1
00:01:14,920 --> 00:01:20,530
Hello, everyone, and welcome to this launch of the Institute for Ethics and I in Oxford.

2
00:01:20,530 --> 00:01:22,570
Thank you for joining us.

3
00:01:22,570 --> 00:01:31,600
This institute is made possible by the historic philanthropic gift from Stephen Schwarzman to create a new centre for the humanities at Oxford.

4
00:01:31,600 --> 00:01:34,810
The centre will bring together seven humanities faculties.

5
00:01:34,810 --> 00:01:42,070
It will create a new state of the art performing arts centre and a new centre for Ethics and I.

6
00:01:42,070 --> 00:01:51,580
The goal is to harness insights from the humanities gleaned from a millennium of scholarship and to apply them to the technologies of the future.

7
00:01:51,580 --> 00:01:54,440
Such as I. To do this,

8
00:01:54,440 --> 00:02:03,670
we will bring together philosophers and other humanists with technical developers and users of A.I. and academia and business and government.

9
00:02:03,670 --> 00:02:08,710
The ethics and governance of the is an exceptionally vibrant area of research at Oxford,

10
00:02:08,710 --> 00:02:19,150
and the institute will build on this work as it explores how A.I. is changing, how we work, how we live, how our democracy functions.

11
00:02:19,150 --> 00:02:24,820
The institute has gotten off to a very strong start by assembling a wonderful group of academics with

12
00:02:24,820 --> 00:02:32,040
more appointments to come and by organising an important series of stimulating events such as this.

13
00:02:32,040 --> 00:02:36,960
This evening to help us explore the questions of A.I. and democratic culture,

14
00:02:36,960 --> 00:02:43,540
we've convened a fabulous panel and I'm delighted to have the chance to introduce them to you.

15
00:02:43,540 --> 00:02:51,370
Professor Sir Nigel Shadbolt is principal of Jesus College and Professorial Research Fellow in the Department of Computer Science.

16
00:02:51,370 --> 00:03:00,370
He's chairman and co-founder of the Open Data Institute and a fellow of the Royal Academy of Engineering and the British Computer Society.

17
00:03:00,370 --> 00:03:03,520
Niger has been unfailingly generous with his time,

18
00:03:03,520 --> 00:03:10,240
wisdom and commitment as we've sought to define the institution in such a way as to mobilise work and

19
00:03:10,240 --> 00:03:18,050
related fields across the university while optimising its impact on the world beyond the university.

20
00:03:18,050 --> 00:03:23,160
Professor Alan Landau is associate professor of political science at Yale.

21
00:03:23,160 --> 00:03:30,750
Her research and teaching interests include democratic theory, political epistemology, theories of justice,

22
00:03:30,750 --> 00:03:37,680
the philosophy of social sciences, constitutional processes and theories and workplace democracy.

23
00:03:37,680 --> 00:03:43,200
Allen is the author of the recent book Open Democracy, published by Princeton A Vision for a new,

24
00:03:43,200 --> 00:03:49,730
more open form of democracy based on non electoral forms of representation.

25
00:03:49,730 --> 00:03:55,190
Professor Joshua Cohen is a political philosopher and is on the faculty of Apple University.

26
00:03:55,190 --> 00:04:00,960
He's also a distinguished senior fellow in law, philosophy and political science at Berkeley.

27
00:04:00,960 --> 00:04:06,000
He has written on issues of democratic theory, freedom of expression, religious freedom,

28
00:04:06,000 --> 00:04:12,180
political equality, democracy and digital technology, good jobs and global justice.

29
00:04:12,180 --> 00:04:18,540
His books include on democracy, democracy and Association Philosophy, Politics, Democracy,

30
00:04:18,540 --> 00:04:25,640
Rousseau, a Free Community of Equals, and the Arc of the Moral Universe and other essays.

31
00:04:25,640 --> 00:04:30,530
And finally, I'd like to introduce the chair of today's panel, Professor John Sourceless,

32
00:04:30,530 --> 00:04:38,600
who is the inaugural director of the Institute for Ethics, and I am professor of ethics and Legal Philosophy here in Oxford.

