1
00:00:00,120 --> 00:00:03,040
Speaker 1: Wrapping up our deep dive this week into AI. Bill

2
00:00:03,120 --> 00:00:06,479
Speaker 1: Simpson Young is co founder and CEO of the Gradient Institute,

3
00:00:06,640 --> 00:00:10,440
Speaker 1: a nonprofit focused on ethical AI systems, is joining us

4
00:00:10,760 --> 00:00:15,480
Speaker 1: to chat about where AI is headed and the everyday implications.

5
00:00:15,520 --> 00:00:19,520
Speaker 2: Good morning, Bell, Morning Bell, Hi, Hi Russell, thank you.

6
00:00:19,480 --> 00:00:23,040
Speaker 1: So much for joining us. So how do you see

7
00:00:23,280 --> 00:00:28,040
Speaker 1: AI tools evolving in personal use over the next few years?

8
00:00:28,080 --> 00:00:28,880
Speaker 1: What's coming.

9
00:00:30,440 --> 00:00:33,600
Speaker 3: Well, I mean you've got to think about how they're

10
00:00:33,640 --> 00:00:36,440
Speaker 3: improving continually. These new models coming out every week is

11
00:00:36,479 --> 00:00:39,239
Speaker 3: probably the four major models released from last week or

12
00:00:39,280 --> 00:00:43,159
Speaker 3: with major new capabilities and there of AI is just

13
00:00:43,200 --> 00:00:45,560
Speaker 3: getting more and more powerful. So I mean, I don't

14
00:00:45,560 --> 00:00:47,519
Speaker 3: know if you've tried to like Nano Banana Pro.

15
00:00:48,120 --> 00:00:51,840
Speaker 2: We heard about that yesterday, about it twice and two.

16
00:00:54,280 --> 00:00:58,160
Speaker 3: Like it's phenomenal and you can creating images video is

17
00:00:58,200 --> 00:01:01,480
Speaker 3: really simply and it's not the only one that all

18
00:01:01,480 --> 00:01:03,120
Speaker 3: the models have come out the last couple of weeks.

19
00:01:03,520 --> 00:01:05,600
Speaker 3: You know, it's a new one from Clawed, from One's Anthropic,

20
00:01:05,720 --> 00:01:08,440
Speaker 3: The new version of claud is amazing, the new vision

21
00:01:08,480 --> 00:01:11,560
Speaker 3: Chatty may do it this incredible and relevant to your

22
00:01:11,600 --> 00:01:13,800
Speaker 3: last song. You're playing everybody once through all the world.

23
00:01:14,120 --> 00:01:18,800
Speaker 3: Companies are all racing to racing to be the most intelligent.

24
00:01:18,920 --> 00:01:20,720
Speaker 2: With so many of them, they have to come up

25
00:01:20,760 --> 00:01:23,560
Speaker 2: with better names and nano bananas. So they are doing themselves.

26
00:01:24,000 --> 00:01:25,800
Speaker 1: But is it too many? I mean, did you not

27
00:01:25,840 --> 00:01:29,440
Speaker 1: see the wood for the trees? It's like every week

28
00:01:29,600 --> 00:01:31,960
Speaker 1: for new things, it's yeah.

29
00:01:32,160 --> 00:01:34,840
Speaker 3: So what's happening, as I think one of you earlier,

30
00:01:34,840 --> 00:01:38,360
Speaker 3: I think I think Sarah said in the week the

31
00:01:38,360 --> 00:01:41,120
Speaker 3: the way model the train is on data. And but

32
00:01:41,200 --> 00:01:43,520
Speaker 3: it turns out something was discovered about eight years ago

33
00:01:43,560 --> 00:01:48,360
Speaker 3: where if you use more data and more what's called computing,

34
00:01:48,400 --> 00:01:52,000
Speaker 3: and more cycles of a of a processor like from

35
00:01:52,120 --> 00:01:54,560
Speaker 3: in video, you just the more you do it, the

36
00:01:54,600 --> 00:01:57,080
Speaker 3: smarter they get. You don't have to design the smart

37
00:01:57,160 --> 00:02:00,280
Speaker 3: you just make the models bigger and faster and with

