Strachey Lecture: Use or Be Used - Regaining Control of AI

1
00:00:12,160 --> 00:00:15,639
I'd like to welcome you to the Trinity Church, Straight G lecture,

2
00:00:15,640 --> 00:00:21,430
the Straight G lectures O the distinguished lecture series from the Department of Computer Science.

3
00:00:21,940 --> 00:00:25,420
We have a hash tag which is in front of me. Hash ox, straight Tilak.

4
00:00:26,050 --> 00:00:29,680
Please do tweet at me. Nice about it. If you are if you are going to tweet.

5
00:00:30,010 --> 00:00:33,220
Before we begin, I would like to acknowledge our sponsors.

6
00:00:33,520 --> 00:00:41,440
Oxford Asset Management now for a number of years have been sponsoring these lectures and it's been a transformative thing for us,

7
00:00:41,440 --> 00:00:48,519
enabling us to bring in speakers and to put on a show that we simply wouldn't have been able to do otherwise.

8
00:00:48,520 --> 00:00:53,280
So we're extremely grateful to Oxford Asset Management for their continuing support.

9
00:00:53,720 --> 00:00:56,560
I I'm sure they would like me to point out that they are hiring.

10
00:00:57,580 --> 00:01:02,140
So if you Google Oxford Asset Management, you will hire, you will find many job opportunities.

11
00:01:02,320 --> 00:01:09,880
But thank you to Oxford Asset Management. So now on to our speakers and I'm delighted to welcome Neil Laurence.

12
00:01:10,570 --> 00:01:18,130
Neil is the inaugural DeepMind Professor of Machine Learning at the University of Cambridge, where he leads the university's flagship mission on A.I.

13
00:01:18,310 --> 00:01:22,690
A.I. At camp. He's been working on machine learning models for 20 years,

14
00:01:23,020 --> 00:01:28,150
and he recently returned to academia after three years of director of Machine Learning Amazon.

15
00:01:28,660 --> 00:01:32,050
His main interest is the interaction of machine learning with the physical world,

16
00:01:32,260 --> 00:01:40,240
and this interest was triggered by deploying machine learning solutions in the African context, where end to end solutions are essential.

17
00:01:40,510 --> 00:01:45,280
And this has inspired new research directions at the interface of machine learning and systems research.

18
00:01:45,760 --> 00:01:50,320
And that work is funded by Senior A.I. Fellowship from the Alan Turing Institute.

19
00:01:50,860 --> 00:01:58,839
He's interim chair of the Advisory Board of the UK Centre for Data Ethics and Innovation and a member of the UK's A.I. Council,

20
00:01:58,840 --> 00:02:05,350
which advises government on all matters. A.I. Neil is a visiting professor at the University of Sheffield,

21
00:02:05,530 --> 00:02:10,090
where he was for a number of years as an academic and is the co-host of Talking Machines.

22
00:02:10,480 --> 00:02:17,889
And the title of Neil's lecture could hardly be more contemporary use or be used.

23
00:02:17,890 --> 00:02:21,610
Regaining Control of Artificial Intelligence. Males, The floor is yours.

24
00:02:29,860 --> 00:02:33,819
Thanks to this chip that's working. Okay. And thank you for that wonderful introduction.

25
00:02:33,820 --> 00:02:39,720
And thank you, Mike, Leslie, everyone, for the invitation and an apology.

26
00:02:39,730 --> 00:02:48,220
This invitation was originally for a year ago, and I caught COVID just before I was due to give the presentation.

27
00:02:48,550 --> 00:02:57,040
The title is actually the same, so imagine how much better the world would have been had I been able to give this lecture a year ago.

28
00:02:57,040 --> 00:03:00,790
Look, another another consequence of the pandemic.

29
00:03:01,720 --> 00:03:05,350
Okay. So I try and use historic.

30
00:03:05,360 --> 00:03:11,260
I'm not going to give a technical thought really today. I'm just going to talk about issues that lead to technical questions.

31
00:03:11,830 --> 00:03:19,989
And when I think about artificial intelligence, I like to sort of use these three apocryphal quotes that will be in the talk today.

32
00:03:19,990 --> 00:03:23,310
So the apocryphal quote about Henry Ford is when?

33
00:03:23,320 --> 00:03:29,050
And then he says that if I'd asked people what they wanted out of a car, they would have asked for faster horse.

34
00:03:30,070 --> 00:03:38,290
And that provokes in my mind the question. So if you ask people what do they want out of artificial intelligence, is the answer a smarter human?

35
00:03:39,820 --> 00:03:43,630
And then the question that follows is, is even what does that mean?

36
00:03:43,750 --> 00:03:47,530
I think, by the way, if we had a faster horse, we'd already have autonomous vehicles.

37
00:03:47,530 --> 00:03:50,500
So that would have been a pretty cool thing. So maybe the customer's always right.

38
00:03:51,970 --> 00:03:59,980
Now, I talk about this a lot and I'm only going to go through it very briefly because I just want to contextualise the way I think about it,

39
00:04:00,550 --> 00:04:08,170
which is when we look at humans in human intelligence, I've been trying to do public understanding, machine learning folks for about ten years,

40
00:04:08,170 --> 00:04:14,920
and I wanted to capture why machine intelligence is so very different from human intelligence and

41
00:04:14,920 --> 00:04:21,190
simply thought at one level it's because I'm speaking to you using sound and machines communicate,

42
00:04:21,310 --> 00:04:24,639
using radio waves or light speed and cables.

43
00:04:24,640 --> 00:04:26,980
So they communicate about a million times faster.

44
00:04:26,980 --> 00:04:34,660
And this leads to all these sort of consequences that for me to talk to you if I'm talking in a good rate of speech,

45
00:04:34,660 --> 00:04:39,370
if we look at how much information I'm sharing, it's about 2000 bits per minute.

46
00:04:39,640 --> 00:04:46,660
So that's a bit as defined by Shannon, an information theory that's the equivalent information as 2000 coin tosses.

47
00:04:46,660 --> 00:04:55,330
So it's about telling you the result of 2000 tennis matches where the odds were even every minute, which is pretty good now.

48
00:04:55,330 --> 00:05:02,770
But to do that, I'm sharing with you the thoughts are inside my head, and it's very hard to estimate how many calculations per seconds we're doing.

49
00:05:02,770 --> 00:05:06,790
But if you look at the best estimates I could find for simulating a brain,

50
00:05:06,790 --> 00:05:12,100
you would need a machine about as fast as the Met Office supercomputer down in Exeter,

51
00:05:12,490 --> 00:05:16,030
which can do about 1,000,000,000 billion calculations per second.

52
00:05:17,380 --> 00:05:21,550
Now, if you compare that to a simple computer, these slides actually date back a few years.

53
00:05:21,550 --> 00:05:30,340
So maybe they're a bit faster than this. Nowadays, a computer can communicate at 60 billion bits per minute if it's using gigabit per second Internet.

54
00:05:30,640 --> 00:05:34,030
So that means if we turn out, that's a nine to salary.

55
00:05:34,030 --> 00:05:39,220
So imagine your university student stipend is $20 a month.

56
00:05:39,670 --> 00:05:42,670
It's not great money, but it's kind of maybe enough to survive on.

57
00:05:43,240 --> 00:05:47,500
And then imagine your university stipend is $60 billion a month.

58
00:05:49,540 --> 00:05:58,810
It's actually closer to the budget of a country, so it's closer to what the UK spends on the whole of health and Social Security $60 billion a month.

59
00:05:59,020 --> 00:06:02,620
So it's operating in terms of this information exchange, much, much faster.

60
00:06:02,770 --> 00:06:08,950
But then our best estimate is that a typical machine in terms of number of calculations, it can do per second.

61
00:06:09,520 --> 00:06:15,339
And I've done discriminant calculations a second distance. You see there are about 100 billion calculations.

62
00:06:15,340 --> 00:06:20,379
So actually a typical machine is doing a lot less compute per second than our brains.

63
00:06:20,380 --> 00:06:25,790
Of course, we can't access all that compute. Neither can the computer access all the compute it does.

64
00:06:25,790 --> 00:06:31,389
So imagine if I wanted to share with you 1 seconds worth of calculations from my head.

65
00:06:31,390 --> 00:06:37,240
If these numbers are right, it would take me a billion years to tell you all the calculations.

