety fifty called Computing, Machinery and Intelligence, and he sort

33
00:02:03,560 --> 00:02:06,680
Speaker 1: of laid out his thought experiment there. And it's interesting

34
00:02:06,680 --> 00:02:08,840
Speaker 1: because what he did was he took this idea of

35
00:02:08,880 --> 00:02:12,320
Speaker 1: a party game and then adapted it for computers. Right.

36
00:02:12,480 --> 00:02:16,800
Speaker 1: The party game that it's based on would have three participants,

37
00:02:16,840 --> 00:02:20,480
Speaker 1: an interrogator, a man, and a woman um all situated

38
00:02:20,480 --> 00:02:23,160
Speaker 1: so that they can't see one another, and the interrogator

39
00:02:23,200 --> 00:02:26,080
Speaker 1: is supposed to ask questions to try to figure out

40
00:02:26,480 --> 00:02:29,400
Speaker 1: which of the participants is the dude and which is

41
00:02:29,400 --> 00:02:32,600
Speaker 1: the lady exactly, And it's the dude's job to try

42
00:02:32,600 --> 00:02:37,200
Speaker 1: and mislead the interrogator to believe that that he, in fact,

43
00:02:37,240 --> 00:02:38,760
Speaker 1: is the lady and the other one is the man.

44
00:02:39,160 --> 00:02:42,160
Speaker 1: It's the lady's job to say, hey, I want you

45
00:02:42,200 --> 00:02:44,440
Speaker 1: to get this right. I'm the lady. That other guy

46
00:02:44,480 --> 00:02:47,760
Speaker 1: that's the dude. And so the interrogator has to ask questions. Now,

47
00:02:48,480 --> 00:02:51,120
Speaker 1: obviously they can't see each other. Because they could see

48
00:02:51,120 --> 00:02:53,920
Speaker 1: each other, then that would probably give things away. Probably,

49
00:02:54,200 --> 00:02:56,800
Speaker 1: they really shouldn't be able to hear each other because

50
00:02:56,800 --> 00:02:59,880
Speaker 1: that could also give things away. Sure, and if you

51
00:03:00,080 --> 00:03:03,399
Speaker 1: use handwritten notes, then you the interrogator might be able

52
00:03:03,440 --> 00:03:06,440
Speaker 1: to make judgments based on the handwriting style. So it

53
00:03:06,440 --> 00:03:08,880
Speaker 1: should really be typewritten, right, So you want to you

54
00:03:08,919 --> 00:03:13,040
Speaker 1: want to remove as many easily identifiable traits from this

55
00:03:13,080 --> 00:03:15,880
Speaker 1: game as possible to make it all about the questions

56
00:03:15,919 --> 00:03:19,480
Speaker 1: and the answers. Now, Touring said, what if we were

57
00:03:19,480 --> 00:03:23,040
Speaker 1: to take the same basic premise, but instead of having

58
00:03:23,160 --> 00:03:29,120
Speaker 1: two human interviewees, replace one of those humans with a machine. Now,

59
00:03:29,480 --> 00:03:33,200
Speaker 1: if that machine can convince the interrogator that the machine

60
00:03:33,240 --> 00:03:36,640
Speaker 1: itself is a human being, it would be a pretty

61
00:03:36,640 --> 00:03:40,360
Speaker 1: phenomenal achievement. Or you know, just generally, if the interrogator

62
00:03:40,560 --> 00:03:45,560
Speaker 1: wasn't sure which of the two interrogate ease, right, which

63
00:03:45,600 --> 00:03:47,360
Speaker 1: one is the human, which one is the machine? Or

64
00:03:47,440 --> 00:03:50,600
Speaker 1: even if there could be a case where you don't know.

65
00:03:50,680 --> 00:03:54,120
Speaker 1: It may be that you have two humans that you're interrogating,

66
00:03:54,200 --> 00:03:56,040
Speaker 1: and it may be one of those things where you

67
00:03:56,120 --> 00:03:57,720
Speaker 1: have to you know, if you don't know for a

68
00:03:57,800 --> 00:03:59,520
Speaker 1: fact that one of them the machine, that makes it

69
00:03:59,560 --> 00:04:02,520
Speaker 1: even hard, right, at least assuming that the computer program

70
00:04:02,560 --> 00:04:06,800
Speaker 1: is sophisticated enough. Now, Touring was saying that we don't

71
00:04:06,800 --> 00:04:08,640
Speaker 1: have any machines right now that can do this, but

72
00:04:08,680 --> 00:04:12,600
Speaker 1: I envision a time when computers will be able to

73
00:04:12,680 --> 00:04:14,880
Speaker 1: do such a thing where if you were to interrogate

74
00:04:14,960 --> 00:04:19,040
Speaker 1: a computer, you would get back responses that would be uh,

75
00:04:19,320 --> 00:04:21,960
Speaker 1: convincing enough for it to make it difficult to determine

76
00:04:21,960 --> 00:04:24,719
Speaker 1: if it were man or machine. And so he said

77
00:04:25,120 --> 00:04:27,640
Speaker 1: that he predicted in fifty years time, which would be

78
00:04:27,680 --> 00:04:31,200
Speaker 1: the year two thousand, that computers and software be sophisticate

79
00:04:31,320 --> 00:04:34,120
Speaker 1: enough that interrogators would only be able to guess correctly

80
00:04:34,520 --> 00:04:37,400
Speaker 1: seventy of the time, meaning they would be fooled by

81
00:04:37,400 --> 00:04:41,240
Speaker 1: the computer of the time. Right, So, uh, this was

82
00:04:41,279 --> 00:04:43,120
Speaker 1: just kind of a thing he was coming up with,

83
00:04:43,160 --> 00:04:47,240
Speaker 1: like an idea, a prediction not necessarily a test, although

84
00:04:47,800 --> 00:04:51,800
Speaker 1: very much like Moore's observation became Moore's law, this became

85
00:04:51,839 --> 00:04:54,600
Speaker 1: what is known as the touring test. People talk about

86
00:04:54,640 --> 00:04:59,240
Speaker 1: a touring test as a machine capable of fooling people

87
00:04:59,320 --> 00:05:02,479
Speaker 1: into thinking it's another person at least thirty of the

88
00:05:02,520 --> 00:05:05,960
Speaker 1: time after five minutes of conversation. Very good point. Yes,

89
00:05:06,040 --> 00:05:07,960
Speaker 1: it needs to be five minutes of conversation. If you

90
00:05:08,000 --> 00:05:10,960
Speaker 1: are just getting maybe two or three responses, that might

91
00:05:11,000 --> 00:05:12,480
Speaker 1: not be enough for you to be able to draw

92
00:05:12,600 --> 00:05:16,440
Speaker 1: a conclusion that you feel good about. If after five

93
00:05:16,520 --> 00:05:21,440
Speaker 1: minutes you still are not entirely, entirely certain, then that

94
00:05:21,560 --> 00:05:24,960
Speaker 1: might say that this machine, in fact has passed the

95
00:05:25,000 --> 00:05:29,440
Speaker 1: touring test. And this has been extrapolated to mean something

96
00:05:29,480 --> 00:05:35,240
Speaker 1: about machine intelligence because Turing himself tied the idea of

97
00:05:35,240 --> 00:05:39,600
Speaker 1: of how we perceive a machines intelligence directly to artificial

98
00:05:39,640 --> 00:05:43,760
Speaker 1: intelligence and and and even how we perceive human intelligence.

99
00:05:43,880 --> 00:05:47,520
Speaker 1: Because here's Touring's idea is a little cheeky, and I

100
00:05:47,600 --> 00:05:52,000
Speaker 1: love the fact that's so cheeky. So Turing kind of said,

101
00:05:53,200 --> 00:05:55,159
Speaker 1: would you say such a machine as intelligent if it

102
00:05:55,200 --> 00:05:57,960
Speaker 1: appears to be intelligent, is it fair to say that

103
00:05:58,200 --> 00:06:02,440
Speaker 1: is intelligent touring. So why not because I'm only able

104
00:06:02,480 --> 00:06:05,960
Speaker 1: to tell that I am intelligent, right, because of my experience.

105
00:06:05,960 --> 00:06:08,520
Speaker 1: I'm only able to have my own personal experience. I

106
00:06:08,520 --> 00:06:12,000
Speaker 1: can't experience what someone else's life is like. All right,

107
00:06:12,000 --> 00:06:14,400
Speaker 1: Sitting across from each other, Jonathan and I can only

108
00:06:14,520 --> 00:06:17,360
Speaker 1: assume that the other one is intelligent, right, And it's

109
00:06:17,440 --> 00:06:20,960
Speaker 1: because the other person is displaying traits that we associate

110
00:06:21,040 --> 00:06:24,599
Speaker 1: with intelligence. They they seem to be able to take

111
00:06:24,640 --> 00:06:28,279
Speaker 1: an information, respond to it, make decisions. And based on

112
00:06:28,320 --> 00:06:30,919
Speaker 1: the fact that we ourselves also do that thing, we

113
00:06:31,040 --> 00:06:33,159
Speaker 1: go ahead and say, all right, well, they clearly have

114
00:06:33,800 --> 00:06:38,960
Speaker 1: the same features that I have, which includes intelligence. Uh. Now,

115
00:06:39,279 --> 00:06:42,440
Speaker 1: he says, why would we not extend that same courtesy

116
00:06:42,520 --> 00:06:46,320
Speaker 1: to a machine if it also appeared to display those

117
00:06:46,360 --> 00:06:49,839
Speaker 1: same uh features? He says, doesn't matter if the computer

118
00:06:49,920 --> 00:06:53,440
Speaker 1: is quote unquote thinking. If it's it can fool you. Yeah,

119
00:06:53,480 --> 00:06:55,720
Speaker 1: if it's able to to simulate it well enough, you

120
00:06:55,800 --> 00:06:58,119
Speaker 1: might as well say it's intelligent, because stimulate is probably

121
00:06:58,120 --> 00:07:01,120
Speaker 1: a kinder phrase for that than fool. Yeah. Yeah, well

122
00:07:01,279 --> 00:07:03,680
Speaker 1: I know it's it is fooling essentially, I mean, because

123
00:07:04,040 --> 00:07:08,200
Speaker 1: ultimately you're talking about a computer programmer, who's making this happen?

124
00:07:08,640 --> 00:07:11,440
Speaker 1: So nowadays we think of this as the Turing test.

