WEBVTT FILE

1
00:00:08.520 --> 00:00:10.740
Hello. This is 6 Minute English from

2
00:00:10.740 --> 00:00:12.420
BBC Learning English. I’m Sam.

3
00:00:12.420 --> 00:00:13.260
And I’m Neil.

4
00:00:13.260 --> 00:00:15.780
In the autumn of 2021, something

5
00:00:15.780 --> 00:00:17.700
strange happened at the Google

6
00:00:17.700 --> 00:00:20.160
headquarters in California’s Silicon

7
00:00:20.160 --> 00:00:22.260
Valley. A software engineer called,

8
00:00:22.260 --> 00:00:24.720
Blake Lemoine, was working on the

9
00:00:24.720 --> 00:00:27.060
artificial intelligence project, ‘Language

10
00:00:27.060 --> 00:00:29.760
Models for Dialogue Applications’, or

11
00:00:29.760 --> 00:00:32.700
LaMDA for short. LaMDA is a

12
00:00:32.700 --> 00:00:34.560
chatbot – a computer programme

13
00:00:34.560 --> 00:00:36.600
designed to have conversations with

14
00:00:36.600 --> 00:00:37.680
humans over the internet.

15
00:00:37.680 --> 00:00:40.500
After months talking with LaMDA

16
00:00:40.500 --> 00:00:42.779
on topics ranging from movies to

17
00:00:42.780 --> 00:00:45.000
the meaning of life, Blake came to

18
00:00:45.000 --> 00:00:47.700
a surprising conclusion: the chatbot

19
00:00:47.700 --> 00:00:50.340
was an intelligent person with wishes

20
00:00:50.340 --> 00:00:52.500
and rights that should be respected.

21
00:00:52.500 --> 00:00:55.140
For Blake, LaMDA was a Google

22
00:00:55.140 --> 00:00:56.519
employee, not a machine.

23
00:00:56.520 --> 00:00:58.680
He also called it his ‘friend’.

24
00:00:58.680 --> 00:01:01.260
Google quickly reassigned Blake from

25
00:01:01.260 --> 00:01:03.120
the project, announcing that his ideas

26
00:01:03.120 --> 00:01:05.340
were not supported by the evidence.

27
00:01:05.340 --> 00:01:08.040
But what exactly was going on?

28
00:01:08.040 --> 00:01:09.900
In this programme, we’ll be

29
00:01:09.900 --> 00:01:12.060
discussing whether artificial intelligence

30
00:01:12.060 --> 00:01:15.060
is capable of consciousness. We’ll hear

31
00:01:15.060 --> 00:01:18.060
from one expert who thinks AI is not as

32
00:01:18.060 --> 00:01:19.620
intelligent as we sometimes think,

33
00:01:19.620 --> 00:01:22.080
and as usual, we’ll be learning some

34
00:01:22.080 --> 00:01:23.400
new vocabulary as well.

35
00:01:23.400 --> 00:01:25.500
But before that, I have a question for

36
00:01:25.500 --> 00:01:27.779
you, Neil. What happened to Blake Lemoine

37
00:01:27.780 --> 00:01:30.720
is strangely similar to the 2013 Hollywood

38
00:01:30.720 --> 00:01:34.140
movie, Her, starring Joaquin Phoenix as

39
00:01:34.140 --> 00:01:36.000
a lonely writer who talks with his

40
00:01:36.000 --> 00:01:38.279
computer, voiced by Scarlett Johansson.

41
00:01:38.280 --> 00:01:40.140
But what happens at the end

42
00:01:40.140 --> 00:01:41.519
of the movie? Is it:

43
00:01:41.520 --> 00:01:43.380
a) the computer comes to life?

44
00:01:43.380 --> 00:01:46.740
b) the computer dreams about the writer? or,

45
00:01:46.740 --> 00:01:48.780
c) the writer falls in love with the computer?

46
00:01:48.780 --> 00:01:52.200
... c) the writer falls in love with the computer.