66
00:06:37,240 --> 00:06:42,100
5 billion years to tell you all the calculations that had occurred in my head in 1 seconds.

67
00:06:42,520 --> 00:06:50,840
So while my battery is running low, this is an. I'm not sure I was referring to my brain or the machine.

68
00:06:50,960 --> 00:06:58,940
Not me. I can see a blob that's not strong. So I meant to infer human.

69
00:06:59,330 --> 00:07:04,310
Human intelligence. So, 5 billion years.

70
00:07:04,430 --> 00:07:07,700
What is a computer? What did you share? 1 seconds worth of computation.

71
00:07:07,700 --> 00:07:11,180
You would take it about 20 minutes. So this means your intelligence is locked in.

72
00:07:11,630 --> 00:07:14,630
So what I mean by locked in is you can hear these stories.

73
00:07:14,630 --> 00:07:19,790
And sometimes in a longer version of this section, I would talk about a guy called Jean-Dominique Bobic,

74
00:07:19,790 --> 00:07:23,900
who was the French editor of Elle magazine, who was locked in, could only communicate by winking.

75
00:07:24,110 --> 00:07:29,270
They made a film about him. He wrote the book about the use the film after he was locked in.

76
00:07:29,480 --> 00:07:33,800
And I think that's a state we all think about. We all think, oh, wow, what would it be like to be in that state?

77
00:07:33,950 --> 00:07:37,160
Well, that's the state you're in, right? You just evolve that way.

78
00:07:37,250 --> 00:07:40,790
You don't notice. But compared to the machine, that's where you're at.

79
00:07:42,050 --> 00:07:48,110
So in some sense, it's already a nonsense to talk about a more intelligent human being,

80
00:07:48,320 --> 00:07:52,730
because the human being is defined by constraints, not by capabilities.

81
00:07:53,030 --> 00:07:58,640
We're defined in terms of our intelligence, in the way we share information by a limited ability to communicate.

82
00:07:58,850 --> 00:08:02,120
And we've evolved in that state for hundreds of thousands of years,

83
00:08:02,330 --> 00:08:09,200
as Homo sapiens should say that in the Natural History Museum and for, you know, millions of years as primates, etc., etc.

84
00:08:09,200 --> 00:08:14,480
And all animals do this to some. Well, mammals, higher mammals do this to someone great, to all that sort of thing.

85
00:08:14,810 --> 00:08:18,950
So second apocryphal quote, the six word novel never written by Ernest Hemingway,

86
00:08:18,950 --> 00:08:23,870
but apocryphal written by honest Hemingway for sale baby shoes never wore.

87
00:08:24,290 --> 00:08:29,750
Now, so that six words, 7 to 2 bits of information on average, 12 bits per word we have.

88
00:08:30,980 --> 00:08:37,790
But there's so much more meaning in those 12 words than 72 bits of information.

89
00:08:38,090 --> 00:08:43,070
And that's because you exist as humans. You sit within a wider culture, a wider society.

90
00:08:43,370 --> 00:08:48,330
And you can imagine what it would feel to be the human that has to place that up.

91
00:08:50,000 --> 00:09:00,830
So we use this information bandwidth in a way that allows us to lean on the fact that we are all human and we understand what it means to be a human,

92
00:09:01,070 --> 00:09:05,830
to communicate things that feel like there's a lot more than 72 bits of information in there,

93
00:09:05,840 --> 00:09:09,570
because we can all imagine what it felt like to write that advert.

94
00:09:10,370 --> 00:09:14,959
So although the information but I'm sure our context is very important and we

95
00:09:14,960 --> 00:09:20,210
can extract a lot of information from very few provided pieces of information.

96
00:09:20,240 --> 00:09:23,300
So does anyone know who this apocryphal quote is? Apocryphal divine.

97
00:09:24,350 --> 00:09:31,610
There are three types of three families and statistics. It is Mark Twain.

98
00:09:32,330 --> 00:09:36,890
Not apocryphal in Mark Twain, but Mark Twain appropriately credits it to Benjamin Disraeli.

99
00:09:37,370 --> 00:09:42,560
But it was said by Lord Balfour, the foreign secretary tried to do it to someone else at some point.

100
00:09:42,770 --> 00:09:48,889
But the point about this quote is that one of the reasons that we end up misunderstanding and misinterpreting

101
00:09:48,890 --> 00:09:54,650
statistics is because we're so information constrained that when we get little bits of information,

102
00:09:54,650 --> 00:10:00,980
we tend to overinterpret. We tend to see more behind that data than we have any right to.

103
00:10:01,340 --> 00:10:03,530
And that comes out in a bad way in statistics.

104
00:10:03,530 --> 00:10:08,810
It comes out in a good way when you're imagining what it felt for the person that had to write that advert.

105
00:10:09,440 --> 00:10:10,940
So I kind of apocryphal.

106
00:10:11,080 --> 00:10:21,440
Benjamin Disraeli I tend to think one of the the modern equivalent of these quotes is this There are three types of lies, lie families and big data.

107
00:10:21,740 --> 00:10:27,410
Because what's happened in this modern data era and that's being driven by the machine interconnection,

108
00:10:27,680 --> 00:10:31,730
is requiring more and more information, more and more data.

109
00:10:31,940 --> 00:10:37,100
And we're basically forgetting the hard lessons of a field known as mathematical statistics.

110
00:10:37,370 --> 00:10:41,479
So that quote from Disraeli predates the mathematical field of statistics.

111
00:10:41,480 --> 00:10:44,510
So I'm quite close to computer science, government statistics.

112
00:10:44,960 --> 00:10:48,830
They study mathematical statistics. They just don't say mathematical statistics.

113
00:10:48,830 --> 00:10:56,660
They study that the they say the process of understanding whether whether you can believe what the statistics is purporting to say or not.

114
00:10:57,020 --> 00:10:59,389
And we don't even say mathematical statistics anymore.

115
00:10:59,390 --> 00:11:05,840
But it was a new field that had to come in to stop people misinterpreting statistics, which is the sort of science mistake.

116
00:11:06,110 --> 00:11:13,459
So that's Karl Pearson, of course, unfortunately associated with that field is a lot of fairly unpleasant things like eugenics.

117
00:11:13,460 --> 00:11:21,440
Most of the early mathematical statisticians were also involved in eugenics to one form or another, because they believe once you could tell this,

118
00:11:21,440 --> 00:11:27,020
we should be looking to improve the human species in some way, and that led to some quite problematic results.

119
00:11:27,260 --> 00:11:32,900
So that is still an open question about what does mathematical data science look like in our era.

120
00:11:33,380 --> 00:11:40,550
And it's being triggered by this sort of new image because we now have an evolved relationship with information.

121
00:11:41,960 --> 00:11:49,610
So in the past, very, very far past 100,000 years ago, we could go in the museum and check,

122
00:11:49,610 --> 00:11:53,570
but I think we would been in groups of about 30 to 50 operating together,

123
00:11:53,870 --> 00:12:00,739
hunter gatherers, maybe slightly larger, trusting each other, basing sort of inferences about what's going on.

124
00:12:00,740 --> 00:12:07,549
Those relationships of trust and understanding who each of us is. Since then, we became we came into cities.

125
00:12:07,550 --> 00:12:11,030
We've now grown things. We have to introduce laws to govern us.

126
00:12:11,870 --> 00:12:15,980
And we also ended up with statistics to look at how those cities are being managed.

127
00:12:16,160 --> 00:12:20,120
And we've had to teach ourselves not to misinterpret those statistics when making decisions.

128
00:12:20,360 --> 00:12:28,400
We removed ourselves from that direct contact of person to person, and that's led to this evolved relationship with information.

129
00:12:28,430 --> 00:12:36,950
So in the past 100 years, you know, this is the situation that you had humans with this very low bandwidth connection.

130
00:12:37,310 --> 00:12:40,730
And a lot of what we were trying to do was manage that low bandwidth connection

131
00:12:40,730 --> 00:12:44,210
to ensure that we weren't misinterpreting the information we were receiving.

132
00:12:45,140 --> 00:12:49,940
But now what we've got is this We've got an enormously high bandwidth connection between

133
00:12:50,780 --> 00:12:56,150
the machine and data and then a low bandwidth connection between us and the machine.

134
00:12:56,900 --> 00:13:05,900
And this is leading to all sorts of problems. You know, and I we can we can look back over the last ten years in terms of election manipulation,

135
00:13:05,900 --> 00:13:09,290
in terms of social media, etc., etc., and what that's done.