125
00:07:11,560 --> 00:07:15,880
Speaker 1: Can a machine of the time are more fool someone

126
00:07:15,920 --> 00:07:19,600
Speaker 1: after five minutes of conversation into thinking it's a human? Now,

127
00:07:21,440 --> 00:07:24,080
Speaker 1: this is really hard to do. This is yeah, this

128
00:07:24,120 --> 00:07:28,000
Speaker 1: is non trivial. I mean it sounds almost simple. Yeah,

129
00:07:28,120 --> 00:07:33,119
Speaker 1: the concept is simple, the execution incredibly difficult. Because here's

130
00:07:33,120 --> 00:07:39,920
Speaker 1: the thing, human language varied. We have unstructured, unpredictable ways

131
00:07:39,960 --> 00:07:42,680
Speaker 1: to say things like if if I were to tell

132
00:07:42,720 --> 00:07:47,640
Speaker 1: you that it's a hundred degrees outside and it is

133
00:07:49,080 --> 00:07:51,800
Speaker 1: humid like the humidities at and you go out there,

134
00:07:52,160 --> 00:07:55,360
Speaker 1: then I'm sure all of our listeners would have slightly

135
00:07:55,400 --> 00:08:00,480
Speaker 1: different ways to express their thoughts on the conditions out side.

136
00:08:00,880 --> 00:08:05,480
Speaker 1: Some of them would probably contain colorful metaphors, probably mine

137
00:08:05,920 --> 00:08:09,640
Speaker 1: very likely when it contain colorful metaphors, particularly if I

138
00:08:09,640 --> 00:08:11,560
Speaker 1: had to be outside for any length of time. But

139
00:08:11,640 --> 00:08:14,080
Speaker 1: that's the point. We would all have different ways of

140
00:08:14,120 --> 00:08:18,440
Speaker 1: saying this. So how do you make a computer program

141
00:08:18,480 --> 00:08:22,760
Speaker 1: able to interpret all the myriad of ways we can

142
00:08:22,840 --> 00:08:27,440
Speaker 1: all express the same thought, let alone any thought. Right,

143
00:08:27,480 --> 00:08:30,600
Speaker 1: this is what's referred to as natural language. Recognition, and

144
00:08:30,640 --> 00:08:34,120
Speaker 1: it's a really huge problem in artificial intelligence and and

145
00:08:34,120 --> 00:08:38,520
Speaker 1: and a lot of other speech related computer programming. Exactly. Yeah,

146
00:08:38,600 --> 00:08:41,319
Speaker 1: this is where a computer program has to be able

147
00:08:41,360 --> 00:08:46,040
Speaker 1: to parse the language so it recognizes things like this

148
00:08:46,280 --> 00:08:50,400
Speaker 1: word is a noun, this word is a verb, this

149
00:08:50,400 --> 00:08:54,600
Speaker 1: this word alters this other word. Uh. And not only

150
00:08:54,600 --> 00:08:56,240
Speaker 1: does it need to be able to recognize it, it it

151
00:08:56,360 --> 00:08:59,440
Speaker 1: needs to be able to respond in kind. Right, So

152
00:08:59,559 --> 00:09:02,240
Speaker 1: if you appropriately in some way or another share, Yeah,

153
00:09:02,280 --> 00:09:05,679
Speaker 1: you could have a computer program that's literally making up

154
00:09:05,720 --> 00:09:09,960
Speaker 1: sentences or you know, the approximation of a sentence randomly,

155
00:09:10,120 --> 00:09:13,600
Speaker 1: where it's just pulling strings of words and placing them

156
00:09:13,679 --> 00:09:16,200
Speaker 1: in a sequence and then presenting them. But that wouldn't

157
00:09:16,200 --> 00:09:18,720
Speaker 1: be convincing at all. If I were to say, Hello,

158
00:09:18,760 --> 00:09:21,480
Speaker 1: how are you today, and Lauren was to say blue

159
00:09:21,480 --> 00:09:26,560
Speaker 1: panther pickup jump down street, I'd be like what? And

160
00:09:26,600 --> 00:09:29,120
Speaker 1: even that was closer to being a sentence than some

161
00:09:29,200 --> 00:09:30,760
Speaker 1: of the random stuff that you would see if it

162
00:09:30,840 --> 00:09:35,800
Speaker 1: was just truly completely So it also has to be

163
00:09:35,920 --> 00:09:39,880
Speaker 1: able to uh endure a five minute long conversation, like

164
00:09:39,920 --> 00:09:41,760
Speaker 1: we said, in order to pass the Turing test. So

165
00:09:42,040 --> 00:09:44,520
Speaker 1: you can't have too much repetition or that gives it

166
00:09:44,520 --> 00:09:46,719
Speaker 1: away absolutely. If you've ever been playing a video game

167
00:09:46,760 --> 00:09:48,720
Speaker 1: and all of the NPCs say the same thing over

168
00:09:48,760 --> 00:09:51,200
Speaker 1: and over and over again. Yeah, if if all the

169
00:09:51,320 --> 00:09:55,320
Speaker 1: chat bot says is hey listen, then you're clearly playing

170
00:09:55,400 --> 00:10:00,200
Speaker 1: Zelda and you're not actually having a decent conversation. That

171
00:10:00,280 --> 00:10:02,760
Speaker 1: being said, there is someone I know who plays a

172
00:10:02,840 --> 00:10:05,440
Speaker 1: ferry at the Georgia Renaissance Festival, and hey listen is

173
00:10:05,840 --> 00:10:10,040
Speaker 1: heavily represented in a repertoire. Pretty amazing, No, it's pretty awesome,

174
00:10:10,320 --> 00:10:12,480
Speaker 1: but at any rate, Yeah, So, so these are these

175
00:10:12,520 --> 00:10:16,000
Speaker 1: are big problems. You have to build a database of words,

176
00:10:16,280 --> 00:10:19,000
Speaker 1: you have to be able to figure out what kind

177
00:10:19,040 --> 00:10:22,280
Speaker 1: of syntax are you going for. It's a wide open,

178
00:10:22,360 --> 00:10:27,040
Speaker 1: huge problem, and solving this problem can be really beneficial

179
00:10:27,040 --> 00:10:28,679
Speaker 1: in lots of ways. We'll talk a little bit about

180
00:10:28,679 --> 00:10:31,960
Speaker 1: that later. It's it's beyond just making a program that

181
00:10:32,040 --> 00:10:35,960
Speaker 1: seems human, right, That's that's one way of looking at it.

182
00:10:36,000 --> 00:10:37,440
Speaker 1: But there are a lot of other benefits that come

183
00:10:37,440 --> 00:10:39,200
Speaker 1: along with it, which will chat about towards the end

184
00:10:39,240 --> 00:10:41,439
Speaker 1: of the show. But for right now, let's go into

185
00:10:41,600 --> 00:10:45,520
Speaker 1: this story behind Eugene Goostman and whether or not it

186
00:10:45,600 --> 00:10:49,360
Speaker 1: was actually the first chat bought to pass the Turing test. Yeah. So,

187
00:10:49,520 --> 00:10:52,600
Speaker 1: first of all, we've got three programmers in this story

188
00:10:52,600 --> 00:10:58,040
Speaker 1: of Vladimir Vassilov, Eugene Demchinko and Sarage you listen. As

189
00:10:58,080 --> 00:11:01,079
Speaker 1: you may guess from their names, they all hail from

190
00:11:02,000 --> 00:11:05,880
Speaker 1: Russia and the Ukraine as well, and although not all

191
00:11:05,920 --> 00:11:08,440
Speaker 1: the not all of them still live there. But starting

192
00:11:08,480 --> 00:11:11,760
Speaker 1: in two thousand one, they got to work on this, right, Yeah,

193
00:11:11,800 --> 00:11:14,320
Speaker 1: they were trying to design a computer program that would

194
00:11:14,360 --> 00:11:18,160
Speaker 1: pose specifically as a thirteen year old boy from Odessa

195
00:11:18,240 --> 00:11:21,720
Speaker 1: in the Ukraine. Yeah, and that meant that they had

196
00:11:21,960 --> 00:11:27,319
Speaker 1: specific parameters that they could work within. It automatically helped

197
00:11:27,520 --> 00:11:31,200
Speaker 1: reduce some of that unpredictability and that lack of restriction

198
00:11:31,280 --> 00:11:33,479
Speaker 1: that you would have if you were to just say,

199
00:11:33,520 --> 00:11:38,000
Speaker 1: this is a fluent adult speaker of a given language. Yeah, yeah,

200
00:11:38,160 --> 00:11:41,800
Speaker 1: giving him these you know, setting up the expectation from

201
00:11:41,840 --> 00:11:43,720
Speaker 1: the judges that you know, this is a non native

202
00:11:43,720 --> 00:11:47,560
Speaker 1: English speaker. It's it's a kid essentially, Um, you know

203
00:11:47,600 --> 00:11:50,840
Speaker 1: that they'll expect him to have limited knowledge of the

204
00:11:50,880 --> 00:11:55,880
Speaker 1: world and and different subject areas and a limited understanding

205
00:11:55,920 --> 00:11:59,240
Speaker 1: of English. Vocabulary and grammar and all that kind of stuff.

206
00:11:59,320 --> 00:12:02,600
Speaker 1: So you're already managing expectations. That's going to come into

207
00:12:02,600 --> 00:12:06,000
Speaker 1: play when we talk about some of the criticisms about this,

208
00:12:06,080 --> 00:12:08,080
Speaker 1: although although I do think it's a very clever way

209
00:12:08,120 --> 00:12:10,560
Speaker 1: and in a lot of previous chat bots have have

210
00:12:10,640 --> 00:12:13,679
Speaker 1: had similar, yes, similar kind of approaches. Yeah, because like

211
00:12:13,679 --> 00:12:16,160
Speaker 1: we said, if you're to take a quote unquote pure

212
00:12:16,280 --> 00:12:21,080
Speaker 1: approach to this, it's really really challenging. So yeah, by

213
00:12:21,200 --> 00:12:25,000
Speaker 1: by limiting this, the judges have an idea of well,

214
00:12:25,040 --> 00:12:26,920
Speaker 1: this this could be a thirteen year old boy or

215
00:12:26,920 --> 00:12:29,200
Speaker 1: it could be a computer program. It means that the

216
00:12:29,200 --> 00:12:32,280
Speaker 1: computer program doesn't have to be as sophisticated as one

217
00:12:32,320 --> 00:12:34,760
Speaker 1: that would be completely fluent and have you know, an

218
00:12:34,760 --> 00:12:39,880
Speaker 1: adult's experiences, uh and ability to communicate. So that was

219
00:12:39,880 --> 00:12:43,760
Speaker 1: step one. And uh. Eugene Goosman took part in the

220
00:12:43,800 --> 00:12:48,760
Speaker 1: competition UH that had five total chatbots. It was it

221
00:12:48,800 --> 00:12:52,120
Speaker 1: was one of five that took place on the sixtieth

222
00:12:52,200 --> 00:12:56,880
Speaker 1: anniversary of Touring's death, and the program managed to full

223
00:12:57,000 --> 00:12:59,679
Speaker 1: thirty three percent of the judges into thinking it was

224
00:12:59,679 --> 00:13:03,079
Speaker 1: actually a person. UH. And it had there were thirty judges.