33
00:04:38,600 --> 00:04:45,010
Professor Thewlis, until recently held the chair of Politics, Philosophy and Law in King's College, London.

34
00:04:45,010 --> 00:04:51,040
He has degrees in law and philosophy from the University of Melbourne and a different philosophy from the University of Oxford,

35
00:04:51,040 --> 00:04:57,520
where he studied as a Rhodes Scholar, previously taught in Glasgow and here in Oxford,

36
00:04:57,520 --> 00:05:03,520
and has also served as a consultant on human rights for the World Bank. Since coming back to Oxford last year,

37
00:05:03,520 --> 00:05:11,920
he has brought insight and energy to the role of director of the Institute for Ethics and A.I. and we're delighted to welcome him back to Oxford.

38
00:05:11,920 --> 00:05:17,580
Over to you, John. Many thanks for that very kind introduction, Louise.

39
00:05:17,580 --> 00:05:25,890
And before beginning discussion, I want to say a few words about the mission of Oxford's Institute of Ethics for Ethics.

40
00:05:25,890 --> 00:05:31,650
In a good place to begin is the word ethics itself for many people.

41
00:05:31,650 --> 00:05:35,610
Ethics is just one form of regulation, amongst others.

42
00:05:35,610 --> 00:05:45,360
Many perhaps big tech companies, amongst them regarded as a soft form of regulation to be distinguished from law, which is enforceable.

43
00:05:45,360 --> 00:05:53,940
Indeed, on a cynical view, ethics is a matter of facile matters like Don't Be Evil, which can be hard to distinguish from corporate PR.

44
00:05:53,940 --> 00:05:56,790
But that is not at all how we understand ethics.

45
00:05:56,790 --> 00:06:06,660
In the Institute for US, Ethics refers to the ultimate values that human beings should respect and pursuit values like justice,

46
00:06:06,660 --> 00:06:16,600
equality and the common good. And two of the reasons for action generated by those values in particular times and places.

47
00:06:16,600 --> 00:06:21,670
On this expansive view of ethics, it is not one form of regulation, amongst others.

48
00:06:21,670 --> 00:06:31,620
Instead, ethics is the foundation of all forms of regulation, whether political, economic, legal or any other kind.

49
00:06:31,620 --> 00:06:37,620
It is ethical judgement, for example, that determines whether a given kind of activity,

50
00:06:37,620 --> 00:06:47,130
such as the use of self driving cars or online microtargeting is an appropriate subject matter for legal regulation.

51
00:06:47,130 --> 00:06:52,620
And if it is, what form those laws should take.

52
00:06:52,620 --> 00:06:59,000
It is because we understand ethics in this way. It is because we understand ethics in this way,

53
00:06:59,000 --> 00:07:04,730
that the institute gives a central role to the discipline of philosophy and why the

54
00:07:04,730 --> 00:07:11,270
institute enjoys the great privilege of being based in Oxford's faculty of philosophy.

55
00:07:11,270 --> 00:07:15,620
Philosophy is the academic discipline with a centuries long cross-cultural track

56
00:07:15,620 --> 00:07:21,360
record of seeking to address questions of ultimate value through the use of reason.

57
00:07:21,360 --> 00:07:26,210
One of the most distinctive features of the institute here at Oxford is that it aims to bring

58
00:07:26,210 --> 00:07:34,730
the clarity and rigour characteristic of the best philosophy to this nascent field of ethics.

59
00:07:34,730 --> 00:07:42,560
But philosophy by itself is not enough. The ethics of A.I. is inescapably an interdisciplinary endeavour.

60
00:07:42,560 --> 00:07:50,840
Most obviously, it requires dialogue with computer scientists who are at the forefront in the development of technology.

61
00:07:50,840 --> 00:07:55,640
And it has been immensely gratifying that the institute has already established

62
00:07:55,640 --> 00:08:00,110
close ties with brilliant colleagues in Oxford's Department of Computer Science,

63
00:08:00,110 --> 00:08:07,600
many of whom have themselves reflected long and hard about the ethical challenges thrown up by their research.

64
00:08:07,600 --> 00:08:11,560
However, there's another reason why philosophy by itself is not enough,

65
00:08:11,560 --> 00:08:18,220
a truly rich and informative ethics debate has to recognise that it's ethical questions assume

66
00:08:18,220 --> 00:08:24,820
a significantly different shape depending on the domain of human life that is in question.