136
00:13:09,290 --> 00:13:14,550
And you can put that down to the fact that the machine is now mediating what information it chooses to show.

137
00:13:14,990 --> 00:13:22,070
I mean, it's actually using classical statistics to work out how to draw us in to our engagement with the machine.

138
00:13:22,280 --> 00:13:27,130
But it's not using classical physics to say, how can I share the truth with this individual?

139
00:13:27,140 --> 00:13:31,610
It's using classical statistics to say, how can I cause this individual to engage with me?

140
00:13:31,850 --> 00:13:39,590
And of course, the answer turns out to be share information with them that, you know, fulfils their prejudices, that type of thing.

141
00:13:40,910 --> 00:13:41,600
She's not great.

142
00:13:42,530 --> 00:13:50,330
So what we have is this as the challenge for data science, that we're trying to deal with this new high bandwidth connection to the machine,

143
00:13:50,450 --> 00:13:57,030
which is then presenting information to us in the low bandwidth way, and we've got classical statistics being the traditional way now.

144
00:13:57,050 --> 00:14:01,250
The first thing I want to emphasise is when you see this connection, you realise, well,

145
00:14:02,060 --> 00:14:06,770
the statistics have a massive role in mediating these new connections and understanding what's going wrong.

146
00:14:06,920 --> 00:14:10,790
But it's not everything. But there is enormous role of computer science as well.

147
00:14:11,450 --> 00:14:16,160
You know, mathematics just in understanding this, this new flow of information.

148
00:14:17,750 --> 00:14:25,280
But then the thing I think we need to also be talking about on top of that is this new wave of movements which are actually,

149
00:14:25,370 --> 00:14:29,420
I would argue, operating almost directly on that link between the human and the machine.

150
00:14:29,770 --> 00:14:33,760
So forming new ways for us to communicate with the machine.

151
00:14:34,420 --> 00:14:40,690
So in the past, what a statistician would do if they wanted to represent data is they would train the human to

152
00:14:40,690 --> 00:14:47,469
understand what a P value is or apparently misunderstand what a P value is or what an error bar is,

153
00:14:47,470 --> 00:14:49,000
or how to interpret a graph.

154
00:14:49,150 --> 00:14:55,990
And then the machine would share, you know, if you were using the machine that information in the form of a graph and provide that information to you.

155
00:14:57,370 --> 00:15:02,800
What we've got now is machines that will just talk to us and when they talk to us,

156
00:15:03,010 --> 00:15:12,520
they are going right to the depths of that desire to overinterpret because that feeding into our desire to anthropomorphise eyes.

157
00:15:13,570 --> 00:15:17,680
And I can't say anthropomorphise very well. So I just say anthrax.

158
00:15:18,190 --> 00:15:28,240
So we all tend to treat complex things as if they are human in some form, whether it's our cat or whether it's our car or whether it's our computer.

159
00:15:28,630 --> 00:15:34,810
And when machines are starting to exhibit language, that tendency is going to increase even more.

160
00:15:35,290 --> 00:15:40,830
But the problem is that we mustn't forget this fundamental difference, that these intelligences are quite different from us.

161
00:15:40,840 --> 00:15:44,980
They are quite alien to ours. They can augment us.

162
00:15:45,400 --> 00:15:51,970
But the point in this tool is they can never replace us and they should never be allowed to replace us,

163
00:15:52,300 --> 00:15:56,020
because at the core of our intelligence is our limitations.

164
00:15:56,410 --> 00:16:03,850
And those are things that right from its birth, however capable the machine has, it will never have those limitations baked into it.

165
00:16:05,780 --> 00:16:10,689
Okay, so when I think about the scale of what we're experiencing now,

166
00:16:10,690 --> 00:16:16,810
I was giving I was convening the UK, AI Fellows, students and postdocs at a hackathon,

167
00:16:17,530 --> 00:16:22,809
partially because I strongly believe that we are so dependent on that generation

168
00:16:22,810 --> 00:16:27,910
of people to deploy and use these technologies in the best way possible,

169
00:16:28,210 --> 00:16:34,860
because it's not going to be the old people like myself who understand the ways in which these things can be deployed and used.

170
00:16:34,880 --> 00:16:41,890
We can see that in every previous revolution it's young people. So I've tried to come up with analogies for how big I think this revolution is.

171
00:16:41,890 --> 00:16:49,780
And sometimes I talk all right about the printing press and the sort of extraordinary effect that had 500 years ago leading to the Enlightenment,

172
00:16:49,780 --> 00:16:53,950
but also sort of 500 years of war around religion and all sorts of things going on in Europe.

173
00:16:54,340 --> 00:16:57,460
But then I kind of caution to the printing press doesn't go far enough.

174
00:16:57,640 --> 00:17:03,850
So the revolution I think we're engaged in at the moment is arguably akin to the revolution of rights.

175
00:17:04,000 --> 00:17:11,800
So this is a sort of early cuneiform tablet that if you go back to well, I started working with a colleague,

176
00:17:12,310 --> 00:17:20,310
geologist in Cambridge who works on Nepal from 1500 B.C. to 1000 B.C. and reads these tablets with,

177
00:17:20,320 --> 00:17:23,860
you know, court cases written down this time that as far as I can make out,

178
00:17:23,860 --> 00:17:28,810
the Greeks were some form of barbarians going around the Mediterranean, attacking cities.

179
00:17:29,650 --> 00:17:38,620
And at the same time, the city of Nepal has 2000 years of written history going back to, you know, before Gilgamesh existed.

180
00:17:38,920 --> 00:17:47,020
So this is a society and a culture that existed some 3000 years ago that had 2000 years of written history,

181
00:17:47,020 --> 00:17:49,480
written in the form of these sort of clay tablets.

182
00:17:49,720 --> 00:17:56,580
But then the thing I sort of think about is what was it like for the first person when when the Epic of Gilgamesh, which was an oral tradition,

183
00:17:56,590 --> 00:18:02,170
like presumably most of these traditions were like The Odyssey and everything else was suddenly written down for the first time,

184
00:18:02,380 --> 00:18:06,640
and the sort of bard who's learned this whole epic and can sing it and do everything else.

185
00:18:06,640 --> 00:18:10,690
He turns up and says, Now this or I may I put it written down, Now don't need you anymore.

186
00:18:11,740 --> 00:18:15,460
We don't need you to remember things because we can remember things.

187
00:18:16,680 --> 00:18:22,230
All a piece of tabloid. And the first form of usage for these tablets was not to write down poems.

188
00:18:22,470 --> 00:18:26,430
It was to write down accounts. So-and-so owes me three sheep.

189
00:18:26,880 --> 00:18:31,770
I owe them four lots of call. And I think there's a tremendous parallel between that and the computer,

190
00:18:32,220 --> 00:18:37,680
which so far has mainly been and somehow a tool of the accountants excel spreadsheets.

191
00:18:37,890 --> 00:18:44,880
It doesn't and can't interface with us in the poetic language we might like to use and what it suddenly gains.

192
00:18:45,270 --> 00:18:47,190
I mean it. Pearson But it's not something.

193
00:18:47,190 --> 00:18:52,440
This has been going on for a number of years, but it's crossed the threshold into, Oh yes, you can talk to the machine.

194
00:18:55,030 --> 00:19:03,009
And I think that this is really one particular reason why this is sending society into these sort of spasms of

195
00:19:03,010 --> 00:19:12,129
excitement is is related to an experience I had in 2017 where we were doing a Royal Society report on machine learning.

196
00:19:12,130 --> 00:19:17,290
So people were thinking about this a long time ago and worrying about how technology would change things.

197
00:19:17,650 --> 00:19:22,410
And we were very focussed at the time as it was on my Richard and Daniel Susskind on the future profession.

198
00:19:22,420 --> 00:19:25,630
It's a very different technological transform, the work of human experts,

199
00:19:25,870 --> 00:19:31,929
and we were doing this work from 2015 to about 2017 and about midway through the work when we were coming up with

200
00:19:31,930 --> 00:19:39,970
a lot of conclusions about how the lives of the middle classes were going to be affected by this revolution,

201
00:19:40,360 --> 00:19:48,459
we had a particular occurrence that really stuck with me and that was the Brexit vote and we were in the wrong society.