47
00:01:52.200 --> 00:01:54.900
OK, Neil, I’ll reveal the answer at the end

48
00:01:54.900 --> 00:01:57.360
of the programme. Although Hollywood is

49
00:01:57.360 --> 00:01:59.700
full of movies about robots coming to life,

50
00:01:59.700 --> 00:02:02.940
Emily Bender, a professor of linguistics and

51
00:02:02.940 --> 00:02:05.040
computing at the University of Washington,

52
00:02:05.040 --> 00:02:09.480
thinks AI isn’t that smart. She thinks the

53
00:02:09.480 --> 00:02:11.280
words we use to talk about technology,

54
00:02:11.280 --> 00:02:14.580
phrases like ‘machine learning’, give a

55
00:02:14.580 --> 00:02:16.200
false impression about what

56
00:02:16.200 --> 00:02:17.940
computers can and can’t do.

57
00:02:17.940 --> 00:02:20.340
Here is Professor Bender discussing

58
00:02:20.340 --> 00:02:22.680
another misleading phrase, ‘speech

59
00:02:22.680 --> 00:02:25.139
recognition’, with BBC World Service

60
00:02:25.139 --> 00:02:26.579
programme, The Inquiry:

61
00:02:27.360 --> 00:02:29.580
If you talk about ‘automatic speech

62
00:02:29.580 --> 00:02:31.500
recognition’, the term ‘recognition’

63
00:02:31.500 --> 00:02:33.600
suggests that there's something

64
00:02:33.600 --> 00:02:35.760
cognitive going on, where I think a

65
00:02:35.760 --> 00:02:37.860
better term would be automatic transcription.

66
00:02:37.860 --> 00:02:39.840
That just describes the input-output

67
00:02:39.840 --> 00:02:43.500
relation, and not any theory or wishful

68
00:02:43.500 --> 00:02:45.779
thinking about what the computer is

69
00:02:45.780 --> 00:02:47.220
doing to be able to achieve that.

70
00:02:47.220 --> 00:02:50.580
Using words like ‘recognition’ in relation

71
00:02:50.580 --> 00:02:52.680
to computers gives the idea that

72
00:02:52.680 --> 00:02:55.439
something cognitive is happening – something

73
00:02:55.440 --> 00:02:57.780
related to the mental processes of

74
00:02:57.780 --> 00:03:00.539
thinking, knowing, learning and understanding.

75
00:03:00.540 --> 00:03:03.240
But thinking and knowing are human,

76
00:03:03.240 --> 00:03:06.300
not machine, activities. Professor Benders

77
00:03:06.300 --> 00:03:08.340
says that talking about them in connection

78
00:03:08.340 --> 00:03:11.910
with computers is wishful thinking -

79
00:03:11.910 --> 00:03:14.220
something which is unlikely to happen.

80
00:03:14.220 --> 00:03:16.440
The problem with using words in this

81
00:03:16.440 --> 00:03:18.540
way is that it reinforces what

82
00:03:18.540 --> 00:03:20.700
Professor Bender calls, technical

83
00:03:20.700 --> 00:03:23.220
bias – the assumption that the computer

84
00:03:23.220 --> 00:03:25.740
is always right. When we encounter

85
00:03:25.740 --> 00:03:27.419
language that sounds natural, but is

86
00:03:27.419 --> 00:03:29.699
coming from a computer, humans

87
00:03:29.700 --> 00:03:32.160
can’t help but imagine a mind behind

88
00:03:32.160 --> 00:03:34.380
the language, even when there isn’t one.

89
00:03:34.380 --> 00:03:36.540
In other words, we anthropomorphise

90
00:03:36.540 --> 00:03:39.060
computers – we treat them as if they

91
00:03:39.060 --> 00:03:41.280
were human. Here’s Professor Bender

92
00:03:41.280 --> 00:03:43.500
again, discussing this idea with

93
00:03:43.500 --> 00:03:46.200
Charmaine Cozier, presenter of BBC

94
00:03:46.200 --> 00:03:47.820
World Service’s, the Inquiry.

95
00:03:48.420 --> 00:03:52.080
So ‘ism’ means system, ‘anthro’ or ‘anthropo’

96
00:03:52.080 --> 00:03:55.080
means human, and ‘morph’ means shape...

97
00:03:55.080 --> 00:03:58.080
And so this is a system that puts the

98
00:03:58.080 --> 00:04:00.240
shape of a human on something, and

99
00:04:00.240 --> 00:04:01.500
in this case the something is a computer.

100
00:04:01.500 --> 00:04:04.680
We anthropomorphise animals all the time,

101
00:04:04.680 --> 00:04:07.740
but we also anthropomorphise action figures,

102
00:04:07.740 --> 00:04:10.620
or dolls, or companies when we talk about

103
00:04:10.620 --> 00:04:12.660
companies having intentions and so on.