225
00:13:03,120 --> 00:13:07,120
Speaker 1: From what I understand. So that was where you got

226
00:13:07,160 --> 00:13:10,320
Speaker 1: all the headlines of chat bought. There are computer beats

227
00:13:10,360 --> 00:13:14,840
Speaker 1: touring tests, which already is not accurate. There were some

228
00:13:15,040 --> 00:13:18,800
Speaker 1: slight um hiccups in a little bit of the news

229
00:13:18,840 --> 00:13:22,319
Speaker 1: reporting that We'll get into that part of the story later,

230
00:13:22,880 --> 00:13:25,360
Speaker 1: but you know, the key takeaways I think here are

231
00:13:25,440 --> 00:13:28,000
Speaker 1: that this was a this was a competition that was

232
00:13:28,040 --> 00:13:33,959
Speaker 1: celebrating touring awesome um and that there were thirty judges,

233
00:13:34,400 --> 00:13:38,160
Speaker 1: some of whom were celebrities, yes, including an actor who

234
00:13:38,280 --> 00:13:41,360
Speaker 1: had appeared on Red Dwarf right. Yes. And then a

235
00:13:41,360 --> 00:13:44,360
Speaker 1: couple of years before that, because because you said they

236
00:13:44,360 --> 00:13:46,559
Speaker 1: started working on this on two th two thousand one,

237
00:13:46,640 --> 00:13:49,719
Speaker 1: this was not the first time that Gustman had entered competition.

238
00:13:50,400 --> 00:13:55,240
Speaker 1: Two years previously, that same software had convinced twenty of

239
00:13:55,320 --> 00:13:58,840
Speaker 1: judges at a similar competition that was held at Bletchley Park. Now,

240
00:13:58,920 --> 00:14:02,640
Speaker 1: but Bletchley Park's where touring helped crack the Enigma machine,

241
00:14:03,440 --> 00:14:07,080
Speaker 1: the the encoding device that the German military was using

242
00:14:07,160 --> 00:14:10,600
Speaker 1: during World War Two. So this was a big celebration.

243
00:14:10,600 --> 00:14:14,120
Speaker 1: It was at the centennial celebrating his birth, and it

244
00:14:14,440 --> 00:14:20,440
Speaker 1: ended up falling just short of passing the Turing Test. Uh.

245
00:14:20,520 --> 00:14:24,520
Speaker 1: The organizer of the the more recent event, the one

246
00:14:24,560 --> 00:14:28,000
Speaker 1: in which Goostman ran away with quote unquote beating the

247
00:14:28,000 --> 00:14:31,960
Speaker 1: Turing test. Uh. That organizer was Kevin Warwick. That name

248
00:14:32,000 --> 00:14:34,080
Speaker 1: may sound familiar to some of our listeners if you've

249
00:14:34,120 --> 00:14:37,480
Speaker 1: ever heard us talk about cyborgs. He's the guy who

250
00:14:37,680 --> 00:14:41,560
Speaker 1: had an r f I D chip surgically implanted into him.

251
00:14:42,000 --> 00:14:45,000
Speaker 1: His wife also did at one point. Yes, they could

252
00:14:45,000 --> 00:14:49,200
Speaker 1: communicate with each other through them. Some unflattering news media

253
00:14:49,360 --> 00:14:53,040
Speaker 1: sometimes refers to him as Captain Cyborg. Yeah. There there

254
00:14:53,080 --> 00:14:56,280
Speaker 1: are some critics who say that he uh, he courts

255
00:14:56,440 --> 00:15:02,000
Speaker 1: publicity in a manner that is um unbeco umming of

256
00:15:02,000 --> 00:15:05,360
Speaker 1: of of a scientist, yes, really of anybody. Yes, yes,

257
00:15:05,960 --> 00:15:08,800
Speaker 1: uh those are the critics who say that. By the way,

258
00:15:08,960 --> 00:15:11,560
Speaker 1: I just want to make that clear. At any rate,

259
00:15:11,720 --> 00:15:13,680
Speaker 1: He said that this was the first time a chat

260
00:15:13,680 --> 00:15:16,320
Speaker 1: bot had passed the Turing test at an event where

261
00:15:16,360 --> 00:15:19,280
Speaker 1: the conversation was open ended, meaning that they had not

262
00:15:19,560 --> 00:15:25,240
Speaker 1: previously uh decided upon a specific topic or line of questioning.

263
00:15:25,640 --> 00:15:28,400
Speaker 1: That the judges were allowed to say whatever they wanted

264
00:15:28,440 --> 00:15:30,960
Speaker 1: to the chat bot and response, which obviously makes it

265
00:15:31,000 --> 00:15:33,640
Speaker 1: harder because you have to have a much wider breadth

266
00:15:33,840 --> 00:15:37,880
Speaker 1: of potential responses. Yeah. Yeah, because again, if you were

267
00:15:37,920 --> 00:15:40,160
Speaker 1: to say, all right, this chat bot is just going

268
00:15:40,200 --> 00:15:43,920
Speaker 1: to talk about, um, I don't know, sporting events from

269
00:15:44,000 --> 00:15:48,560
Speaker 1: last year, Well, then you can prepare pretty well for that. Yeah, exactly.

270
00:15:48,600 --> 00:15:50,160
Speaker 1: So it's one of those again, it's one of those

271
00:15:50,160 --> 00:15:54,720
Speaker 1: things where the unrestricted nature adds in a degree of difficulty.

272
00:15:54,800 --> 00:15:57,440
Speaker 1: So why would you need to make the qualification that

273
00:15:57,520 --> 00:15:59,320
Speaker 1: this is an open ended approach and this is the

274
00:15:59,320 --> 00:16:03,840
Speaker 1: first chat to manage it. That would be because despite

275
00:16:03,880 --> 00:16:06,960
Speaker 1: what you may have heard, Eugene Goosband was not the

276
00:16:07,000 --> 00:16:09,960
Speaker 1: first program to beat the Turing test, not by a

277
00:16:10,000 --> 00:16:14,720
Speaker 1: long shot. So work on these sort of chatbots, these

278
00:16:14,800 --> 00:16:19,520
Speaker 1: kind of artificial conversationalists. That's real recent, right, I mean

279
00:16:19,600 --> 00:16:21,880
Speaker 1: they just started doing that like what like like maybe

280
00:16:21,880 --> 00:16:23,840
Speaker 1: three or four years ago or two thousand, one of

281
00:16:23,880 --> 00:16:26,560
Speaker 1: the earliest, I mean that's when they started with Goosman. Yeah. No,

282
00:16:26,640 --> 00:16:31,320
Speaker 1: in the nineteen sixties and seventies, say, back in the

283
00:16:31,440 --> 00:16:35,160
Speaker 1: mid nineteen sixties there was Eliza, which was written by

284
00:16:35,360 --> 00:16:38,800
Speaker 1: Joseph Weisenbaum. If you're pronouncing it in the correct German,

285
00:16:39,000 --> 00:16:42,160
Speaker 1: that's correct. Excellent, I'm finally learning. I'm sure, I'm sure

286
00:16:42,240 --> 00:16:45,920
Speaker 1: he pronounces it Wisenbaum, but yeah, but no, it would

287
00:16:45,920 --> 00:16:49,840
Speaker 1: be Weisenbaum at any rate. Um. This this was a

288
00:16:49,880 --> 00:16:53,400
Speaker 1: program UM that would respond to human conversation and what

289
00:16:53,880 --> 00:16:57,720
Speaker 1: ideally would be a relevant way. Yes, it was obviously

290
00:16:57,760 --> 00:17:01,040
Speaker 1: an early attempt. It was not meant to be a

291
00:17:01,920 --> 00:17:04,320
Speaker 1: program that takes on the Turing test. It was really

292
00:17:04,359 --> 00:17:06,439
Speaker 1: a thought again, kind of like a thought experiment, the

293
00:17:06,440 --> 00:17:09,320
Speaker 1: idea of what does it take to create a piece

294
00:17:09,320 --> 00:17:12,680
Speaker 1: of software that can react to questions and make it

295
00:17:12,840 --> 00:17:15,080
Speaker 1: make sense. At that point, it was more a can

296
00:17:15,160 --> 00:17:17,480
Speaker 1: we do this? Then let's do this for real? Yeah,

297
00:17:17,520 --> 00:17:20,879
Speaker 1: and these are the sort of foundations that you have

298
00:17:20,920 --> 00:17:23,280
Speaker 1: to lay in order for other things like the Eugene

299
00:17:23,320 --> 00:17:28,840
Speaker 1: Goostman program to be uh successful. So he created a

300
00:17:28,920 --> 00:17:33,000
Speaker 1: language analyzer. Now this specifically would look at words that

301
00:17:33,280 --> 00:17:36,000
Speaker 1: users would put in and then compare them against a

302
00:17:36,080 --> 00:17:38,720
Speaker 1: database of words that were stored in the computer's memory.

303
00:17:39,240 --> 00:17:42,840
Speaker 1: And then also created scripts. Now, in this case, the

304
00:17:42,880 --> 00:17:45,119
Speaker 1: scripts were sets of rules. They're kind of like, you know,

305
00:17:45,160 --> 00:17:48,200
Speaker 1: like a protocol or an algorithm in a way. These

306
00:17:48,280 --> 00:17:52,640
Speaker 1: rules dictated how Eliza would respond to messages in order

307
00:17:52,680 --> 00:17:56,160
Speaker 1: to cut down on that huge, massive number of variables

308
00:17:56,160 --> 00:17:59,719
Speaker 1: we were talking about, the whole unrestricted, unpredictable thing, and

309
00:17:59,800 --> 00:18:03,679
Speaker 1: so they would have different uh kind of like like overlays.

310
00:18:03,720 --> 00:18:05,520
Speaker 1: Think of him as an overlay that would kind of

311
00:18:05,840 --> 00:18:09,359
Speaker 1: guide Aliza's responses. And the most well known one was

312
00:18:09,400 --> 00:18:13,040
Speaker 1: called doctor, which put Eliza in the role of a

313
00:18:13,200 --> 00:18:18,480
Speaker 1: Rogerian psychiatrist. Uh. This is the person who responds to

314
00:18:18,480 --> 00:18:20,600
Speaker 1: everything with a question. Al Right. It's that it's that

315
00:18:20,800 --> 00:18:25,080
Speaker 1: passive interview style where where you know, you repeat back.

316
00:18:25,640 --> 00:18:27,560
Speaker 1: You know, if if if I go, oh, man, like

317
00:18:27,600 --> 00:18:30,440
Speaker 1: I'm I'm I'm really sad about my cat, Oh, tell

318
00:18:30,920 --> 00:18:32,640
Speaker 1: what is it about your cat that makes you sad?