202
00:19:48,460 --> 00:19:55,660
About a week later and we were sat around a table and we were all digesting the consequences of this vote.

203
00:19:55,660 --> 00:20:00,910
And there were employees from all parties and there were policy experts and there were domain experts.

204
00:20:01,300 --> 00:20:05,170
And one of the MP said, Well, I bet no one in this room voted for Brexit.

205
00:20:05,800 --> 00:20:08,860
And it suddenly occurred to me, I'll bet your rights.

206
00:20:09,100 --> 00:20:10,300
And isn't that the problem?

207
00:20:11,650 --> 00:20:20,799
Because the people who voted for Brexit are the ones who were disengaged from the decision making, and that's why they voted for Brexit.

208
00:20:20,800 --> 00:20:28,330
And you can see that in populism wherever it exists. And that gave me a great fear because this book speaks to the empowerment.

209
00:20:28,990 --> 00:20:32,590
It speaks to all of us that are already engaged in decision making.

210
00:20:33,580 --> 00:20:41,110
And that's also what you see going on now. Now, I'm not saying the allies aren't going to be transformed as educated people,

211
00:20:41,740 --> 00:20:47,920
but my experience is that educated people are amongst the best equipped to handle such change.

212
00:20:49,480 --> 00:20:53,560
And mean what I have in my head is that of the coin pushing machine.

213
00:20:54,010 --> 00:20:57,910
So this is a coin which I used to love before I as a kid, I don't know if anyone knows what they are.

214
00:20:59,170 --> 00:21:02,910
These are the more fair grounds. You've got to bring the Blackpool or whatever else in you.

215
00:21:03,070 --> 00:21:07,000
You've got these two pennies and they go on the sets and the sets move in and out and you draw the

216
00:21:07,000 --> 00:21:12,960
coin and the coin sort of goes in the sand here and then it lands and it should push these coins off.

217
00:21:12,970 --> 00:21:18,280
You can see these coins are deformed and then when the step pushes out, some coins will fall off and hopefully you'll win.

218
00:21:18,520 --> 00:21:22,900
Now, which coins fall off? Is it the coins that move? No, it's the coins on the edge.

219
00:21:23,680 --> 00:21:29,800
Because those people who are already on the edge of society, who are already challenged by things,

220
00:21:30,070 --> 00:21:33,070
are the ones that are going to be affected by this form of disruption.

221
00:21:33,400 --> 00:21:39,760
And my sense is, yes, those of us in the professions, those of us who are educated can do things well.

222
00:21:39,820 --> 00:21:45,190
We'll have to learn. We'll be disrupted in a major way. But we actually have the intelligence and capability to do your jobs.

223
00:21:45,340 --> 00:21:49,630
It's the knock on effects that it has on those people that have already given up

224
00:21:49,990 --> 00:21:55,629
that I met a guy in Lancaster once who used to who decided not to go to university,

225
00:21:55,630 --> 00:21:59,620
wanted to be in a band. He went catering, called into catering. He worked at the prison.

226
00:21:59,800 --> 00:22:00,700
The president got shot.

227
00:22:00,820 --> 00:22:06,969
And then, of course, I met him because he was driving a taxi, because that's where, if you ask any taxi driver why they're driving a taxi,

228
00:22:06,970 --> 00:22:11,410
you'll always get this sequence of stories about, you know, because anyone can sort of drive a taxi.

229
00:22:11,620 --> 00:22:14,800
But there are lots of people's where there isn't much demand for taxi.

230
00:22:16,300 --> 00:22:23,200
So when we did have a call for the Royal Society back in 2017, it was called machine made the complex computers that lead by example.

231
00:22:23,230 --> 00:22:26,920
I still think it's a really valid report that is actually still referenced today,

232
00:22:27,100 --> 00:22:32,860
but one of the things we did is we spoke to people and we asked them what they would like machine learning solutions to do.

233
00:22:34,000 --> 00:22:38,559
And so we did a it was the Ipsos MORI poll on public views, machine learning.

234
00:22:38,560 --> 00:22:43,060
So this is qualitative research, which I think is going to be increasingly important in this domain.

235
00:22:43,330 --> 00:22:48,400
And I understand in qualitative research, even if you're a quantitative scientist, needs to improve dramatically.

236
00:22:48,730 --> 00:22:53,380
So these are the sort of things they said, and I'm not sure if you can read, but these are sort of areas we looked at.

237
00:22:53,740 --> 00:23:00,640
So we asked about health and health and they can see the greatest potential benefit to individuals and society.

238
00:23:00,970 --> 00:23:02,890
We asked about social care,

239
00:23:03,280 --> 00:23:10,030
so they saw the potential for resourcing issues that they fear to mobilise machines with due to involvement and emotional contact.

240
00:23:10,600 --> 00:23:18,339
Right. So that's clearly a worry and sort of thing. But they could see that there was benefit, the best case scenario, unachievable.

241
00:23:18,340 --> 00:23:21,100
She and they were keeping carers to spend more time with patients.

242
00:23:22,000 --> 00:23:26,530
I mean that would be brilliant if that's what we've done in any of our hospitals or social care.

243
00:23:26,530 --> 00:23:28,209
I don't think it's a very good record.

244
00:23:28,210 --> 00:23:34,660
But marketing maybe even thought marketing was, Oh yeah, we could be used to tailoring marketing and that might not be a bad idea.

245
00:23:34,870 --> 00:23:37,959
Transport driver's cars could have benefits.

246
00:23:37,960 --> 00:23:39,040
These are just regular people.

247
00:23:39,040 --> 00:23:46,240
They were sort of in Birmingham in a sort of like small groups finance universe and thought about our to monitor potentially fraudulent activity.

248
00:23:46,480 --> 00:23:50,200
People are really, really smart about what they want technology to do.

249
00:23:50,620 --> 00:23:54,110
Participants tend to think machine learning and crime the thought patterns in.

250
00:23:54,480 --> 00:23:57,240
Good idea. So let's see how might work accurately in practice.

251
00:23:57,870 --> 00:24:03,179
Smart people, they saw it as a useful tool to aid the military police resources but concerned about consequence to stereotyping.

252
00:24:03,180 --> 00:24:09,030
Individual people are smart. This is why at a time, by the way, when most people had not even heard what machine learning was,

253
00:24:09,030 --> 00:24:12,000
that was the first thing you had to do is educate them what machine learning was.

254
00:24:13,620 --> 00:24:17,939
Public education partisans are concerned that tainted education based on machine learning

255
00:24:17,940 --> 00:24:21,929
would result in killing and limiting people to some career path that too young and again,

256
00:24:21,930 --> 00:24:24,990
really, really smart. But they also sort of see benefits in that.

257
00:24:25,350 --> 00:24:30,239
What is the one area that all participants thought was ridiculous?

258
00:24:30,240 --> 00:24:34,080
That we should not be trying to do that? They could see no point in it whatsoever.

259
00:24:35,250 --> 00:24:41,639
RS They still fail to see the purpose of machine learning and read the poetry for all.

260
00:24:41,640 --> 00:24:42,450
Other case study.

261
00:24:42,450 --> 00:24:48,209
Participants recognise that machine might not do a better job than a human, however, but they do not think this application creates art,

262
00:24:48,210 --> 00:24:53,670
as doing so is considered to be a fundamental human activity that machines could only mimic best.

263
00:24:54,420 --> 00:25:00,600
Where have we made the greatest progress in machine learning in the ensuing six years?

264
00:25:01,260 --> 00:25:03,060
In mimicking what humans do.

265
00:25:03,090 --> 00:25:10,920
So in honour of that, I often get to go right to Silicon Valley That introduces a large language model us spoken by Mickey McHugh.

266
00:25:11,160 --> 00:25:18,150
This was chapter 54. Chapter 3.5 didn't do quite as good a job, but yes,

267
00:25:19,350 --> 00:25:24,720
I don't know who pockets to enable new POC is as if woven by is POC inappropriate person to create?

268
00:25:24,840 --> 00:25:28,320
Largely because of different play, but it's not a fixed.

269
00:25:28,870 --> 00:25:36,969
Right. Well, is the crossover plan. It's the one where they create large language models and there's roaming now.

270
00:25:36,970 --> 00:25:41,110
And you know, it's in nowadays. You know, you've got better call Saul, all sorts going on.