38
00:02:00,360 --> 00:02:03,080
Speaker 3: more data. And so what's happening is these models are

39
00:02:03,080 --> 00:02:07,280
Speaker 3: getting They're pumping in more chips, more GPUs from the video,

40
00:02:07,600 --> 00:02:09,920
Speaker 3: They're putting in more data, and they're just getting smart

41
00:02:09,960 --> 00:02:12,120
Speaker 3: and getting more capabilities. Now the's other stuff happening as well,

42
00:02:12,320 --> 00:02:16,040
Speaker 3: but that change is making making them get more and

43
00:02:16,040 --> 00:02:18,919
Speaker 3: more intelligence. So now you've got models that will pass

44
00:02:19,880 --> 00:02:26,720
Speaker 3: do PhD level biology, chemistry, mathematics solve problems at a

45
00:02:26,840 --> 00:02:30,400
Speaker 3: really sophisticated level across a whole bunch of fields. And

46
00:02:30,560 --> 00:02:32,400
Speaker 3: now these companies are trying to get to know what

47
00:02:32,440 --> 00:02:35,240
Speaker 3: it's called artificial general intelligence, where you've got you know,

48
00:02:35,400 --> 00:02:37,480
Speaker 3: an AI that can do a whole bunch of things

49
00:02:37,480 --> 00:02:40,840
Speaker 3: across the holes off different areas that are greater than

50
00:02:41,360 --> 00:02:44,160
Speaker 3: as good or better than human level capability. And that's

51
00:02:44,960 --> 00:02:48,079
Speaker 3: sort of starting to happen. But any of this, all

52
00:02:48,120 --> 00:02:52,320
Speaker 3: of them are fairly dodgy as well. The problems of

53
00:02:52,680 --> 00:02:55,639
Speaker 3: a thing called spiky capabilities, where you know, any any

54
00:02:55,639 --> 00:02:58,360
Speaker 3: model you have, it'll do something's really amazingly and then

55
00:02:58,440 --> 00:03:02,760
Speaker 3: just something really stupid and cause to realize that my humans, yeah,

56
00:03:02,760 --> 00:03:05,880
Speaker 3: but different. The ways they go wrong are often different

57
00:03:05,880 --> 00:03:07,640
Speaker 3: from the ways that humans go wrong, So you have

58
00:03:07,639 --> 00:03:10,120
Speaker 3: to always understand even if they're smart, they're still pretty

59
00:03:10,120 --> 00:03:11,000
Speaker 3: stupid in some ways.

60
00:03:11,560 --> 00:03:16,320
Speaker 1: Bill governments and big companies preparing for the possible risks.

61
00:03:15,919 --> 00:03:20,280
Speaker 3: Of AI, you actually, it's a great question, and this

62
00:03:20,320 --> 00:03:22,720
Speaker 3: week is a big week in that because on Tuesday,

63
00:03:23,040 --> 00:03:25,320
Speaker 3: I don't know if you've said this, but the Australian

64
00:03:25,320 --> 00:03:27,480
Speaker 3: government announced that they're going to set up an AI

65
00:03:27,520 --> 00:03:31,440
Speaker 3: safety institute. Now, this is a fantastic thing, I think,

66
00:03:31,600 --> 00:03:35,200
Speaker 3: and the UK have done it, the US have don't

67
00:03:35,240 --> 00:03:37,160
Speaker 3: to a lesser extent. There's a bunch of other companies

68
00:03:37,480 --> 00:03:39,400
Speaker 3: Seamplet have done it as well, where you've got sort

69
00:03:39,400 --> 00:03:42,880
Speaker 3: of technical people working on the scientific and technical issues

70
00:03:42,920 --> 00:03:45,120
Speaker 3: with models, where you're looking for how they can go wrong,

71
00:03:45,520 --> 00:03:47,760
Speaker 3: looking at how you can fix them, and I'm actually

72
00:03:47,800 --> 00:03:53,480
Speaker 3: really pleased that's coming. Not yet. Then there's also there's

73
00:03:53,520 --> 00:03:56,560
Speaker 3: various other things that come asuming list, some new guidance

74
00:03:56,600 --> 00:03:58,440
Speaker 3: on how to use AI well and so on, so.