319
00:18:33,320 --> 00:18:35,159
Speaker 1: And then you can say things like you know, And

320
00:18:35,359 --> 00:18:38,919
Speaker 1: that's one of those things where as a conversation starts

321
00:18:38,960 --> 00:18:42,560
Speaker 1: to wind down, you then have another line of questions,

322
00:18:42,600 --> 00:18:45,280
Speaker 1: So tell me more about your mother, like that the

323
00:18:45,560 --> 00:18:47,960
Speaker 1: whole tell me about your mother thing. That's really coming

324
00:18:47,960 --> 00:18:50,160
Speaker 1: back to this kind of model of psychiatrists. But yeah,

325
00:18:50,200 --> 00:18:52,600
Speaker 1: if you've ever heard the joke about them responding to

326
00:18:52,640 --> 00:18:56,040
Speaker 1: anything with another question, just just taking the last word

327
00:18:56,080 --> 00:18:58,600
Speaker 1: and turning it into a question, what am I paying

328
00:18:58,600 --> 00:19:01,800
Speaker 1: you for Why do you think your page for that

329
00:19:01,880 --> 00:19:05,840
Speaker 1: kind of thing? Uh So that's what That's how Eliza responded,

330
00:19:05,880 --> 00:19:10,040
Speaker 1: And in fact, you can find examples of the Eliza

331
00:19:10,400 --> 00:19:14,159
Speaker 1: program script transcripts and even even the actual there's there.

332
00:19:14,160 --> 00:19:17,399
Speaker 1: There are ports, you could say, or people have essentially

333
00:19:17,440 --> 00:19:20,280
Speaker 1: created their own version of Eliza just using the original

334
00:19:20,320 --> 00:19:22,800
Speaker 1: Liza as a guide. You can find tons of them

335
00:19:22,800 --> 00:19:25,000
Speaker 1: on the web, and you can attempt to have a

336
00:19:25,040 --> 00:19:29,200
Speaker 1: conversation with them. It's not terribly compelling, but it's it's

337
00:19:29,280 --> 00:19:33,240
Speaker 1: kind of fun. Usually within maybe four or five exchanges,

338
00:19:33,240 --> 00:19:35,120
Speaker 1: you've already run into something where you're like, well, this

339
00:19:35,160 --> 00:19:38,120
Speaker 1: can't be a person, or if it's a person, it's

340
00:19:38,119 --> 00:19:41,040
Speaker 1: the weirdest person I've ever conversed with. But but again,

341
00:19:41,080 --> 00:19:44,600
Speaker 1: it wasn't really an earnest attempt to to create something

342
00:19:44,640 --> 00:19:48,159
Speaker 1: that would pass the quote unquote tearing tests, which I

343
00:19:48,200 --> 00:19:49,800
Speaker 1: don't think was being referred to as such at that

344
00:19:49,840 --> 00:19:52,240
Speaker 1: point yet. Yeah, it was kind of people knew about

345
00:19:52,280 --> 00:19:55,520
Speaker 1: tourings prediction, but it wasn't so much called a turing test.

346
00:19:56,000 --> 00:19:59,080
Speaker 1: Another chat bought that premiered in the early nineties seventies

347
00:19:59,119 --> 00:20:02,880
Speaker 1: went even further and actually was an attempt to try

348
00:20:03,000 --> 00:20:06,000
Speaker 1: and pass the Turing test in a very specific approach,

349
00:20:06,080 --> 00:20:08,560
Speaker 1: kind of like you know, Eugene Goostman, is a very

350
00:20:08,560 --> 00:20:12,200
Speaker 1: specific approach to narrow down that those parameters. In this case,

351
00:20:12,280 --> 00:20:15,639
Speaker 1: this one was called Perry p A R R Y

352
00:20:15,920 --> 00:20:19,520
Speaker 1: A Right, and Kenneth Colby created it too to emulate

353
00:20:19,800 --> 00:20:24,080
Speaker 1: a patient who has paranoid schizophrenia. Yeah, someone who has

354
00:20:24,200 --> 00:20:28,360
Speaker 1: you know, sort of a persecution complex. Uh. They imagined

355
00:20:28,440 --> 00:20:32,639
Speaker 1: that's that there are people or other entities that are

356
00:20:32,680 --> 00:20:36,000
Speaker 1: out to get them. And in this uh specific case,

357
00:20:36,320 --> 00:20:40,080
Speaker 1: he he kinda he kind of really embraced this approach.

358
00:20:40,600 --> 00:20:43,840
Speaker 1: It reminds me of people who create a like a

359
00:20:43,960 --> 00:20:46,080
Speaker 1: Dungeons and Dragon's character, but then give their character an

360
00:20:46,200 --> 00:20:49,960
Speaker 1: entire backstory. Yeah. Yeah, yeah, this this Perry persona was

361
00:20:50,040 --> 00:20:52,040
Speaker 1: an entire persona. It was a twenty eight year old

362
00:20:52,080 --> 00:20:54,400
Speaker 1: man with a job as a post office clerk who

363
00:20:54,440 --> 00:20:57,240
Speaker 1: was single, had no brothers and sisters, and rarely saw

364
00:20:57,359 --> 00:21:00,119
Speaker 1: his parents. He had specific hobbies. He liked to of

365
00:21:00,160 --> 00:21:03,240
Speaker 1: the movies and horse racing. Yeah, he liked to bet

366
00:21:03,240 --> 00:21:06,720
Speaker 1: on the horses and placed a bet with bookies in

367
00:21:06,720 --> 00:21:11,240
Speaker 1: the past. Right, and that he realized later bookies have

368
00:21:11,359 --> 00:21:15,639
Speaker 1: an association with the criminal underworld, and that therefore the

369
00:21:15,800 --> 00:21:19,240
Speaker 1: mafia knew about him and we're out to get him. Now. Now,

370
00:21:19,280 --> 00:21:21,479
Speaker 1: all of that might sound very ridiculous to you if

371
00:21:21,520 --> 00:21:24,400
Speaker 1: you never had any kind of interaction with someone who

372
00:21:24,440 --> 00:21:27,560
Speaker 1: suffers from paranoid schizophrenia, that can seem like, well, that

373
00:21:27,640 --> 00:21:31,600
Speaker 1: seems cartoonish. But no, this, this kind of thinking is

374
00:21:31,680 --> 00:21:34,719
Speaker 1: not uncommon, you know, whether it's whether it's a criminal

375
00:21:34,800 --> 00:21:40,760
Speaker 1: organization or the government or some other even unnamed entity. Oh. Absolutely, Uh,

376
00:21:40,880 --> 00:21:44,000
Speaker 1: it's it's very much realistic in terms of that kind

377
00:21:44,000 --> 00:21:46,560
Speaker 1: of diagnosis. And and if I was speaking about it

378
00:21:46,600 --> 00:21:48,320
Speaker 1: with humor in my voice a moment ago, it's it's

379
00:21:48,320 --> 00:21:51,800
Speaker 1: only because I am absolutely tickled that the programmers of

380
00:21:52,000 --> 00:21:56,200
Speaker 1: this of this program built it, built it in. Yeah,

381
00:21:56,880 --> 00:21:59,800
Speaker 1: it's it's actually pretty entertaining that they went so far

382
00:21:59,840 --> 00:22:03,199
Speaker 1: as to make this whole uh, this whole backstory to explain,

383
00:22:03,240 --> 00:22:06,280
Speaker 1: because that's what gives it the believability, right, and then

384
00:22:06,440 --> 00:22:09,359
Speaker 1: they ended up testing it in conversation with a human.

385
00:22:09,359 --> 00:22:13,080
Speaker 1: Perry would gradually start to introduce his thoughts quote unquote

386
00:22:13,200 --> 00:22:17,560
Speaker 1: thoughts about being persecuted, and would respond sensitively to anything

387
00:22:17,600 --> 00:22:21,480
Speaker 1: said about his appearance, family, or religious beliefs. I've actually

388
00:22:21,520 --> 00:22:24,760
Speaker 1: seen lots of transcripts of conversations with Perry, and sure enough,

389
00:22:25,000 --> 00:22:27,360
Speaker 1: it's one of those things where you know, you might

390
00:22:27,720 --> 00:22:30,080
Speaker 1: have a few exchanges and then Perry ends up saying

391
00:22:30,240 --> 00:22:34,000
Speaker 1: something that seems really odd, but not so odd as

392
00:22:34,040 --> 00:22:37,120
Speaker 1: to seem artificial. It just seems like it's a non sequitur,

393
00:22:37,720 --> 00:22:39,760
Speaker 1: you know, something like, well, that's what they want you

394
00:22:39,800 --> 00:22:41,240
Speaker 1: to think. And if you were to say, who are

395
00:22:41,280 --> 00:22:44,280
Speaker 1: they the mafia? Who did you think I was talking about?

396
00:22:45,000 --> 00:22:48,199
Speaker 1: Like the mafia after you? Of course they're after me.

397
00:22:48,720 --> 00:22:50,320
Speaker 1: They know who I am. That kind of stuff, and

398
00:22:50,560 --> 00:22:53,439
Speaker 1: it's disturbing, like it's you know, when you know what

399
00:22:53,520 --> 00:22:55,680
Speaker 1: it is, it's kind of amusing, but like if you're

400
00:22:55,680 --> 00:23:00,320
Speaker 1: in the middle of a conversation, oh, this poor person. Yeah.

401
00:23:00,680 --> 00:23:03,920
Speaker 1: And so in order to test it, uh Kolby did

402
00:23:03,920 --> 00:23:06,400
Speaker 1: a couple of different things. He did one test where

403
00:23:06,440 --> 00:23:11,240
Speaker 1: eight psychiatrists interviewed both Perry and a human patient via teletypewriter.

404
00:23:11,359 --> 00:23:15,919
Speaker 1: So in both cases, the the the psychiatrists could not

405
00:23:16,080 --> 00:23:18,000
Speaker 1: see who they were interviewing. This is going back to

406
00:23:18,040 --> 00:23:20,720
Speaker 1: the kind of the original touring test. Idea, or at

407
00:23:20,760 --> 00:23:25,119
Speaker 1: least Touring's proposed experiment. And in this case, only two

408
00:23:25,240 --> 00:23:27,240
Speaker 1: of the eight were able to identify that one of

409
00:23:27,240 --> 00:23:29,680
Speaker 1: the interviewees was human and the other was a machine.

410
00:23:30,200 --> 00:23:32,439
Speaker 1: In a second test, Colby presented a group of a

411
00:23:32,520 --> 00:23:40,159
Speaker 1: hundreds psychiatrists, most a psychics, transcripts of interviews between Perry. Yeah,

412
00:23:40,200 --> 00:23:44,920
Speaker 1: he had, he had pairs of exactly food, I foresee. No.