271
00:25:41,110 --> 00:25:45,790
The characters appear in other things. Digital Athenaeum.

272
00:25:46,150 --> 00:25:51,010
My goodness. But what would that fundamentally be?

273
00:25:51,010 --> 00:25:54,760
So I want to say, does anyone recognise what this map is of?

274
00:25:56,920 --> 00:26:00,820
You're not. What's. What's the map of. Western Europe.

275
00:26:01,680 --> 00:26:05,250
Yeah. And can you guess what year it is and what the circumstances of the matter.

276
00:26:06,750 --> 00:26:17,010
Well, go to that site and can you guess the date? There's only a few significant dates in World War Two before, but again, in 44 June, Portugal.

277
00:26:17,010 --> 00:26:21,390
Yeah, June the fifth, 1944. And can you guys see my penis?

278
00:26:23,960 --> 00:26:27,620
As a former teacher, it's a job and not of the weather stations.

279
00:26:28,400 --> 00:26:34,490
Yeah. So it's the pressure and the different motivations. And and actually, you know, I get it from the ice baths.

280
00:26:35,120 --> 00:26:38,780
So what was going on in June 44 was that there was a storm in the channel.

281
00:26:39,080 --> 00:26:44,060
What was also going on was that the tides were at the right level for an invasion of Europe to occur.

282
00:26:44,900 --> 00:26:52,820
What was difficult for the Germans is that the weather comes from the West and that way the stations were on France.

283
00:26:53,240 --> 00:27:02,510
Does anyone know where Rommel was on the day of the invasion at his wife's 50th birthday party in Germany, because he got the weather report and said,

284
00:27:02,510 --> 00:27:10,040
Well, there's no way there'll be an invasion tomorrow so I can go home to my wife's birthday party and everything will be fine.

285
00:27:10,070 --> 00:27:15,380
Of course, he was kind of rudely awoken when I say and had to drive rapidly back to the front.

286
00:27:16,340 --> 00:27:22,159
Now, if we look on the alternative map, this is the allied view of the weather on that day.

287
00:27:22,160 --> 00:27:26,729
And the allies actually on the same day, the front says, and they understand that there's going to be a gap in the weather.

288
00:27:26,730 --> 00:27:30,410
And that's actually multiple weather reports that are in conflict in this even controversy

289
00:27:30,410 --> 00:27:33,950
today over who agreed that there was going to be a significant gap in the weather.

290
00:27:34,430 --> 00:27:40,130
But the interesting thing about weather forecasting in both these cases is it was being done by interpolation.

291
00:27:40,340 --> 00:27:44,150
So you had a number of weather stations spaced across Britain.

292
00:27:44,300 --> 00:27:52,820
I mean, actually it's interesting to me. Martin I'm not sure saying we should have had weather information from France because of Enigma decrypts,

293
00:27:53,060 --> 00:27:54,680
but whether they would put it on the map or not.

294
00:27:54,830 --> 00:27:59,420
So we could actually interpolate across the channel when we did our weather reports, whereas the Germans,

295
00:27:59,780 --> 00:28:03,440
I don't think, could have some weather information on the previous map, I think in Cornwall.

296
00:28:03,470 --> 00:28:06,590
So I'm not sure how they're getting it and what the CORNISH doing.

297
00:28:11,030 --> 00:28:18,560
But basically you're having to interpret. But for Britain, you're interpreting in the right direction because the weather is moving in the right way.

298
00:28:19,190 --> 00:28:25,490
A modern weather forecast is, of course, done using Navier-stokes equation on that big computer done in Exeter,

299
00:28:25,730 --> 00:28:32,389
and they use Navier-stokes equation that allows them to extrapolate. And that extrapolation is sort of key in giving us these long term predictions

300
00:28:32,390 --> 00:28:35,480
up to five or six days because we understand the physics of the weather.

301
00:28:35,480 --> 00:28:42,860
We're not just looking at what the pressure is instead of having a stab at where it might move through some minimal sort of examination over time.

302
00:28:43,130 --> 00:28:46,930
So that's why the Germans got the weather wrong. Now, what's that got to do with TATP?

303
00:28:48,080 --> 00:28:53,410
Well, fundamentally, all it's doing is it's operating. It doesn't have access.

304
00:28:53,420 --> 00:28:59,780
It's in this position. It doesn't have access to the future where humans are given this tool and they

305
00:28:59,780 --> 00:29:04,010
start to work with this tool and they start doing what they will do with that.

306
00:29:04,610 --> 00:29:12,590
Why does that come about? Well, the enormous quantity of data we fed these machines is basically at the primary level.

307
00:29:12,800 --> 00:29:16,160
It's giving a feature space in which interpolation works.

308
00:29:16,490 --> 00:29:21,410
And all these capabilities like logical reasoning stuff are emergent because that

309
00:29:21,410 --> 00:29:25,280
feature space is emergent by just looking at enormous quantities of our data.

310
00:29:25,610 --> 00:29:30,050
But it like in the final layer of this thing, it's still having to interpolate.

311
00:29:30,260 --> 00:29:37,730
It can't build that feature space and extrapolate in such a way that says how well, how would McCue show actually react?

312
00:29:38,150 --> 00:29:41,750
You know, what would they do, first of all, which you could see it could say something,

313
00:29:42,020 --> 00:29:46,490
but we'll only know what McCue would do when the modern McCue shows get given the tool.

314
00:29:46,910 --> 00:29:54,800
And so at this very key point, given that humans, whatever their limitations, are coming out, this problem from particular perspective,

315
00:29:55,430 --> 00:29:59,690
you can never replace them because as soon as you augment them with the tool,

316
00:29:59,870 --> 00:30:03,710
they are something that is far greater than either themselves or the tool on its own.

317
00:30:03,860 --> 00:30:10,009
Just like we became far greater than ourselves by writing things in clay tablets, you know, for a large amount of time.

318
00:30:10,010 --> 00:30:15,290
That was how we shared knowledge and built upon knowledge and understood geometry and all sorts of other things.

319
00:30:15,890 --> 00:30:19,550
So those models are only interpolated now.

320
00:30:20,150 --> 00:30:25,790
This is something I talk about a lot, and I should not question what I would have said about this a little bit.

321
00:30:26,270 --> 00:30:32,060
Because when people talk about artificial intelligence, a question occurred to me is what do they mean?

322
00:30:32,570 --> 00:30:34,640
Because different people mean different things.

323
00:30:35,180 --> 00:30:41,830
I was at the Natural History Museum in London talking with some of the scientists there the other week and on my way,

324
00:30:41,840 --> 00:30:46,900
and I got chatting to the security guard. And I always like to ask people, what do you think it is?

325
00:30:47,480 --> 00:30:54,379
And it's quite hard to understand. A common thing is behind this because intelligence is such an emotive word.

326
00:30:54,380 --> 00:30:57,380
So they sort of think it's like them. It's like, well, it can't be quite like you.

327
00:30:57,860 --> 00:31:05,570
But I do think want to say to you that this technology might and could do is be the first technology that when it automates,

328
00:31:06,230 --> 00:31:10,310
tries to adapt to who we are or I think that's what it purports to do.

329
00:31:10,490 --> 00:31:17,430
Right. So all previous generations of automation, if you invented a weaving in a room or if you invented a spinning journey,

330
00:31:17,690 --> 00:31:21,420
it required everyone to turn up at the factory and service the machine.

331
00:31:21,460 --> 00:31:23,910
So Samuel Butler. It's about this in the 19th century,

332
00:31:24,120 --> 00:31:31,080
the way in which these forms of automation actually enslave us to the machine because we're adaptable and the machine isn't.

333
00:31:31,470 --> 00:31:34,950
So if you look at railways, you know, you all have to turn up on time station.

334
00:31:34,950 --> 00:31:39,810
You didn't even have universal time across the United Kingdom until they had railways.

335
00:31:40,110 --> 00:31:46,050
So every previous generation of automation requires us as flexible entities to adapt to the machine.

336
00:31:46,260 --> 00:31:51,930
And I think what people expect that they're getting from an AI is an entity that will adapt to them.

337
00:31:54,030 --> 00:31:56,759
I think it's a fallacy, but maybe I'm wrong.

338
00:31:56,760 --> 00:32:01,830
And maybe what we're seeing now is when you have machines that can understand the whole of human history and culture,

339
00:32:02,160 --> 00:32:05,490
they can mimic adaptation to a certain it will feel a bit more.