75
00:04:00,080 --> 00:04:02,640
Speaker 2: Of that and going back and watching RoboCop a few

76
00:04:02,680 --> 00:04:05,280
Speaker 2: times just to see where it can go wrong.

77
00:04:06,120 --> 00:04:06,880
Speaker 3: Yeah, exactly.

78
00:04:08,880 --> 00:04:13,720
Speaker 2: So one of the big issues around AI for the

79
00:04:13,760 --> 00:04:17,279
Speaker 2: average person is what is it going to do to jobs,

80
00:04:17,560 --> 00:04:20,839
Speaker 2: to employment? Is there going to be a whole layer

81
00:04:20,880 --> 00:04:24,039
Speaker 2: and they obviously one of the big areas is white

82
00:04:24,040 --> 00:04:27,600
Speaker 2: collar employment. The change that it's going to make that

83
00:04:28,200 --> 00:04:31,239
Speaker 2: in your opinion, your expert opinion, where do you see

84
00:04:31,279 --> 00:04:33,880
Speaker 2: that taking us in the next I don't know, five years,

85
00:04:34,000 --> 00:04:35,039
Speaker 2: probably even less.

86
00:04:36,200 --> 00:04:39,120
Speaker 3: Yeah, I think it will have a massive impact on

87
00:04:39,160 --> 00:04:42,679
Speaker 3: all white college jobs. I think that pretty much every job,

88
00:04:43,440 --> 00:04:48,320
Speaker 3: or most jobs is college jobs will be will become

89
00:04:48,440 --> 00:04:51,680
Speaker 3: AI supervision jobs. So some will go away but many

90
00:04:51,760 --> 00:04:55,120
Speaker 3: will stay. But the responsibilities will change where it be

91
00:04:55,160 --> 00:04:58,080
Speaker 3: tasks that you would have typically previously done, but those

92
00:04:58,120 --> 00:05:00,520
Speaker 3: tasks will be done by AI. But you're job as

93
00:05:00,520 --> 00:05:03,360
Speaker 3: a supervisor like that, any supervisor is just to make

94
00:05:03,400 --> 00:05:06,039
Speaker 3: sure they're being done well. And given that spiky capabilities

95
00:05:06,080 --> 00:05:08,800
Speaker 3: I've talked about, you'll need to know, you know, no,

96
00:05:08,920 --> 00:05:10,880
Speaker 3: the basis of how they work, know how they go wrong,

97
00:05:11,120 --> 00:05:12,839
Speaker 3: so that you can actually pick them up when they're

98
00:05:12,839 --> 00:05:15,560
Speaker 3: doing wrong, and just be a good manager, you know, basically,

99
00:05:16,040 --> 00:05:18,880
Speaker 3: and so a lot of you know, if you're in

100
00:05:18,880 --> 00:05:20,480
Speaker 3: a white college job, you need to make sure you're

101
00:05:20,480 --> 00:05:22,760
Speaker 3: on top of AI. You need to get the experience

102
00:05:22,839 --> 00:05:25,920
Speaker 3: using it, but you also need to learn you know,

103
00:05:26,920 --> 00:05:28,680
Speaker 3: where it's going to get wrong. And there will always

104
00:05:28,680 --> 00:05:31,000
Speaker 3: be problems, right, It's not the case that tomorrow will

105
00:05:31,040 --> 00:05:32,200
Speaker 3: wake up and to be a new model that just

106
00:05:32,240 --> 00:05:34,440
Speaker 3: does everything perfectly. That's not going to happen. It will

107
00:05:34,480 --> 00:05:37,279
Speaker 3: be for the next five years. There will be problems

108
00:05:37,720 --> 00:05:40,200
Speaker 3: and they'll need to be really good human supervision.

109
00:05:40,279 --> 00:05:43,360
Speaker 1: Everyone's worried about the job situation. But will it create

110
00:05:43,480 --> 00:05:44,120
Speaker 1: new jobs?