413
00:23:45,119 --> 00:23:49,600
Speaker 1: But they gave these psychiatrists transcripts of interviews between an

414
00:23:49,640 --> 00:23:52,359
Speaker 1: interviewer and Perry, and an interviewer and a human patient,

415
00:23:53,119 --> 00:23:56,439
Speaker 1: and forty out of the one responded. I don't know

416
00:23:56,480 --> 00:23:58,879
Speaker 1: if the other sixty just never got it or if

417
00:23:58,920 --> 00:24:02,239
Speaker 1: they didn't take The response rates are variable, right, so

418
00:24:02,359 --> 00:24:05,680
Speaker 1: I'll the forty who responded, nineteen of them guessed incorrectly.

419
00:24:06,040 --> 00:24:10,159
Speaker 1: So that's almost a fifty percent, you know, getting you know,

420
00:24:10,320 --> 00:24:15,160
Speaker 1: right up there with a pretty impressive amount. Now, again,

421
00:24:16,280 --> 00:24:19,359
Speaker 1: we have to look at the fact that Perry is

422
00:24:19,400 --> 00:24:24,080
Speaker 1: operating under a very restricted set of rules. We're talking

423
00:24:24,119 --> 00:24:28,320
Speaker 1: about paranoid schizophrenic, someone who we would assume would occasionally

424
00:24:28,760 --> 00:24:33,560
Speaker 1: have a non normative answers exactly to conversational pieces. And

425
00:24:33,560 --> 00:24:36,120
Speaker 1: and again it's it's a limited time that you're having

426
00:24:36,160 --> 00:24:38,920
Speaker 1: with this person or this entity in this case, this

427
00:24:38,920 --> 00:24:45,840
Speaker 1: this program, and uh, because the psychiatrists had a specific

428
00:24:45,920 --> 00:24:48,480
Speaker 1: expectation of the type of interactions they were going to

429
00:24:48,520 --> 00:24:56,200
Speaker 1: see that could have affected their their answer right. So ideally,

430
00:24:56,640 --> 00:25:00,359
Speaker 1: in the perfect situation, you would have this in review

431
00:25:00,440 --> 00:25:03,879
Speaker 1: happening where you have no expectation as to what the

432
00:25:03,920 --> 00:25:07,719
Speaker 1: answers should be. In other words, you don't know ahead

433
00:25:07,720 --> 00:25:12,320
Speaker 1: of time that the interviewee is having any kind of

434
00:25:12,320 --> 00:25:15,199
Speaker 1: other any kind of restrictions upon that person, so that

435
00:25:15,280 --> 00:25:18,560
Speaker 1: you would be interviewing anyone like any average person. But

436
00:25:18,640 --> 00:25:21,040
Speaker 1: that's obviously not what we're talking about here, nor was

437
00:25:21,080 --> 00:25:25,720
Speaker 1: it the one for Eugene Goosband. So I've also seen

438
00:25:25,800 --> 00:25:30,119
Speaker 1: by the way transcripts of people who set up Eliza

439
00:25:30,240 --> 00:25:32,720
Speaker 1: and Perry to talk to each other. So you have

440
00:25:32,760 --> 00:25:37,640
Speaker 1: Eliza acting as the Rogerian psychiatrist and Perry the paranoid schizophrenic,

441
00:25:38,000 --> 00:25:42,520
Speaker 1: having bizarre conversations, and they usually don't last very long

442
00:25:42,560 --> 00:25:47,159
Speaker 1: because Perry gets upset, And obviously by gets upset, I

443
00:25:47,200 --> 00:25:49,399
Speaker 1: just mean that Perry ends up essentially shutting down the

444
00:25:49,440 --> 00:25:53,280
Speaker 1: conversation because Eliza just wants to ask questions, and Perry

445
00:25:53,320 --> 00:25:56,439
Speaker 1: gets suspicious of people who are asking questions, and by

446
00:25:56,560 --> 00:25:59,479
Speaker 1: again get suspicious, I'm saying following specific rules that make

447
00:25:59,520 --> 00:26:02,800
Speaker 1: it feel like this computer programs getting suspicious, but they

448
00:26:02,800 --> 00:26:06,000
Speaker 1: are entertaining. If you ever do a search online, just

449
00:26:06,080 --> 00:26:09,560
Speaker 1: look for Eliza Perry transcripts. There. There are a few

450
00:26:09,600 --> 00:26:12,560
Speaker 1: of them, and they're all pretty entertaining. But so since

451
00:26:12,600 --> 00:26:15,440
Speaker 1: since that time, I mean obviously, lots and lots of

452
00:26:15,520 --> 00:26:18,719
Speaker 1: chatbots have been created for multiple reasons. Oh yeah, well,

453
00:26:18,760 --> 00:26:20,720
Speaker 1: you know, some of them are trying to test the

454
00:26:20,760 --> 00:26:23,919
Speaker 1: Turing test, and others are trying to fool you into

455
00:26:24,320 --> 00:26:27,240
Speaker 1: giving out your credit card information or planing on a

456
00:26:27,320 --> 00:26:31,800
Speaker 1: link that has malicious uh links to malicious software. Yeah.

457
00:26:32,359 --> 00:26:37,080
Speaker 1: Anyone who's been on any kind of chat program, specifically

458
00:26:37,119 --> 00:26:40,760
Speaker 1: like AIM has probably encountered this at least once or twice,

459
00:26:40,800 --> 00:26:45,680
Speaker 1: where they're they're getting an unsolicited message from someone or

460
00:26:45,720 --> 00:26:48,280
Speaker 1: something or something. Yeah, and if you type a couple

461
00:26:48,280 --> 00:26:51,280
Speaker 1: of times, you realize, oh, this is not actually this

462
00:26:51,320 --> 00:26:54,080
Speaker 1: is an attempt for to either get information from me

463
00:26:54,200 --> 00:26:56,800
Speaker 1: or have me click on a link. Yeah, that's a

464
00:26:57,200 --> 00:27:01,760
Speaker 1: that's the thing. But there are some example of I

465
00:27:01,800 --> 00:27:05,160
Speaker 1: guess saying legitimate is weird. But there are some examples

466
00:27:05,200 --> 00:27:09,600
Speaker 1: of other more and more scholarly attempts. Yeah, like a

467
00:27:09,600 --> 00:27:15,960
Speaker 1: PC therapist. Uh. That one was by Joseph Weintraub, and

468
00:27:16,119 --> 00:27:19,280
Speaker 1: it fooled of its judges into thinking that it was human.

469
00:27:20,760 --> 00:27:23,639
Speaker 1: Of ten judges, of course, so if I'm doing my

470
00:27:23,680 --> 00:27:26,760
Speaker 1: math correctly, that's five that I believe you are. Unless

471
00:27:26,800 --> 00:27:30,439
Speaker 1: we're talking about quantum judges. Um, these judges were both

472
00:27:30,560 --> 00:27:33,719
Speaker 1: right and wrong at the same time. And um, and

473
00:27:33,800 --> 00:27:36,600
Speaker 1: it was it was a whimsical program. I guess you

474
00:27:36,600 --> 00:27:39,400
Speaker 1: could say, yeah, it had come some I think it's

475
00:27:39,400 --> 00:27:42,760
Speaker 1: fair to say its answers could be pretty smart ass. Also,

476
00:27:43,000 --> 00:27:45,840
Speaker 1: I read some of these transcripts and it actually surprised

477
00:27:45,880 --> 00:27:48,600
Speaker 1: me that enough judges thought that it was a person.

478
00:27:49,080 --> 00:27:51,760
Speaker 1: Maybe they thought it was a person who was purposefully

479
00:27:51,880 --> 00:27:55,040
Speaker 1: attempting to fool them into thinking it was a computer. Uh,

480
00:27:55,280 --> 00:27:57,320
Speaker 1: you know, because it would say things like I compute,

481
00:27:57,359 --> 00:27:59,919
Speaker 1: therefore I am that kind of stuff where it was

482
00:28:00,000 --> 00:28:03,760
Speaker 1: specifically Yeah, so you're looking at this stuff and you're thinking,

483
00:28:03,840 --> 00:28:05,400
Speaker 1: all right, well, maybe this is kind of going back

484
00:28:05,440 --> 00:28:09,240
Speaker 1: to that original touring test party game where the the

485
00:28:09,280 --> 00:28:12,119
Speaker 1: effort is for a person like the person who's being

486
00:28:12,119 --> 00:28:15,240
Speaker 1: interviewed is trying to throw everybody off, And that's perfectly

487
00:28:15,280 --> 00:28:18,240
Speaker 1: within the rules unless you stay upfront. No, just be

488
00:28:18,320 --> 00:28:20,920
Speaker 1: honest in your in your answers. There's nothing in here,

489
00:28:20,960 --> 00:28:23,800
Speaker 1: by the way that says that the interviewee has to

490
00:28:23,840 --> 00:28:26,800
Speaker 1: tell the truth necessarily unless you just state that as

491
00:28:26,800 --> 00:28:29,480
Speaker 1: a parameter at the beginning. So in other words, you

492
00:28:29,520 --> 00:28:31,679
Speaker 1: could you could be like, I'm totally the computer and

493
00:28:31,720 --> 00:28:34,639
Speaker 1: you're the human being interviewed. Uh. I don't know if

494
00:28:34,640 --> 00:28:37,440
Speaker 1: that's a fair way of saying that the device won

495
00:28:37,560 --> 00:28:40,480
Speaker 1: or lost, but it is a possibility. Uh. Then we

496
00:28:40,560 --> 00:28:42,160
Speaker 1: have a two thousand eleven. Now, this one is a

497
00:28:42,200 --> 00:28:46,640
Speaker 1: really pretty impressive one. And this was clever Butt, which

498
00:28:46,680 --> 00:28:49,800
Speaker 1: was made by a fellow named Rollo Carpenter, and it

499
00:28:49,920 --> 00:28:53,520
Speaker 1: fooled fifty nine point three percent of a live audience

500
00:28:53,560 --> 00:28:57,960
Speaker 1: and an event in India with more than a thousand people. Yeah,

501
00:28:58,000 --> 00:29:00,160
Speaker 1: the way this worked was that the audience watched as

502
00:29:00,200 --> 00:29:04,440
Speaker 1: interviewers interacted via text with either clever Bot or a

503
00:29:04,520 --> 00:29:08,120
Speaker 1: human in the course of a four minute interview. So

504
00:29:08,160 --> 00:29:11,240
Speaker 1: it's a little shorter than what Turing had said, but

505
00:29:11,280 --> 00:29:13,600
Speaker 1: not by not by a whole lot. In fourmants is

506
00:29:13,640 --> 00:29:15,800
Speaker 1: still a good amount of time. Sure that that is

507
00:29:15,840 --> 00:29:20,560
Speaker 1: twent less, that's true. That's true, so keep that in mind.