67
00:08:24,820 --> 00:08:31,750
The use of robots and A.I. to carry out surgery or to make medical diagnoses, for example,

68
00:08:31,750 --> 00:08:40,620
raises very different ethical issues from their use as judges, law enforcers or lethal weapons and warfare.

69
00:08:40,620 --> 00:08:49,140
This is one reason why it is significant that the Institute for Ethics and I will be housed in the Schwartzmann Centre for the Humanities,

70
00:08:49,140 --> 00:08:56,910
drawing on expertise from a range of other humanities disciplines, from classics to music and also wine.

71
00:08:56,910 --> 00:09:00,990
We're establishing close connexions with other parts of the university,

72
00:09:00,990 --> 00:09:09,960
such as the Oxford Internet Institute and the Blowback School of Government, with whom we will soon be advertising joint posts.

73
00:09:09,960 --> 00:09:16,200
One final aspect of the institute's mission I want to stress is public engagement.

74
00:09:16,200 --> 00:09:21,540
This doesn't simply mean that my colleagues and I are interested in producing work that is accessible

75
00:09:21,540 --> 00:09:29,260
to a wider public beyond academia and which can inform policy decisions and enhance democratic debate,

76
00:09:29,260 --> 00:09:38,850
although it does mean that, too. It also means that we hope that the institute will operate as an inclusive forum in which a broader public

77
00:09:38,850 --> 00:09:46,370
in this country and globally can engage in discussion about the challenges and opportunities of I.

78
00:09:46,370 --> 00:09:49,160
In his work and political philosophy, The Politics,

79
00:09:49,160 --> 00:09:56,300
Aristotle wrote that the person who lives in a house often knows more about it than the architect who designed it.

80
00:09:56,300 --> 00:10:01,160
All of us are increasingly living in a society shaped by.

81
00:10:01,160 --> 00:10:06,740
So it is vital that the voices of those who are affected by applications such

82
00:10:06,740 --> 00:10:11,780
as contact tracing and facial recognition are heard not just those who create,

83
00:10:11,780 --> 00:10:15,530
deploy or profit from those applications.

84
00:10:15,530 --> 00:10:22,640
And this is what brings me to the topic of our discussion tonight, philosophers can play an important role in ethics,

85
00:10:22,640 --> 00:10:28,760
but they cannot resolve these conundrums by themselves, not because they are not wise enough,

86
00:10:28,760 --> 00:10:33,320
but because they have no political authority to do so in a democracy.

87
00:10:33,320 --> 00:10:40,680
That authority ultimately rests in the citizenry. And that is why we thought that is a topic for a launch event.

88
00:10:40,680 --> 00:10:44,780
We could not do better than to explore the interplay of democracy.

89
00:10:44,780 --> 00:10:51,320
And democracy is not just one of the demands of activity in which operates.

90
00:10:51,320 --> 00:11:01,130
It is also the overarching system of governance that is ultimately tasked with addressing its challenges and its opportunities.

91
00:11:01,130 --> 00:11:06,530
So it's my pleasure now to invite our three panellists to join me.

92
00:11:06,530 --> 00:11:15,410
And as they do so, let me remind you that there will be a Q&A session towards the very end of the last half hour of this event.

93
00:11:15,410 --> 00:11:22,700
So please do use the chat function in YouTube to ask your questions.

94
00:11:22,700 --> 00:11:31,700
So thank you so much and welcome, everybody. Perhaps we can begin in the sort of characteristically, perhaps the Danto way of philosophers.

95
00:11:31,700 --> 00:11:38,660
And just to clarify the key terms in both democracy or democratic culture and artificial intelligence.

96
00:11:38,660 --> 00:11:46,820
So when a lot of people think about democracy, they think about Majoritarian Decision-Making, especially in the context of elections,

97
00:11:46,820 --> 00:11:53,210
and then they think about formal political institutions like legislatures that enact laws,

98
00:11:53,210 --> 00:12:00,080
courts that adjudicate upon them, administrators that implement them, police that enforce them.