340
00:32:05,580 --> 00:32:10,050
I don't know if we'll truly feel that in two, three years time. I think people are feeling that today.

341
00:32:10,380 --> 00:32:17,700
I can have a conversation with a machine about my software that feels like it is adapting to who I am and what I like.

342
00:32:18,870 --> 00:32:23,060
But when it comes to then applying that in critical decision making,

343
00:32:23,070 --> 00:32:31,170
so a lot of what we care about in society is what one might think of as consequential decisions being made by people about us and who we are.

344
00:32:31,530 --> 00:32:36,870
I think there's kind of a problem with this point of view because the thing I sort of started noticing,

345
00:32:36,870 --> 00:32:40,490
and I think it's also written about I'm not the first to notice it and people use different words,

346
00:32:40,590 --> 00:32:45,090
but everyone in machine learning talks about we want decisions that are fair.

347
00:32:45,900 --> 00:32:53,040
But when they say that, we want fair decisions. I've noticed that people mean a couple of different things.

348
00:32:53,040 --> 00:32:53,399
And in fact,

349
00:32:53,400 --> 00:33:02,280
it was Karl Rasmussen that he went out for a run once because I was talking about the complexity of a fair decision and how difficult it is.

350
00:33:02,640 --> 00:33:07,200
If I'm going to make a nuanced decision about some individuals in front of me from

351
00:33:07,200 --> 00:33:11,160
different backgrounds who all appear qualified to enter the University of Cambridge,

352
00:33:11,340 --> 00:33:18,240
and I'm taking into account all their different backgrounds, what they've done up until now, etc., etc., because they can all manage the degree.

353
00:33:19,080 --> 00:33:26,010
I'm using an immense amount of my understanding of these individuals as a human being and trying to find out about who they are.

354
00:33:26,460 --> 00:33:30,870
And it seems fair that I should do that because I'm taking into account these individuals

355
00:33:30,870 --> 00:33:34,140
experience of life and the struggles they've had to get through to get there.

356
00:33:34,770 --> 00:33:38,150
But it's another form of fairness that says, no, the rules should be clear. Now.

357
00:33:38,160 --> 00:33:43,290
Basically, you should have to get these grades and then you should be accepted into university.

358
00:33:43,860 --> 00:33:50,730
And the same thing applies for any consequential decision making, like giving loans or, you know, court cases.

359
00:33:51,150 --> 00:33:55,050
And this is now reflected in the General Data Protection regulation,

360
00:33:55,740 --> 00:34:03,240
which says that there are certain protected characteristics and you should not ever decide about someone on the basis of those characteristics.

361
00:34:03,810 --> 00:34:07,980
But what I find interesting about this is it's always like this is political alley.

362
00:34:07,980 --> 00:34:12,390
We decide which side, which politic politics is, errs on which side.

363
00:34:12,600 --> 00:34:20,759
To me, both of these extremes are dystopias. The extreme where all you're doing is considering everyone's history and and if only they be given

364
00:34:20,760 --> 00:34:26,309
the chances and giving everyone the same opportunity is one dystopia and the extreme way you say,

365
00:34:26,310 --> 00:34:35,670
no, this is the fixed rule, and only if you pass this rule do you get to have the beneficial outcome or the negative outcome is another dystopia.

366
00:34:36,270 --> 00:34:39,900
And in practice, we're always operating somewhere between.

367
00:34:40,680 --> 00:34:44,159
And it was sort of struggling to think about this because it's really important.

368
00:34:44,160 --> 00:34:50,820
How can you have something that is procedurally fair and simultaneously nuanced and substantively fair?

369
00:34:51,480 --> 00:34:53,280
Because it's something to be perceived.

370
00:34:53,280 --> 00:34:58,980
To be fair, we all have to understand, like, I don't think the tax code is fair because no one understands the tax code.

371
00:34:59,250 --> 00:35:05,250
So if you pay someone a large amount of money, they can find a hole in the tax code and you can pay less tax.

372
00:35:05,460 --> 00:35:08,610
So that's not procedurally fair, even though it's a process. Right.

373
00:35:08,610 --> 00:35:14,520
And it's not procedure fair because it's difficult to understand. So for something to be perceived fair, it has to be clear to everyone.

374
00:35:14,850 --> 00:35:18,300
But simultaneously, we're going to take into account all these nuances.

375
00:35:18,510 --> 00:35:22,440
How can we have something that is taking into account nuances and also be simple?

376
00:35:23,280 --> 00:35:25,979
Well, the answer goes back to that word.

377
00:35:25,980 --> 00:35:32,670
And folks, if you gather a group of people together, if you convene them really well and teach them how to make decisions,

378
00:35:32,970 --> 00:35:39,480
you can have a decision making process that is simultaneously nuanced and procedurally fair,

379
00:35:39,810 --> 00:35:45,720
because you can have people I mean, the most complex things in this room are the other people in this room.

380
00:35:46,170 --> 00:35:49,400
But we all think we have a deep and intuitive understanding of those people.

381
00:35:49,410 --> 00:35:54,930
I mean, when we do action because we co-evolved together, right? So that goes right to the heart of who we are as humans.

382
00:35:55,320 --> 00:36:00,600
So I call this a marvellous resolution between these two extremes that the only way you

383
00:36:00,600 --> 00:36:05,370
can break these two is to have a procedure where you've got well trained human beings.

384
00:36:05,370 --> 00:36:08,010
And of course, very often we're not well trained with bias or whatever.

385
00:36:08,330 --> 00:36:14,360
If you train us humans well about good decision making and then you also have a procedure that sort of says, yeah,

386
00:36:14,370 --> 00:36:18,810
we can how to get these marks, and then you'll be before a committee, and then that committee will make the final decision.

387
00:36:19,020 --> 00:36:22,140
Everyone says, Sure, that sounds fine if you say deal terms of.

388
00:36:22,240 --> 00:36:27,420
Say. And then the machine looks at your entire history and does really complicated calculations

389
00:36:27,420 --> 00:36:32,430
about what's going on that that's very hard for people to understand that sort of procedure,

390
00:36:32,430 --> 00:36:38,370
because it's a procedure, but one that we don't feel an intuition about unless we're assuming that machine is operating like you.

391
00:36:39,030 --> 00:36:42,359
So for me, this says that for consequential decision making,

392
00:36:42,360 --> 00:36:49,470
in order to have decisions that can balance between these two forms of fairness, we must always have humans in the loop.

393
00:36:51,520 --> 00:36:56,530
But what does that mean? Because we know machines can be a great help in these decisions.

394
00:36:58,390 --> 00:36:58,780
Okay.

395
00:36:58,780 --> 00:37:06,250
So I kind of think that the other side to that and I love these lectures, I listen to them recently, these are the best lectures on air quality here.

396
00:37:06,790 --> 00:37:13,750
And from 2002 and then on Al, because what I, Norah O'Neill is talking about is a question of trust.

397
00:37:13,960 --> 00:37:21,610
And she's talking about a world where everything has become more procedural and we're calling everything to account through processes.

398
00:37:21,820 --> 00:37:29,350
And she's talking about the way that erodes our professional faith in individuals who are paid to make good decisions.

399
00:37:29,530 --> 00:37:35,230
Now, there are clearly things that we all know about professional individuals who have made corrupted decisions that are in the papers all the time.

400
00:37:35,650 --> 00:37:42,730
But what Neil is saying is that by increasing accountability in terms of scoring them and marking them and making measures,

401
00:37:42,940 --> 00:37:45,550
you're actually eroding that professional duty.

402
00:37:45,880 --> 00:37:52,560
And what she talks about a lot is that all rights come with duties and that you can't have rights with duties.

403
00:37:52,720 --> 00:37:56,560
And what you will increasingly here, certainly if you hear this, is lapses again,

404
00:37:57,070 --> 00:38:02,680
is is that that's kind of the way that people talk about things like certainly even in academic politics,

405
00:38:03,220 --> 00:38:08,230
I very often don't hear my colleagues talking about what their duties are towards their students,

406
00:38:08,230 --> 00:38:11,590
but I often hear them talking about what their rights are around their intellectual property.

407
00:38:12,730 --> 00:38:16,570
And it would be good if they were focusing a bit more on what their duties are.

408
00:38:16,630 --> 00:38:21,550
And the point O'Neill is making is you can't have rights without duties now.