111
00:05:45,720 --> 00:05:52,160
Speaker 3: Yeah, I think it will, but also will change. There'll

112
00:05:52,200 --> 00:05:56,120
Speaker 3: be new skills that everybody needs, and there will be

113
00:05:56,200 --> 00:05:59,040
Speaker 3: new jobs. I mean one of the big jobs. I

114
00:05:59,240 --> 00:06:02,960
Speaker 3: coming back to your rubberic. You know, there will be

115
00:06:03,040 --> 00:06:06,400
Speaker 3: jobs that are just about There'll be an overallance on AI.

116
00:06:06,480 --> 00:06:08,000
Speaker 3: I think it's a lot of businesses using it, and

117
00:06:08,000 --> 00:06:10,040
Speaker 3: then it's going to go wrong in unexpected ways, and

118
00:06:10,080 --> 00:06:13,200
Speaker 3: there's going to be people out there trying to stop

119
00:06:13,240 --> 00:06:15,440
Speaker 3: the trying to try to stop, to find the problem,

120
00:06:15,440 --> 00:06:19,760
Speaker 3: stop the problems, and try to correct from when problems happen.

121
00:06:19,760 --> 00:06:21,400
Speaker 3: And I think a lot of jobs in that area

122
00:06:21,520 --> 00:06:29,320
Speaker 3: sort of trouble exactly.

123
00:06:29,440 --> 00:06:32,280
Speaker 1: So you're one bit of advice, I imagine, Billy, you've

124
00:06:32,279 --> 00:06:33,800
Speaker 1: already answered this. So I was going to say, what

125
00:06:33,839 --> 00:06:35,600
Speaker 1: is the one thing If you could recommend one thing

126
00:06:35,600 --> 00:06:37,440
Speaker 1: that we should all do to prepare for the future

127
00:06:37,440 --> 00:06:40,280
Speaker 1: of AI, you would say, get your head around to

128
00:06:40,400 --> 00:06:41,040
Speaker 1: learn about it.

129
00:06:41,680 --> 00:06:44,320
Speaker 3: Well, that's one. That's one thing, But another thing is actually,

130
00:06:44,560 --> 00:06:46,600
Speaker 3: you know, I think a lot of jobs in the past,

131
00:06:46,680 --> 00:06:48,200
Speaker 3: we've sort of spent a long time trying to make

132
00:06:48,200 --> 00:06:50,839
Speaker 3: out jobs more mechanical. Yeah, and trying to make humans

133
00:06:50,839 --> 00:06:52,440
Speaker 3: more like machines. I think we need to step back

134
00:06:52,480 --> 00:06:54,680
Speaker 3: and say, look, actually humans are not machines. We've got

135
00:06:54,720 --> 00:06:57,440
Speaker 3: machines to be machines. Let's get the humans really good

136
00:06:57,440 --> 00:06:59,839
Speaker 3: at being human. Let's be really good at working with

137
00:06:59,839 --> 00:07:02,920
Speaker 3: other people. Let's understand what people are good at, understand

138
00:07:02,920 --> 00:07:05,599
Speaker 3: how human values and just and then sit back. And

139
00:07:05,640 --> 00:07:08,120
Speaker 3: I mean not sit back, but make sure the machines

140
00:07:08,120 --> 00:07:10,640
Speaker 3: are doing what we want them to do and not

141
00:07:10,760 --> 00:07:13,240
Speaker 3: something else. So yeah, I think two things can be

142
00:07:13,320 --> 00:07:16,480
Speaker 3: about a human and and learn about AI want to

143
00:07:16,560 --> 00:07:16,840
Speaker 3: use it.

144
00:07:17,000 --> 00:07:20,600
Speaker 1: Yeah, so that we remain in charge exactly.

145
00:07:21,160 --> 00:07:24,280
Speaker 2: We need to know whose boss.

146
00:07:24,480 --> 00:07:27,600
Speaker 1: Yeah, Bill, thank you so much for joining us this morning.

147
00:07:28,520 --> 00:07:29,000
Speaker 3: No worries.

148
00:07:29,560 --> 00:07:32,360
Speaker 2: He's laid out some of our fews, well a little bit.

149
00:07:32,440 --> 00:07:35,040
Speaker 2: He's put it in perspective. He's put it in perspective

150
00:07:37,080 --> 00:07:37,880
Speaker 2: apart from the road