508
00:29:20,640 --> 00:29:23,320
Speaker 1: But but at any rate, it was you know, pretty

509
00:29:23,320 --> 00:29:27,760
Speaker 1: interesting experience. And also from why I read they misidentified,

510
00:29:28,640 --> 00:29:31,800
Speaker 1: they thought that the human was a computer sixty of

511
00:29:31,840 --> 00:29:36,240
Speaker 1: the time. Um, because they didn't necessarily just say that

512
00:29:36,360 --> 00:29:39,880
Speaker 1: it was a computer or it was clever. But so

513
00:29:41,000 --> 00:29:44,120
Speaker 1: now we see that there are a few examples of

514
00:29:44,240 --> 00:29:49,000
Speaker 1: chat bots quote unquote passing the Turing test. So what

515
00:29:49,040 --> 00:29:51,760
Speaker 1: does that mean? Does it mean that the machines are

516
00:29:51,760 --> 00:29:57,800
Speaker 1: actually thinking? Um? No, I mean, and it's not to

517
00:29:57,840 --> 00:29:59,880
Speaker 1: say that that computers don't have a certain amount of

518
00:30:00,200 --> 00:30:04,000
Speaker 1: machine intelligence, but there's absolutely a distinction between that and

519
00:30:04,200 --> 00:30:07,400
Speaker 1: what we consider to be human intelligence. That's true. The

520
00:30:07,440 --> 00:30:10,560
Speaker 1: programmers themselves have said that this doesn't mean a machine

521
00:30:10,600 --> 00:30:14,920
Speaker 1: is able to think. Uh, they're just able to interpret

522
00:30:15,720 --> 00:30:18,440
Speaker 1: commands then to follow a set of rules to make

523
00:30:18,480 --> 00:30:21,880
Speaker 1: a response, which is still pretty cool. And and it

524
00:30:21,920 --> 00:30:24,680
Speaker 1: doesn't It certainly doesn't mean that the Turing test is

525
00:30:25,160 --> 00:30:28,320
Speaker 1: worthless as an exercise. No, it is. In fact, it's

526
00:30:28,360 --> 00:30:33,360
Speaker 1: improving our ability to create programs that that can understand

527
00:30:34,000 --> 00:30:37,800
Speaker 1: or at least respond to natural language. Natural language recognition

528
00:30:37,800 --> 00:30:40,120
Speaker 1: is one of those big things where if you're really

529
00:30:40,160 --> 00:30:42,960
Speaker 1: able to crack it, then you can have some amazing

530
00:30:43,040 --> 00:30:45,640
Speaker 1: opportunities open up. And we've seen this recently with stuff

531
00:30:45,680 --> 00:30:49,160
Speaker 1: like Siri. Oh absolutely being able to speak to your

532
00:30:49,200 --> 00:30:52,400
Speaker 1: computer rather than having to I mean, even if you

533
00:30:52,400 --> 00:30:54,880
Speaker 1: could speak to your computer through a keyboard and have

534
00:30:54,920 --> 00:30:57,520
Speaker 1: it understand what you're what you're saying, I mean, like

535
00:30:57,840 --> 00:31:00,960
Speaker 1: it's it's the reason why Google spend so much time

536
00:31:00,960 --> 00:31:03,560
Speaker 1: and money, and it's search algorithms of of trying to

537
00:31:03,560 --> 00:31:06,640
Speaker 1: figure out what you really mean when you search for

538
00:31:06,680 --> 00:31:12,000
Speaker 1: a certain phrase, because traditionally, you know, before we really

539
00:31:12,000 --> 00:31:16,440
Speaker 1: got into the natural language recognition era, it meant that

540
00:31:16,480 --> 00:31:18,040
Speaker 1: in order to work with a computer, you had to

541
00:31:18,040 --> 00:31:20,480
Speaker 1: work with a computer on the computer's terms. You had

542
00:31:20,520 --> 00:31:22,720
Speaker 1: to learn the commands, you had to learn the way

543
00:31:22,760 --> 00:31:25,120
Speaker 1: to navigate a computer system in order for it to

544
00:31:25,120 --> 00:31:27,520
Speaker 1: do what you wanted it to do. Once you get

545
00:31:27,560 --> 00:31:31,520
Speaker 1: to a point where natural language recognition software is robust

546
00:31:31,640 --> 00:31:35,560
Speaker 1: enough the computer is working on your terms, you you

547
00:31:35,600 --> 00:31:39,920
Speaker 1: can put in however you're thinking, like whatever, whatever mental

548
00:31:40,080 --> 00:31:44,080
Speaker 1: exercise you've gone through to ask this computer to do something.

549
00:31:44,480 --> 00:31:48,160
Speaker 1: Whatever you do to kind of express a thought, uh

550
00:31:48,440 --> 00:31:51,040
Speaker 1: as a command to this computer. The computer can then

551
00:31:51,720 --> 00:31:55,160
Speaker 1: interpret then respond And it's not just for serving you

552
00:31:55,200 --> 00:31:58,400
Speaker 1: back whatever information you happen to be looking for. It's

553
00:31:58,560 --> 00:32:00,840
Speaker 1: I mean, I mean, we're talking about being able to

554
00:32:00,920 --> 00:32:04,160
Speaker 1: just look at a computer and say, you know, I

555
00:32:04,240 --> 00:32:08,200
Speaker 1: really want a graph that looks blue and has these

556
00:32:08,240 --> 00:32:11,880
Speaker 1: percentages in it and it is about this thing, and

557
00:32:11,920 --> 00:32:14,040
Speaker 1: it just does it. Yeah, like I want to see

558
00:32:14,040 --> 00:32:17,720
Speaker 1: what the population distribution of Atlanta is in a bar

559
00:32:17,880 --> 00:32:20,840
Speaker 1: chart or something, and then it could bring that, goes out,

560
00:32:20,920 --> 00:32:24,080
Speaker 1: finds that information, put it into a bar chart, and yeah,

561
00:32:24,120 --> 00:32:26,800
Speaker 1: that's pretty phenomenal stuff. We see other examples of machine

562
00:32:26,840 --> 00:32:32,480
Speaker 1: intelligence everywhere, things like pattern recognition, probabilistic predictions, for example, Pandora.

563
00:32:32,880 --> 00:32:35,560
Speaker 1: You know the Music Genome project. It's yeah, that's that's

564
00:32:35,560 --> 00:32:39,920
Speaker 1: pattern recognition. It's looking for elements of songs that you

565
00:32:39,960 --> 00:32:42,600
Speaker 1: say you like, and then looking for other stuff that's

566
00:32:42,640 --> 00:32:45,640
Speaker 1: not in the specific category you mentioned or the specific

567
00:32:45,720 --> 00:32:49,640
Speaker 1: examples you mentioned, and there's something else you probably will

568
00:32:49,680 --> 00:32:51,360
Speaker 1: like because you like these other things that also have

569
00:32:51,440 --> 00:32:54,720
Speaker 1: this stuff in it. Sure, Uh, you know sometimes that's

570
00:32:54,800 --> 00:32:57,640
Speaker 1: less functional than other times. It makes me think of

571
00:32:57,800 --> 00:33:00,880
Speaker 1: Patton Oswald has a great routine about Evo that does

572
00:33:00,920 --> 00:33:02,720
Speaker 1: the same sort of thing where he says, you know,

573
00:33:02,760 --> 00:33:05,480
Speaker 1: TiVo is great. I mean, I like, I love Westerns,

574
00:33:05,520 --> 00:33:08,400
Speaker 1: and I'll have it set to TiVo a Western for me,

575
00:33:08,440 --> 00:33:09,920
Speaker 1: and I come back, and then there'll be all these

576
00:33:09,920 --> 00:33:12,720
Speaker 1: other Westerns that will be suggested. I didn't even know about.

577
00:33:13,160 --> 00:33:16,080
Speaker 1: Thank you, TiVo. But then sometimes TiVo gets it wrong

578
00:33:16,400 --> 00:33:19,200
Speaker 1: and I come back and everything has got horses in it,

579
00:33:19,760 --> 00:33:22,080
Speaker 1: because Westerns have horses in it. So I've got my

580
00:33:22,160 --> 00:33:25,000
Speaker 1: little pony and cartoons with horses and unicorns and things,

581
00:33:25,200 --> 00:33:28,160
Speaker 1: and I have to say, no, TiVo, that's a bad TiVo.

582
00:33:28,400 --> 00:33:32,080
Speaker 1: But TiVo says, but you said you liked horses. Same

583
00:33:32,120 --> 00:33:34,920
Speaker 1: sort of thing. Like when you get more sophisticated than

584
00:33:35,160 --> 00:33:39,320
Speaker 1: the the computer program starts to anticipate things and makes

585
00:33:39,360 --> 00:33:42,920
Speaker 1: these probabilistic models. These these models where there there are

586
00:33:42,960 --> 00:33:48,280
Speaker 1: certain percentages associated with various responses, and it goes with

587
00:33:48,320 --> 00:33:51,880
Speaker 1: whichever one seems to be the most prevalent, assuming it

588
00:33:51,920 --> 00:33:54,920
Speaker 1: meets a threshold. If this sounds familiar to you, it's

589
00:33:54,920 --> 00:33:59,280
Speaker 1: because that's how ibms Watson worked, which is a really

590
00:33:59,320 --> 00:34:03,280
Speaker 1: good example of natural language recognition absolutely because not only

591
00:34:03,360 --> 00:34:06,240
Speaker 1: was it able to recognize natural language, it had to

592
00:34:06,960 --> 00:34:11,440
Speaker 1: interpret things like word play Jeopardy. This is the machine

593
00:34:11,480 --> 00:34:14,480
Speaker 1: that went up on Jeopardy and beat the returning champions

594
00:34:15,360 --> 00:34:18,520
Speaker 1: or former champions. Uh And you know, if you've ever

595
00:34:18,640 --> 00:34:21,759
Speaker 1: played Jeopardy or watch Jeopardy, you know that there are

596
00:34:21,880 --> 00:34:26,319
Speaker 1: categories that depend on things like puns or hominem's or

597
00:34:26,680 --> 00:34:29,720
Speaker 1: other forms of word play. So it has to parse

598
00:34:29,800 --> 00:34:32,359
Speaker 1: all of that, and that's even more complicated than just

599
00:34:32,400 --> 00:34:35,120
Speaker 1: taking a simple sentence and figuring out, all, right, what

600
00:34:35,200 --> 00:34:39,240
Speaker 1: are the potential responses to whatever this this this phrase

601
00:34:39,480 --> 00:34:42,799
Speaker 1: is so great, great example. You know, it would end

602
00:34:42,840 --> 00:34:46,480
Speaker 1: up coming up with a potential answer. It would assign

603
00:34:46,480 --> 00:34:49,520
Speaker 1: a percentage of how quote unquote sure it was that

604
00:34:49,520 --> 00:34:52,279
Speaker 1: that was the right answer, and if the percentage was

605
00:34:52,360 --> 00:34:55,520
Speaker 1: higher than its threshold, which I think was something like that,

606
00:34:56,200 --> 00:34:58,240
Speaker 1: it would buzz in and give that as a guess.