409
00:38:22,520 --> 00:38:27,870
I mean, she can't have duties because the machine itself does not participate in the society.

410
00:38:27,940 --> 00:38:31,179
It can't be trusted in the same way. And that's the point I Nora O'Neill makes.

411
00:38:31,180 --> 00:38:33,640
And it's a very strong point that sticks with me,

412
00:38:33,880 --> 00:38:38,680
because the first time she made that point was when I was interviewing her for the Royal Society Report,

413
00:38:38,680 --> 00:38:43,410
and I asked her if we could trust the machine. And she told me very specifically, you couldn't.

414
00:38:43,740 --> 00:38:48,900
But listening to her ideas of why that is the case is really about this is the same thing.

415
00:38:48,910 --> 00:38:54,910
You need humans with a human understanding of what professional duty is to be in the loop around this decision making.

416
00:38:55,720 --> 00:38:56,770
So what does that look like?

417
00:38:56,830 --> 00:39:03,130
So they actually she found a quote that relates to this example that I was using before I even commissioned for the elections.

418
00:39:03,340 --> 00:39:07,260
So again, universities are, I think I could say, on the basis of ability and promise,

419
00:39:07,270 --> 00:39:12,830
but they're supposed to also administration will represent the intent. There's no guarantee that the process meets the target.

420
00:39:13,060 --> 00:39:14,320
And I mean, she doesn't say this.

421
00:39:14,320 --> 00:39:21,400
What I'm trying to say is that the only way you can do that is if you have trusted individuals who are balancing those things as part of that process.

422
00:39:21,610 --> 00:39:25,360
And the only way you can trust those individuals is they're well trained and well convened,

423
00:39:25,360 --> 00:39:28,760
etc., etc., and they're held to account by their position in society.

424
00:39:28,780 --> 00:39:32,440
You can't trust the machine to do that because it doesn't have a position in society.

425
00:39:33,850 --> 00:39:38,200
So what does this mean for in terms of what we're doing with artificial intelligence today?

426
00:39:38,290 --> 00:39:44,110
Well, I kind of think as a result, artificial intelligence could only ever be seen as a tool.

427
00:39:44,290 --> 00:39:48,550
And if anyone is telling you it's a replacement for us, I think that's a ridiculous notion.

428
00:39:48,970 --> 00:39:54,460
That always has to be a ridiculous notion. Apart from the following, what does it mean for it to be a tool?

429
00:39:55,030 --> 00:40:01,030
I mean, I that's I want to talk about is the National Advisory Committee of Aeronautics in Langley Field,

430
00:40:02,020 --> 00:40:08,530
who participated in the Second World War and before that in testing aircraft and when they would test the aircraft.

431
00:40:09,370 --> 00:40:15,120
One of the things they did, a guy called Bob Gilruth wrote, is the handbook on how an aircraft should handle.

432
00:40:16,240 --> 00:40:18,639
So what he did and he's a member, the test pilots,

433
00:40:18,640 --> 00:40:24,640
they had what was called spectrum came out of I don't know which one, but I read a lot about seven come out race 47.

434
00:40:25,390 --> 00:40:29,770
When they got a new plan they went up and flew these plans and they moved the

435
00:40:29,770 --> 00:40:33,579
stick around and the plane did things and pilots would say flies like a blue,

436
00:40:33,580 --> 00:40:40,490
fly like a brick, whatever. But they didn't know what that meant in terms of how they should change the plane to fly that, sir.

437
00:40:41,260 --> 00:40:45,850
So what Bob Gilbert did is he wrote a paper that you can go and read it online somewhere

438
00:40:45,850 --> 00:40:50,620
that talks about how a plane should respond when you put £10 of pressure on the stick.

439
00:40:51,160 --> 00:40:55,360
Right. So if you put £10 of pressure on the left and it's a fighter, it should roll at a certain rate.

440
00:40:55,540 --> 00:41:01,809
If you put £10 of stick pressure on the stick and it's a bomber, it should roll at a slower rate As you approach your stall,

441
00:41:01,810 --> 00:41:05,290
the section JUDDER on the run into the stall to warn the pilot.

442
00:41:05,560 --> 00:41:07,630
He quantified all of this. Right.

443
00:41:07,990 --> 00:41:16,090
So when these pilots were flying planes, they could feel the plane and the plane was communicating in a way that was well understood.

444
00:41:16,390 --> 00:41:23,560
Now that control of a stick. Well, the other thing about nationalise the Commission Aeronautics, it became a founding group of method.

445
00:41:24,010 --> 00:41:32,140
And the people that did this work also designed the crewed capsules, the Mercury capsule, the Gemini capsule, and the Apollo capsule.

446
00:41:32,410 --> 00:41:40,690
The actually indulged in providing that feedback to pilots and the control to pilots on entities that were entirely controlled by computers.

447
00:41:40,930 --> 00:41:44,320
So the Apollo guidance computer was the thing that was firing the thrusters.

448
00:41:44,560 --> 00:41:49,800
But Neil Armstrong is moving the stick. So in that case, you've got this really interesting situation where the.

449
00:41:49,820 --> 00:41:55,880
Neil Armstrong is not in control because the computer's in control, but Neil Armstrong feels like he's in control.

450
00:41:57,970 --> 00:42:03,550
And that's the kind of balance we're looking for, because for the things that Neil Armstrong cares about, he is in control.

451
00:42:04,120 --> 00:42:07,270
And the same is true for the pilot of an A380 today.

452
00:42:07,600 --> 00:42:08,860
When they move that stick,

453
00:42:09,070 --> 00:42:14,770
the plane that's going into a computer and the computer is deciding what to do and sometimes the computer overrules the pilot.

454
00:42:15,370 --> 00:42:21,850
It's also true for flight fighter pilots today where fighter pilots would not be capable of flying, obviously, without the computer systems.

455
00:42:22,270 --> 00:42:26,140
So the notion of control you're looking for is something that's akin to this,

456
00:42:26,320 --> 00:42:31,900
where you want the intent of the human being to be understood and projected and enhanced by the machine.

457
00:42:32,290 --> 00:42:36,580
But the machine itself is doing things that are beyond the capability of any given human.

458
00:42:37,060 --> 00:42:38,379
Now, this is incredibly difficult,

459
00:42:38,380 --> 00:42:46,870
and I find this is a really ironic that the control stick was first developed by the Wright Brothers in 1904 or whenever they flew for their flyer,

460
00:42:47,530 --> 00:42:50,740
and it just went forward and backwards because they could only go up and down.

461
00:42:50,950 --> 00:42:54,160
And within 36 years we characterise that interface.

462
00:42:54,640 --> 00:42:57,430
The interface we're now faced with now is one, two.

463
00:42:57,820 --> 00:43:09,040
Let's just talk to our machines and our ability to characterise that interface between the computer and the human is minimal,

464
00:43:09,880 --> 00:43:16,060
like in terms of how we can be manipulated or how we should present uncertain information or how

465
00:43:16,060 --> 00:43:22,360
we share an understanding of a patient with a machine or how it should share it back is minimal.

466
00:43:23,200 --> 00:43:28,659
We don't have the equivalent of £10 left should lead to this and we urgently need to

467
00:43:28,660 --> 00:43:33,070
get on top of that because so far we've been manipulated in fairly simplistic ways.

468
00:43:33,250 --> 00:43:40,809
Like here's a like button or here's an image you'll like or click on this, or here's a question about We'll never believe what so-and-so looked like.

469
00:43:40,810 --> 00:43:44,200
Now, when you knew them in the 1980s, apparently you will take on that.

470
00:43:45,670 --> 00:43:46,630
That's simplistic.

471
00:43:46,640 --> 00:43:53,320
But now there's all sorts of opportunities for machines to interact with this in this highly complex way that we haven't quantified.

472
00:43:53,770 --> 00:44:00,459
So I kind of think that the main thing that we need is the equivalent of a proving ground where when we're looking at these models,

473
00:44:00,460 --> 00:44:04,270
we can understand how they are being deployed in practice.

474
00:44:05,440 --> 00:44:09,040
So understand the nature of the tool, what's the potential, what it pitfalls.

475
00:44:09,370 --> 00:44:12,370
And that's the way we can build cyber capability.

476
00:44:12,880 --> 00:44:18,190
There are going to be loads of disruptions, but there's loads of things we need to do and loads of ways we can be empowered.