99
00:12:00,080 --> 00:12:07,250
But the notion of a democratic culture goes beyond purely that formal set of institutions.

100
00:12:07,250 --> 00:12:12,050
And I think one of the things that's most impressive about the work that both Alan and Josh

101
00:12:12,050 --> 00:12:18,320
have done is that that emphasised the need for democracy beyond this institutional context.

102
00:12:18,320 --> 00:12:22,670
But I also know that you somewhat disagree about the nature of that broader context.

103
00:12:22,670 --> 00:12:31,530
Perhaps Joshie might begin by explaining in what sense democracy needs to be broader than just the institutional focus.

104
00:12:31,530 --> 00:12:37,340
Yes, thanks. Thanks very much, John. It's really a privilege to be here.

105
00:12:37,340 --> 00:12:46,460
And I want to, if I may, begin by offering to congratulations one to you for your position as director of the institute

106
00:12:46,460 --> 00:12:52,850
and the other to Oxford for having had the good fortune to be able to bring you in.

107
00:12:52,850 --> 00:13:00,800
As the head of this. I really can't think of anyone who's got the combination of intellectual breadth

108
00:13:00,800 --> 00:13:07,790
and ethical sensibilities that better suits him or her or they to play this role.

109
00:13:07,790 --> 00:13:14,060
So congratulations to both of you. Thanks for the opportunity.

110
00:13:14,060 --> 00:13:19,400
You know, I teach courses in Democratic theory, as does Alan.

111
00:13:19,400 --> 00:13:29,480
And often you begin the course in Democratic theory by discussing what are commonly called minimalist theories of democracy and minimalist theories.

112
00:13:29,480 --> 00:13:40,520
Democracy is a matter of competition for power that's resolved through elections and then a peaceful transfer of power subsequent to the elections.

113
00:13:40,520 --> 00:13:46,970
And this is a view that's presented classically by Joseph Schumpeter in his book Capitalism,

114
00:13:46,970 --> 00:13:54,530
Socialism and Democracy, Norberto Bobbio and Adam Sawhorse Gear, contemporary proponent of it.

115
00:13:54,530 --> 00:14:03,920
And and then when you're teaching the course, you always say, well, that's fine, but it sets the Democratic bar pretty low after January six.

116
00:14:03,920 --> 00:14:08,570
That doesn't seem like such a bad idea to a peaceful transfer.

117
00:14:08,570 --> 00:14:12,620
Powers are looking better. Really good.

118
00:14:12,620 --> 00:14:23,210
Still, I don't think we should let those hideous events of January 6th changed the basic conception of democracy as something more demanding,

119
00:14:23,210 --> 00:14:30,080
a form of self-government, a form of self-government in which the people understood as free and equal,

120
00:14:30,080 --> 00:14:37,310
have good expansive opportunities for open communication and organisation to form ideas

121
00:14:37,310 --> 00:14:45,530
and opinions that they aim to translate into programmes to turn into laws and policies.

122
00:14:45,530 --> 00:14:52,790
Democracy isn't simply a peacefully regulated contest for power, but also an informal,

123
00:14:52,790 --> 00:14:58,730
open ended, fluid, dispersed public discussion of matters of common concern.

124
00:14:58,730 --> 00:15:07,400
And that discussion, in turn, shapes public opinion and shapes and is shaped by civic activism and ultimately the exercise of formal political power.

125
00:15:07,400 --> 00:15:15,120
I think of it is a form of power based on an ideal of substantively equal communicative.

126
00:15:15,120 --> 00:15:21,690
Freedom and that system requires good and equal chances for people to express views, access,

127
00:15:21,690 --> 00:15:28,980
it's instructive information from reliable sources to hear a wide range of views and to coordinate with other people

128
00:15:28,980 --> 00:15:36,990
to explore interests and ideas with an eye to developing common understandings and advancing common concerns.

129
00:15:36,990 --> 00:15:41,670
And you know, that system of rights and opportunities that substantively equal communicative

130
00:15:41,670 --> 00:15:48,930
freedom depends in turn on a set of norms and dispositions of the participants.