607
00:34:58,880 --> 00:35:02,319
Speaker 1: Sometimes it was wrong, but it was right a lot

608
00:35:02,360 --> 00:35:05,960
Speaker 1: of the time, so that's kind of cool. Uh So,

609
00:35:07,040 --> 00:35:10,279
Speaker 1: getting back to Eugene Gene, the main machine as I

610
00:35:10,280 --> 00:35:13,240
Speaker 1: called him in in my notes, at one point uh,

611
00:35:13,320 --> 00:35:15,879
Speaker 1: and of course I'm anthropomorphizing when I say him, it's

612
00:35:15,880 --> 00:35:18,000
Speaker 1: a it's a it's an it. Well, it has a

613
00:35:18,080 --> 00:35:19,880
Speaker 1: dude name. Yeah, it has a dude name and a

614
00:35:19,960 --> 00:35:26,120
Speaker 1: dude persona. But it's ultimately, isn't it. Would you say

615
00:35:26,200 --> 00:35:30,279
Speaker 1: that perhaps some of the reporting around this was was

616
00:35:30,360 --> 00:35:34,680
Speaker 1: maybe a little misleading or at least hype is well,

617
00:35:35,360 --> 00:35:39,920
Speaker 1: you know, okay, the entire Eugene Gooseman chat bought sounds

618
00:35:40,040 --> 00:35:43,520
Speaker 1: really cool. I haven't met it personally, No, I haven't either,

619
00:35:43,560 --> 00:35:46,440
Speaker 1: although you can. There is an Internet version, and I'm

620
00:35:46,480 --> 00:35:48,399
Speaker 1: not sure that it's the same version that's being used

621
00:35:48,400 --> 00:35:51,240
Speaker 1: in competition, because I've seen some transcripts from the Internet

622
00:35:51,320 --> 00:35:55,600
Speaker 1: version and they don't seem good at all. They seem bad, right,

623
00:35:55,800 --> 00:35:58,040
Speaker 1: I guess you'd have to talk to some actual thirteen

624
00:35:58,120 --> 00:36:02,279
Speaker 1: year old boys from actually yeah, that well, this is

625
00:36:02,360 --> 00:36:05,120
Speaker 1: part of it. You know. There's certainly been some some

626
00:36:05,280 --> 00:36:11,200
Speaker 1: questions among uh natural language AI enthusiasts online about whether

627
00:36:11,800 --> 00:36:15,839
Speaker 1: we're really just lowering our expectations for human communication. Which, yeah,

628
00:36:15,840 --> 00:36:18,440
Speaker 1: that's that's a totally different way of looking at and

629
00:36:18,480 --> 00:36:21,440
Speaker 1: a depressing one to say that, Oh, well, if you

630
00:36:21,560 --> 00:36:24,520
Speaker 1: come from this place and if you are of this age,

631
00:36:24,600 --> 00:36:27,320
Speaker 1: then I expect you to only be able to communicate

632
00:36:27,360 --> 00:36:33,479
Speaker 1: at this level, right, which is depressing. It certainly is well,

633
00:36:33,560 --> 00:36:35,439
Speaker 1: but you know, but it's a valid point. I think.

634
00:36:35,480 --> 00:36:37,480
Speaker 1: I think it's a good thing to be thinking about

635
00:36:37,520 --> 00:36:41,760
Speaker 1: in this kind of situation. UM. You know. Beyond that though,

636
00:36:41,800 --> 00:36:47,080
Speaker 1: and I certainly don't want to downplay the apparent achievements

637
00:36:47,160 --> 00:36:50,960
Speaker 1: of its programmers, because I haven't programmed any capable chatbots

638
00:36:51,000 --> 00:36:56,839
Speaker 1: today ever since I did. But there are a few

639
00:36:56,880 --> 00:36:58,960
Speaker 1: things that are just a little bit shady about the

640
00:36:59,000 --> 00:37:03,480
Speaker 1: news UM. First off, the original press release, which came

641
00:37:03,480 --> 00:37:07,319
Speaker 1: out of the University of Reading, I believe so UM

642
00:37:07,520 --> 00:37:12,239
Speaker 1: stated that a quote supercomputer had achieved this feat, and

643
00:37:12,520 --> 00:37:16,920
Speaker 1: perhaps charitably, it was a mistake or misunderstanding on the

644
00:37:16,960 --> 00:37:19,239
Speaker 1: part of the writer of the press release, But some

645
00:37:19,280 --> 00:37:23,040
Speaker 1: skeptics have suggested that it was in fact a purposeful

646
00:37:23,080 --> 00:37:25,920
Speaker 1: publicity play that in fact worked, because a whole lot

647
00:37:25,960 --> 00:37:31,279
Speaker 1: of news headlines around the interwebs repeated the error very excitedly. Yes,

648
00:37:31,320 --> 00:37:33,640
Speaker 1: because it was not a supercomputer. It was a computer

649
00:37:33,760 --> 00:37:36,719
Speaker 1: running a piece of software program. Yes, the program that

650
00:37:36,760 --> 00:37:40,359
Speaker 1: did the work. I mean, the computer just provided the horsepower. Right,

651
00:37:40,400 --> 00:37:42,560
Speaker 1: it's the software that did all the action work, and

652
00:37:42,560 --> 00:37:47,080
Speaker 1: it was not it wasn't on a supercomputer. Little little

653
00:37:47,680 --> 00:37:51,319
Speaker 1: known fact. Supercomputers have better things to do than run

654
00:37:51,400 --> 00:37:56,279
Speaker 1: chatbot software generally. Yeah. Yeah, we're talking about things like

655
00:37:56,320 --> 00:38:00,560
Speaker 1: figuring out global weather changing change pad and some things

656
00:38:00,600 --> 00:38:03,480
Speaker 1: like that, you know, or or the way that money works.

657
00:38:04,760 --> 00:38:07,880
Speaker 1: Chat Bots low on their priority list. It's like number

658
00:38:07,960 --> 00:38:12,680
Speaker 1: seven at least. Um And and also that this press

659
00:38:12,719 --> 00:38:17,840
Speaker 1: release in question was largely a quotation from Dr Kevin Warwick.

660
00:38:18,640 --> 00:38:22,319
Speaker 1: Kevin Warwick, of course being the fellow who organized this

661
00:38:22,600 --> 00:38:27,080
Speaker 1: entire competition, who who's an engineer and a futurist um

662
00:38:27,160 --> 00:38:31,280
Speaker 1: and also the instigator and or enjoyer of a certain

663
00:38:31,320 --> 00:38:37,440
Speaker 1: amount of hype and debate about future technologies. Yeah, he is,

664
00:38:37,640 --> 00:38:41,160
Speaker 1: obviously you can tell. This is the guy who elected

665
00:38:41,160 --> 00:38:43,399
Speaker 1: to have surgery performed on him so he could have

666
00:38:44,200 --> 00:38:46,200
Speaker 1: h that r F I D chip and call himself

667
00:38:46,239 --> 00:38:51,359
Speaker 1: a cybors. This is someone who not only embraces this

668
00:38:51,600 --> 00:38:54,960
Speaker 1: these ideas of futurism, but is actively trying to promote

669
00:38:55,040 --> 00:38:57,560
Speaker 1: them and get to them that we're not even saying

670
00:38:57,600 --> 00:38:59,319
Speaker 1: that that's a bad thing. What we are saying is

671
00:38:59,320 --> 00:39:03,360
Speaker 1: that may give him somewhat of a bias when it

672
00:39:03,400 --> 00:39:07,600
Speaker 1: comes to proclaiming a computer software piece of computer software

673
00:39:07,760 --> 00:39:11,680
Speaker 1: being an amazing achievement that beat the Turing test, right sure.

674
00:39:11,760 --> 00:39:15,239
Speaker 1: I mean he admits basically to to being a provocateur.

675
00:39:15,400 --> 00:39:17,520
Speaker 1: He he says that that's really his job, you know,

676
00:39:17,800 --> 00:39:22,040
Speaker 1: is to get people excited about tech and engineering and

677
00:39:22,040 --> 00:39:25,000
Speaker 1: and the future. And we get that like that that

678
00:39:25,200 --> 00:39:27,360
Speaker 1: we agree that a job too. We think it's red

679
00:39:27,560 --> 00:39:30,759
Speaker 1: We think that perhaps I don't want to put words

680
00:39:30,800 --> 00:39:33,600
Speaker 1: into your mouth, Lauren, I think perhaps that there's a

681
00:39:33,600 --> 00:39:35,800
Speaker 1: different way of going about it where you can still

682
00:39:36,120 --> 00:39:40,160
Speaker 1: be excited, but you can be a little more grounded

683
00:39:40,200 --> 00:39:42,440
Speaker 1: in the way you present things, because I I also

684
00:39:42,520 --> 00:39:45,040
Speaker 1: think the achievement of creating a chatbot that could be

685
00:39:45,600 --> 00:39:49,440
Speaker 1: uh convincing is a fantastic achievement. I mean, it's something that,

686
00:39:49,760 --> 00:39:52,759
Speaker 1: even under any number of qualifications, incredibly challenging to do,

687
00:39:52,840 --> 00:39:56,400
Speaker 1: no matter how you frame it. Um. I do think, however,

688
00:39:56,520 --> 00:40:00,120
Speaker 1: that if you seem to over inflate the achieve it,

689
00:40:00,480 --> 00:40:03,280
Speaker 1: you run the danger of making people feel jaded about

690
00:40:03,320 --> 00:40:08,360
Speaker 1: it later, which I think Computer Wolf exactly, yeah, exactly.