477
00:44:18,520 --> 00:44:22,420
So the question is how do we get this understanding as widespread as possible?

478
00:44:23,050 --> 00:44:31,030
Like I said in the introduction, one of the things that has been enormously inspiration to me is working on point data science solutions in Africa.

479
00:44:31,030 --> 00:44:35,200
So Data Science Africa is a bottom up initiative for capacity building.

480
00:44:35,200 --> 00:44:42,010
They've signed machinery and they are on the African continent. And what I love about it is you're always trying to deploy end to end solutions.

481
00:44:42,280 --> 00:44:47,229
So you don't sit in an office and imagine what a farmer might want or a clinician might want.

482
00:44:47,230 --> 00:44:51,430
You actually go out to the health centre or the field and you don't imagine

483
00:44:51,430 --> 00:44:54,579
what the Ministry of Health might want or Minister of Agriculture might want.

484
00:44:54,580 --> 00:45:01,629
You actually go and talk to the Ministry of Agriculture and you find most things that we think about machine learning are utterly irrelevant and the

485
00:45:01,630 --> 00:45:08,800
most of the challenges that people are genuinely experiencing are not being properly researched because they're all at the interfaces with people.

486
00:45:09,730 --> 00:45:13,389
So the thing we need to urgently do is to get on top of working with those people.

487
00:45:13,390 --> 00:45:23,350
So there's an awful mess. In 2015, when we first set up weeks of talks about how we got in and finally get some key facts on the alarm.

488
00:45:23,350 --> 00:45:29,510
So you'll find all these notes and everything. Another aspect to this is how do people maintain power?

489
00:45:29,530 --> 00:45:32,770
How do people maintain control of the system in a societal way?

490
00:45:33,130 --> 00:45:36,670
Well, actually, all this information is coming from us as individuals.

491
00:45:37,060 --> 00:45:40,060
It's our collective eyes data that is being put into these machines.

492
00:45:40,240 --> 00:45:41,980
And we have rights over that data,

493
00:45:42,130 --> 00:45:46,330
whether they are intellectual property rights or whether they're coming from the general Data Protection regulation.

494
00:45:46,720 --> 00:45:49,000
But one challenge is that there's an asymmetry of power.

495
00:45:49,030 --> 00:45:53,470
So if I want to call up Google and say, Can you stop doing this with my data, they're not going to answer my call.

496
00:45:53,860 --> 00:45:57,849
So the notion of a data trust and there's a whole family of mainstream media is now is an

497
00:45:57,850 --> 00:46:03,520
entity which collectivise these people's data rights and operates on this ecosystem to say,

498
00:46:03,520 --> 00:46:09,070
well, basically you can't do that with these models because we want you to delete our data if you're going to do that.

499
00:46:09,430 --> 00:46:17,780
So there's a lot of work that's been going on. We've got some in this area reports 18 months of their trust initiative, 200 experts,

500
00:46:17,830 --> 00:46:23,260
some 70 or so and exciting free real world data transfer funding as pilot Wonga,

501
00:46:23,260 --> 00:46:30,159
which is a sort of local data trust people around bricks in the definition two of which are associated with medical data,

502
00:46:30,160 --> 00:46:36,220
one with people who opted out for the general practitioners recent data sharing scheme

503
00:46:36,310 --> 00:46:40,420
and another one for the Born in Scotland cohort saying which is just starting out.

504
00:46:41,620 --> 00:46:45,790
The other thing I think is vitally important and this is our funding,

505
00:46:45,970 --> 00:46:53,590
we also have to make funding here from Schmidt futures for a project money for four years now is the accelerated programme for scientific discovery,

506
00:46:53,860 --> 00:47:02,530
where the main focus is to go out to data. Science to go out to scientific domain experts and talk to them about their scientific problems

507
00:47:02,680 --> 00:47:06,650
and try and work out what the solutions they need from a machine learning perspective are.

508
00:47:06,880 --> 00:47:11,230
Because only by understanding the problems they face is it's entirely inspired by the

509
00:47:11,410 --> 00:47:16,390
talk to the farmer in the field approach that the DSA has to take the South Africa.

510
00:47:17,020 --> 00:47:22,000
But working with scientists and we go beyond science to engineering and humanities got project,

511
00:47:22,010 --> 00:47:26,680
as I said in my serology friend to try and understand what is it they're looking at?

512
00:47:26,690 --> 00:47:29,110
How how would they like their lives to be better?

513
00:47:29,290 --> 00:47:35,020
Rather than building a neural network and imposing on people, talking to people about the war, the problems you are facing.

514
00:47:35,260 --> 00:47:40,690
And it's exactly inspired by the fact that nurses aren't getting to spend more time with patients through computers.

515
00:47:40,690 --> 00:47:46,330
They're getting to spend more time with computers. And the reason is because that computer solution is imposed from the centre,

516
00:47:46,510 --> 00:47:54,790
rather than by talking to the nurse about what they need and these new capabilities in terms of our ability to code and communicate with machines,

517
00:47:55,060 --> 00:48:02,200
offer an opportunity to turn that around. And this stuff about there won't be any jobs, there's a lot of jobs doing that.

518
00:48:02,380 --> 00:48:09,420
We are utterly failed to deliver on this for like the last eight years, as we've just seen from what people ask for from the Royal Society poll.

519
00:48:09,700 --> 00:48:12,340
Of course, jobs will be different and of course it will be disruptive.

520
00:48:13,450 --> 00:48:20,439
So the final thing is that we're trying to sort of bring that together with the CAM flagship mission.

521
00:48:20,440 --> 00:48:24,759
And if you read and do feel free, we published this report before we got the mission approved,

522
00:48:24,760 --> 00:48:30,790
because the thing I hate about grant proposals is when they've rejected it and you don't get to do anything with your proposal.

523
00:48:30,790 --> 00:48:35,950
So I said, Well, well, we're going to write a report first, publish it, and then you can decide whether to fund it or not.

524
00:48:36,400 --> 00:48:42,910
Part of the reason for that is because, rightly or wrongly, people look to institutions like Cambridge and Oxford to provide leadership.

525
00:48:43,210 --> 00:48:48,430
So this report brought together a number of these strands of thinking communication across the university.

526
00:48:48,470 --> 00:48:52,420
What people expect to see from our and, you know, we hope other people will copy it,

527
00:48:52,420 --> 00:48:57,310
because what it's trying to do is start from what the societal challenges are at MIT,

528
00:48:57,460 --> 00:49:00,970
that Cambridge is not positioned to solve all societal challenges,

529
00:49:01,210 --> 00:49:07,900
but it's position to be a partner in helping people solve those challenges and try and get people to think about how we contribute in that way.

530
00:49:08,080 --> 00:49:12,580
Because it's up to leading institutions like Oxford and Cambridge and Imperial and UCL and

531
00:49:12,580 --> 00:49:17,920
Sheffield and Manchester to really stop all those coins falling off the side of the step.

532
00:49:19,840 --> 00:49:23,829
So Professor following that is just convening and I love it.

533
00:49:23,830 --> 00:49:27,730
To see what's so cool about this is relatively small amounts of funding.

534
00:49:28,150 --> 00:49:32,350
If you're funding the right people, it tends to be young people in an interdisciplinary way.

535
00:49:32,350 --> 00:49:38,860
You can just see beautiful things happen. I think we had this whole meeting where we just funded these groups to the tune of £10,000,

536
00:49:38,860 --> 00:49:42,130
and I was thinking, Yeah, that's about the cost of a business class flight.

537
00:49:42,820 --> 00:49:51,940
And yet some of these groups had produced like workshops, working papers, discussions that were just thinking in ways that only young people can.

538
00:49:51,940 --> 00:49:56,110
The old people who've been stuck in this area for too long are just not seeing the possibilities.

539
00:49:56,260 --> 00:49:59,470
And so it's really vital. We saw funding at that level.

540
00:49:59,920 --> 00:50:07,930
And with that, I'll just say thanks. I don't have any conclusions other than as the enormous amount of work to do, but we're going to do it together.

541
00:50:08,080 --> 00:50:13,960
And if we do it together, well, I think we can lead to a society that is much better for everyone,

542
00:50:14,110 --> 00:50:18,250
not just the professionals, but also all those people on the edge.

543
00:50:18,400 --> 00:50:19,120
Thank you very much.