131
00:15:48,930 --> 00:15:54,630
So there needs to be a concern for truth, for getting things right and a conception,

132
00:15:54,630 --> 00:16:04,050
not Thrasymachus conception of conception of politics as an activity in which the concern for truth matters.

133
00:16:04,050 --> 00:16:08,070
Doesn't that go without saying? Well, apparently not.

134
00:16:08,070 --> 00:16:16,200
Moreover, a concern for the common good politics is about competing and conflicting views about how do we address common shared concerns.

135
00:16:16,200 --> 00:16:22,050
And I think also a concern for civility, which I don't understand, is politeness or etiquette,

136
00:16:22,050 --> 00:16:30,090
but is acknowledging that the way you should argue and justify is in ways that respect and acknowledge the equal standing of other people.

137
00:16:30,090 --> 00:16:33,040
That's what I think Rawls meant by the duty of civility.

138
00:16:33,040 --> 00:16:41,580
Now, some of us are old enough to remember when digital technology was supposed to be great for all of this.

139
00:16:41,580 --> 00:16:49,920
When we finally had the technological foundations for a renewed and reinvigorated Hober Mocean public sphere,

140
00:16:49,920 --> 00:16:57,420
lowering the barriers to entry of speaking and listening and engaging.

141
00:16:57,420 --> 00:17:02,790
But, you know, the bloom fell pretty quickly from the digital rose.

142
00:17:02,790 --> 00:17:09,060
Still, I think the right way to think about the Democratic idea and the technological

143
00:17:09,060 --> 00:17:15,630
challenges to the Democratic idea is by ruling the understanding in this more ambitious

144
00:17:15,630 --> 00:17:21,180
conception of democracy is something more than a peaceful contest for power with a

145
00:17:21,180 --> 00:17:29,370
peaceful transition as incredibly important as that peaceful contest in transition is.

146
00:17:29,370 --> 00:17:31,800
This is a much richer notion of democracy.

147
00:17:31,800 --> 00:17:39,750
It involves citizens getting together, as it were, as free and equal citizens to deliberate about the common good,

148
00:17:39,750 --> 00:17:48,660
to make decisions about the common good, and that deliberations are structured by norms like respect for the truth, civility and so forth.

149
00:17:48,660 --> 00:17:52,260
Then you would agree with most of that, I think, wouldn't you?

150
00:17:52,260 --> 00:18:01,000
Yes, I think I share a vision of, you know, democracy is a deliberative democracy where laws are legitimate,

151
00:18:01,000 --> 00:18:07,860
don't need they've been the product of a reasoned exchange of arguments amongst free and equal.

152
00:18:07,860 --> 00:18:13,040
So that's sort of the ideal. But I think that the way I look at democracy in general,

153
00:18:13,040 --> 00:18:21,940
it a as a procedure to solve collective decisions where power is distributed equally discursive and voting power,

154
00:18:21,940 --> 00:18:28,640
but it's also more generally a regime form. And even more broadly, a type of society,

155
00:18:28,640 --> 00:18:32,480
and it's highly plausible that you can have a functional democracy in the narrow

156
00:18:32,480 --> 00:18:38,970
sense of a procedure or a regime unless you have a functional democratic society.

157
00:18:38,970 --> 00:18:41,270
And what do I mean by a democratic society?

158
00:18:41,270 --> 00:18:49,340
I mean, what Chuck Hagel meant when he when he first visited, basically, it's a society characterised by an equality of conditions.

159
00:18:49,340 --> 00:18:59,180
And it's really something that I you fully understand only when you come from a country where this equality of condition is lesser or in existence.

160
00:18:59,180 --> 00:19:03,350
For me, it was a revelation when I when I moved to the US, even as interestingly,

161
00:19:03,350 --> 00:19:13,580
the actual American regime is not that democratic at all, but in fact, in recent years has become expressly plutocracy and oligarchy.

162
00:19:13,580 --> 00:19:23,110
But the equality of conditions is still there. And it's it's I think it's essential to the stability and viability of a democratic regime.

163
00:19:23,110 --> 00:19:27,470
And can I just pick up one point? You published this important book recently, Open Democracy,

164
00:19:27,470 --> 00:19:33,530
which takes issue with the kind of view of democracy that Josh and Habermas and roles have.