691
00:40:08,880 --> 00:40:11,719
Speaker 1: So it's one of those things where you know, you

692
00:40:11,800 --> 00:40:15,320
Speaker 1: have to take the context into account, right and don't

693
00:40:15,360 --> 00:40:18,400
Speaker 1: don't downplay the achievement. But don't sit there and say like,

694
00:40:18,440 --> 00:40:21,120
Speaker 1: ah ha, now we have intelligent computers everywhere. That's not

695
00:40:21,200 --> 00:40:24,800
Speaker 1: that's not true either. I saw there's a great Wired

696
00:40:24,880 --> 00:40:30,200
Speaker 1: article that specifically went into uh, kind of debunking the

697
00:40:30,239 --> 00:40:33,960
Speaker 1: whole beating the Turing test thing and again kind of

698
00:40:33,960 --> 00:40:36,080
Speaker 1: saying the same thing we're saying, like take the context

699
00:40:36,120 --> 00:40:39,239
Speaker 1: into account, and and part of that article they ended

700
00:40:39,320 --> 00:40:43,520
Speaker 1: up asking a cognitive scientist named Gary Marcus of n

701
00:40:43,640 --> 00:40:48,040
Speaker 1: y U about this, and Marcus proposed a new version

702
00:40:48,200 --> 00:40:51,040
Speaker 1: of the Turing test because he says, the old version

703
00:40:51,160 --> 00:40:54,480
Speaker 1: is not really a measurement of machine intelligence. Uh, it

704
00:40:54,880 --> 00:40:58,880
Speaker 1: does kind of illustrate ways of creating natural language recognition

705
00:40:59,000 --> 00:41:03,080
Speaker 1: and clever way is to fool the human side, uh huh.

706
00:41:03,160 --> 00:41:06,040
Speaker 1: And that it was very valid historically at the time

707
00:41:06,160 --> 00:41:10,720
Speaker 1: because you know, text textual communication was new and exciting

708
00:41:10,800 --> 00:41:13,320
Speaker 1: and it was you know, it pushed the field forward,

709
00:41:13,360 --> 00:41:15,319
Speaker 1: it really did. But now we've gotten to a point

710
00:41:15,320 --> 00:41:17,880
Speaker 1: where fooling the person on the other side of a

711
00:41:17,960 --> 00:41:21,600
Speaker 1: keyboard is not necessarily the goal that we should be

712
00:41:21,640 --> 00:41:24,279
Speaker 1: looking at he proposes. The next version of the Turing

713
00:41:24,360 --> 00:41:28,520
Speaker 1: test should be that a computer software, uh, like any

714
00:41:28,600 --> 00:41:30,600
Speaker 1: kind of program that wants to beat it, what it

715
00:41:30,640 --> 00:41:34,880
Speaker 1: has to do is first quote unquote watch a movie,

716
00:41:34,920 --> 00:41:38,279
Speaker 1: television show, YouTube video, something, some kind of video media,

717
00:41:38,560 --> 00:41:41,680
Speaker 1: and then be able to respond to questions about it.

718
00:41:41,760 --> 00:41:43,759
Speaker 1: So sort of like here, let me show you this

719
00:41:43,840 --> 00:41:48,560
Speaker 1: ten minute video on car safety, and then asking questions

720
00:41:48,560 --> 00:41:52,360
Speaker 1: about specifically about the video, what happened after they fastened

721
00:41:52,360 --> 00:41:54,279
Speaker 1: the seat belt, that kind of thing, and if the

722
00:41:54,280 --> 00:41:57,359
Speaker 1: computer program is able to answer it, then that would

723
00:41:57,400 --> 00:42:00,960
Speaker 1: be a much more convincing touring test than just kind

724
00:42:00,960 --> 00:42:05,200
Speaker 1: of spewing out a script, which is, you know, it's

725
00:42:05,200 --> 00:42:08,280
Speaker 1: adding another layer of difficulty on top of an already

726
00:42:08,280 --> 00:42:10,719
Speaker 1: difficult task. But that's that's the whole only way you

727
00:42:10,760 --> 00:42:14,120
Speaker 1: can go forward. Otherwise we're just going to see increasingly

728
00:42:14,160 --> 00:42:17,480
Speaker 1: sophisticated chat box. Yeah, and and that is actually a

729
00:42:17,600 --> 00:42:20,720
Speaker 1: very difficult and interesting problem. What wasn't it just recently

730
00:42:20,760 --> 00:42:23,080
Speaker 1: that there was a computer that we we taught I

731
00:42:23,080 --> 00:42:26,239
Speaker 1: mean not us personally, but that humanity, some researchers, and

732
00:42:26,320 --> 00:42:30,719
Speaker 1: we're taught taught to identify cats pictures of cats that

733
00:42:30,800 --> 00:42:34,319
Speaker 1: was the AI program that essentially went through thousands and

734
00:42:34,360 --> 00:42:37,560
Speaker 1: thousands of uh I think it was images and videos

735
00:42:37,680 --> 00:42:41,319
Speaker 1: and then became able to identify cats, essentially defined what

736
00:42:41,480 --> 00:42:44,480
Speaker 1: a cat was because no one taught it right. It

737
00:42:44,880 --> 00:42:48,640
Speaker 1: learned what cats are based upon their appearance, on their

738
00:42:48,640 --> 00:42:51,560
Speaker 1: appearance and can look at pictures of cats and say

739
00:42:51,600 --> 00:42:54,719
Speaker 1: that is totally a cat, essentially saying that thing that

740
00:42:54,840 --> 00:42:56,960
Speaker 1: is in that video is the same as this other

741
00:42:57,040 --> 00:42:58,880
Speaker 1: thing that's in this picture. That's the same as this

742
00:42:58,960 --> 00:43:01,560
Speaker 1: other thing that's in this totally different video, which sounds

743
00:43:01,800 --> 00:43:05,840
Speaker 1: trivial and hilarious, and it kind of is hilarious. Also

744
00:43:05,960 --> 00:43:08,320
Speaker 1: easy because I mean, come on, everything on the internet

745
00:43:08,360 --> 00:43:10,759
Speaker 1: has cats in it. Yeah, that is That is kind

746
00:43:10,760 --> 00:43:12,799
Speaker 1: of a gimme, isn't it? But still no, it is.

747
00:43:12,840 --> 00:43:15,440
Speaker 1: It is cool. I mean because just just that level

748
00:43:15,480 --> 00:43:18,319
Speaker 1: of image recognition, I mean being able to take an

749
00:43:18,320 --> 00:43:20,960
Speaker 1: object and look at it from a different angle than

750
00:43:21,000 --> 00:43:25,879
Speaker 1: you were taught or that's a different color, different different size. Yeah,

751
00:43:25,920 --> 00:43:28,279
Speaker 1: all these things. All of these things are easy for us,

752
00:43:28,480 --> 00:43:32,400
Speaker 1: hard for computers. So seeing something make that breakthrough is

753
00:43:32,440 --> 00:43:35,560
Speaker 1: really exciting. Anyway, we thought we would take that story

754
00:43:35,719 --> 00:43:37,560
Speaker 1: and kind of break it down for you guys, explain

755
00:43:38,000 --> 00:43:41,000
Speaker 1: how it's still cool but maybe not as cool as

756
00:43:41,000 --> 00:43:43,440
Speaker 1: the way some of the headlines are saying. Right, and

757
00:43:43,480 --> 00:43:47,759
Speaker 1: also say, hey internet journalists, um, step up your game. Yeah,

758
00:43:47,880 --> 00:43:50,960
Speaker 1: I understand that you want people to read your stuff.

759
00:43:51,000 --> 00:43:53,840
Speaker 1: Oh yeah, and you're under deadline pressure and that's terrible.

760
00:43:53,960 --> 00:43:59,200
Speaker 1: But that's that's hard. Let's reality. We don't just don't

761
00:43:59,200 --> 00:44:02,640
Speaker 1: just spit up us releases the way that you found them. Yeah,

762
00:44:02,680 --> 00:44:05,560
Speaker 1: and uh, and I would be criminal to to neglect

763
00:44:05,719 --> 00:44:11,040
Speaker 1: to mention, knowl Our producer reminded me that obviously this

764
00:44:11,080 --> 00:44:14,680
Speaker 1: is a very important field of study because we want

765
00:44:14,719 --> 00:44:16,920
Speaker 1: to be able to tell the difference between computers and

766
00:44:17,040 --> 00:44:22,279
Speaker 1: humans when the future of Blade Runner becomes our reality

767
00:44:22,400 --> 00:44:24,680
Speaker 1: and you're chasing down a replicant and you have to

768
00:44:24,680 --> 00:44:27,399
Speaker 1: determine if it's actually a replicate or human. Being very

769
00:44:27,440 --> 00:44:31,960
Speaker 1: good point, computer programmers, just make sure you don't explain

770
00:44:32,080 --> 00:44:33,680
Speaker 1: sort of you know, what you do if you see

771
00:44:33,680 --> 00:44:36,319
Speaker 1: in turtle laying on back and you've decided not to

772
00:44:36,360 --> 00:44:40,960
Speaker 1: turn it over, because that's like that's like our gimme, Yeah, yeah,

773
00:44:41,120 --> 00:44:45,319
Speaker 1: we got yeah, we need that. So everything else fair game.

774
00:44:45,520 --> 00:44:48,399
Speaker 1: All right, So that wraps up this discussion. Guys. First

775
00:44:48,440 --> 00:44:51,840
Speaker 1: of all, thank you so much Nick for for asking

776
00:44:52,040 --> 00:44:54,719
Speaker 1: us about that. And if anyone else wants to ask

777
00:44:54,840 --> 00:44:57,840
Speaker 1: us to talk about anything in particular right now, I

778
00:44:57,840 --> 00:45:01,440
Speaker 1: would highly suggest you use Twitter, Facebook, or Tumbler. We

779
00:45:01,480 --> 00:45:04,520
Speaker 1: have to handle tech stuff hs W. We will soon

780
00:45:04,640 --> 00:45:07,359
Speaker 1: have an email address, but coming any day now. Yeah,

781
00:45:07,360 --> 00:45:11,359
Speaker 1: we need we're making this transition from one kind of

782
00:45:11,400 --> 00:45:14,440
Speaker 1: email server to a different one. We're getting new email addresses,

783
00:45:14,880 --> 00:45:17,520
Speaker 1: and as of right now, as we're recording this podcast,

784
00:45:17,560 --> 00:45:20,200
Speaker 1: I think tech Stuff does not yet have one. Our

785
00:45:20,239 --> 00:45:24,439
Speaker 1: future address will be tech stuff at how stuff works

786
00:45:24,480 --> 00:45:28,959
Speaker 1: dot com. So if you're listening to this in try it. Yeah,

787
00:45:28,960 --> 00:45:31,279
Speaker 1: it should be fine. But if you're listening to this

788
00:45:31,920 --> 00:45:34,239
Speaker 1: the day it comes out and you wanted to send

789
00:45:34,280 --> 00:45:36,840
Speaker 1: us a message and it's bouncing back, try Twitter, Facebook

790
00:45:36,920 --> 00:45:39,960
Speaker 1: or Tumbler and we will get your message there. And yes,

791
00:45:40,040 --> 00:45:41,680
Speaker 1: we are working on this. We'll have it up as

792
00:45:41,680 --> 00:45:43,839
Speaker 1: soon as we possibly can, and we'll talk to you

793
00:45:43,880 --> 00:45:49,960
Speaker 1: again really soon for more on this and bathens of

794
00:45:49,960 --> 00:46:02,000
Speaker 1: other topics. Because it how stuff works dot com