104
00:04:12.660 --> 00:04:15.540
We very much are in the habit of seeing

105
00:04:15.540 --> 00:04:17.160
ourselves in the world around us.

106
00:04:17.160 --> 00:04:19.380
And while we’re busy seeing ourselves

107
00:04:19.380 --> 00:04:21.480
by assigning human traits to things that

108
00:04:21.480 --> 00:04:24.000
are not, we risk being blindsided.

109
00:04:24.000 --> 00:04:26.400
The more fluent that text is, the more

110
00:04:26.400 --> 00:04:29.100
different topics it can converse on, the

111
00:04:29.100 --> 00:04:31.020
more chances there are to get taken in.

112
00:04:31.860 --> 00:04:34.619
If we treat computers as if they could think,

113
00:04:34.620 --> 00:04:37.140
we might get blindsided, or

114
00:04:37.140 --> 00:04:40.320
unpleasantly surprised. Artificial intelligence

115
00:04:40.320 --> 00:04:42.540
works by finding patterns in massive

116
00:04:42.540 --> 00:04:45.060
amounts of data, so it can seem like

117
00:04:45.060 --> 00:04:46.920
we’re talking with a human, instead

118
00:04:46.920 --> 00:04:49.140
of a machine doing data analysis.

119
00:04:49.140 --> 00:04:53.159
As a result, we get taken in – we’re tricked

120
00:04:53.160 --> 00:04:55.080
or deceived into thinking we’re dealing

121
00:04:55.080 --> 00:04:58.083
with a human, or with something intelligent.

122
00:04:58.083 --> 00:05:01.401
Powerful AI can make machines appear conscious,

123
00:05:01.401 --> 00:05:04.346
but even tech giants like Google are years

124
00:05:04.346 --> 00:05:06.753
away from building computers that can

125
00:05:06.753 --> 00:05:09.108
dream or fall in love. Speaking of which,

126
00:05:09.108 --> 00:05:11.779
Sam, what was the answer to your question?

127
00:05:11.779 --> 00:05:14.530
I asked what happened in the 2013 movie, Her.

128
00:05:14.531 --> 00:05:16.431
Neil thought that the main character

129
00:05:16.431 --> 00:05:18.678
falls in love with his computer, which

130
00:05:18.678 --> 00:05:20.039
was the correct answer!

131
00:05:20.880 --> 00:05:23.520
OK. Right, it’s time to recap the vocabulary

132
00:05:23.520 --> 00:05:25.860
we’ve learned from this programme about AI,

133
00:05:25.860 --> 00:05:29.040
including chatbots - computer programmes

134
00:05:29.040 --> 00:05:30.419
designed to interact with

135
00:05:30.420 --> 00:05:31.980
humans over the internet.

136
00:05:31.980 --> 00:05:34.620
The adjective cognitive describes

137
00:05:34.620 --> 00:05:36.480
anything connected with the mental

138
00:05:36.480 --> 00:05:37.800
processes of knowing,

139
00:05:37.800 --> 00:05:39.179
learning and understanding.

140
00:05:39.180 --> 00:05:41.820
Wishful thinking means thinking that

141
00:05:41.820 --> 00:05:43.920
something which is very unlikely to happen

142
00:05:43.920 --> 00:05:45.960
might happen one day in the future.

143
00:05:45.960 --> 00:05:48.360
To anthropomorphise an object means

144
00:05:48.360 --> 00:05:49.860
to treat it as if it were human,

145
00:05:49.860 --> 00:05:51.060
even though it’s not.

146
00:05:51.060 --> 00:05:53.580
When you’re blindsided, you’re

147
00:05:53.580 --> 00:05:55.080
surprised in a negative way.

148
00:05:55.080 --> 00:05:57.960
And finally, to get taken in by someone means

149
00:05:57.960 --> 00:05:59.820
to be deceived or tricked by them.

150
00:05:59.820 --> 00:06:02.460
My computer tells me that our six minutes

151
00:06:02.460 --> 00:06:05.040
are up! Join us again soon, for now

152
00:06:05.040 --> 00:06:06.179
it’s goodbye from us.

153
00:06:06.180 --> 00:06:06.680
Bye!