165
00:19:33,530 --> 00:19:42,110
So the idea there is that there is a twin track of Democratic deliberation, decision making, the formal track of elected representatives and others,

166
00:19:42,110 --> 00:19:49,730
and then this informal track that sort of sets the agenda but doesn't actually decide the decision making is made in the formal track.

167
00:19:49,730 --> 00:19:58,910
You've argued for, in some sense, a less emphasis on elections and more on citizens making decisions themselves.

168
00:19:58,910 --> 00:20:03,550
Can you just bring up bring out that disagreement a little bit with Josh?

169
00:20:03,550 --> 00:20:07,300
Yeah. So I think it's it's just that I think in the 18th century,

170
00:20:07,300 --> 00:20:13,070
we took the wrong turn by reinventing democracy in an electoral form when I think there were other forms of

171
00:20:13,070 --> 00:20:18,170
democratic representation that could have been available because of different ideological and conceptual apparatus,

172
00:20:18,170 --> 00:20:22,460
including what I call autocratic representation based on random selection,

173
00:20:22,460 --> 00:20:30,470
something closer to the ancient Greek practises and also to some Swiss practise.

174
00:20:30,470 --> 00:20:36,860
At the time, there were not all that democracy, to be honest, but at least somewhat equally amongst the pool of candidates.

175
00:20:36,860 --> 00:20:44,900
And maybe it's not a real problem, I guess, but my, my, my, my divergence made my difference in disagreement.

176
00:20:44,900 --> 00:20:56,150
That way with Josh is that is still wedded to this dichotomous Habermas and vision of a space where you have like the

177
00:20:56,150 --> 00:21:04,370
politicians and the officials who make the decisions and then the space for a citizen which is wild and free flowing,

178
00:21:04,370 --> 00:21:10,790
and where magically, somehow an agenda for the official sphere is supposed to emerge.

179
00:21:10,790 --> 00:21:12,590
And I don't buy that,

180
00:21:12,590 --> 00:21:23,810
that there's this invisible market of one deliberate the invisible hand of a person in the wild can set up that agenda perfectly efficiently,

181
00:21:23,810 --> 00:21:32,180
efficiently. And I also think that distinguishing between a professional class of politicians and ordinary citizens is,

182
00:21:32,180 --> 00:21:37,490
you know, in the end, I think it's it's not a good thing that if you want an authentic democracy,

183
00:21:37,490 --> 00:21:45,260
we need to rotate ordinary citizens in and out of power and specifically in and out of legislative functions,

184
00:21:45,260 --> 00:21:51,530
but also in and out of the technical administrative structure while most of the laws are being made, actually.

185
00:21:51,530 --> 00:21:56,990
And meanwhile, we should also kerb the power of executives and return it to the function of

186
00:21:56,990 --> 00:22:02,840
executing the law and not making it so that that's that's probably the reason.

187
00:22:02,840 --> 00:22:07,790
So open democracies, not direct democracy, not all decisions being made by the whole populace.

188
00:22:07,790 --> 00:22:15,530
It is citizens representing citizens rather than an elite class representing citizens as elected representatives.

189
00:22:15,530 --> 00:22:19,700
Now, what I think would be interesting to come back to this issue a bit later to see

190
00:22:19,700 --> 00:22:24,500
whether a digital technology affects how we should adjudicate these two views.

191
00:22:24,500 --> 00:22:31,460
But, Nigel, I want to move on to you and just ask you the question, what do we mean by artificial intelligence?

192
00:22:31,460 --> 00:22:38,510
And also to ask you perhaps to expand on something that you said in your book, The Digital Age.

193
00:22:38,510 --> 00:22:47,720
You said that quite strikingly, that you're more afraid of what natural stupidity rather than artificial intelligence

194
00:22:47,720 --> 00:22:54,260
might wreak in the next 50 years of gradually more pervasive machines and smartness.

195
00:22:54,260 --> 00:22:56,420
Yes, thank you, John.

196
00:22:56,420 --> 00:23:06,020
It's it's very instructive to remember that the 1957 paper by Turing Computing Machine with Intelligence actually was published in mind in

197
00:23:06,020 --> 00:23:14,900
a philosophical journal which really kicked off 70 years now of intense work around the methods and techniques of artificial intelligence.

198
00:23:14,900 --> 00:23:22,700
Those methods and techniques have expanded extraordinarily when you try and define a little bit like trying to define life.

199
00:23:22,700 --> 00:23:26,730
There are many variations in many variants. The thing to remember,

200
00:23:26,730 --> 00:23:32,430
I think over the 70 years of the expansion of methods and techniques and we now have a huge

201
00:23:32,430 --> 00:23:38,880
toolbox of methods from structural encoding of causal principles to statistical methods,

202
00:23:38,880 --> 00:23:47,580
the latest forms of machine learning, many different types of ways of representing and reasoning about the world,

203
00:23:47,580 --> 00:23:56,850
that the thing that's really powered much of this acceleration has been the extraordinary and exponential increase in raw compute power,

204
00:23:56,850 --> 00:24:01,230
the technology underpinning the algorithms so that we now live in a world that's

205
00:24:01,230 --> 00:24:05,610
completely diffused with ubiquitous or pervasive supercomputer supercomputers,

206
00:24:05,610 --> 00:24:11,250
that we hardly acknowledge winning loads of algorithms to recognise speech faces,

207
00:24:11,250 --> 00:24:16,860
shopping patterns, suspicious regions of medical images, and the list goes on.

208
00:24:16,860 --> 00:24:24,840
What we have here is an extraordinary set of task, achieving programmes, very narrow, specific.

209
00:24:24,840 --> 00:24:34,320
Azis tuned to the specifics of particular tasks. But those are tasks that we've determined are important, worthwhile and profitable.

210
00:24:34,320 --> 00:24:38,850
And I think the application of extraordinary computing power driving A.I. with

211
00:24:38,850 --> 00:24:44,640
access to huge swathes of our data presents us with a host of ethical challenges,

212
00:24:44,640 --> 00:24:55,140
dilemmas of governance and choices about the ends to which that's all put my view in that quote was that I augments us in the best case,

213
00:24:55,140 --> 00:24:59,040
it empowers us and in the worst case, it will oppress and disenfranchise us.

214
00:24:59,040 --> 00:25:05,430
And those choices are ours alone and not likely to be emerging from any self-aware machine any time soon.

215
00:25:05,430 --> 00:25:13,380
So let's leave that to one side. And my point in that quote was that we might choose to put the machines in ascendancy over us,

216
00:25:13,380 --> 00:25:17,910
but that would be a case of natural stupidity rather than artificial intelligence.

217
00:25:17,910 --> 00:25:25,140
And I think we're seeing fantastic progress in the taming of particular problems that have large search spaces

218
00:25:25,140 --> 00:25:31,050
where the nature of the space being searched and what counts as a successful outcome can be very well specified.

219
00:25:31,050 --> 00:25:41,430
And we've just seen recently the fantastic results in protein folding, obstacle avoidance, face recognition, many, many examples.

220
00:25:41,430 --> 00:25:48,540
But I think what we're appreciating and in this conversation would explore more is that all of this can't just be instrumentalist.

221
00:25:48,540 --> 00:25:54,030
A.I.s use will reflect values that we collectively take to be important.

222
00:25:54,030 --> 00:26:02,430
And I think along with this idea of democracy that was in its origination and through its philosophical history,

223
00:26:02,430 --> 00:26:09,330
a host of ideas around the values that we most prise transparency, accountability,

224
00:26:09,330 --> 00:26:14,250
tolerance, autonomy, dignity, self-determination, reason, evidence, equity.

225
00:26:14,250 --> 00:26:22,470
And I think that's really important. You're seeing a real flourishing at the moment within departments of computer science, within our conferences,

226
00:26:22,470 --> 00:26:28,920
within the the community itself of discussions around value based computing,

227
00:26:28,920 --> 00:26:36,240
human centric A.I. And I think we all now share this deeper concern actually is back to Bobbio.

228
00:26:36,240 --> 00:26:43,800
Very powerful observation that the problem is that it's turning out that representation

229
00:26:43,800 --> 00:26:49,860
doesn't seem to be reflecting the interests and opinions of individuals, but rather organised group interests.

230
00:26:49,860 --> 00:26:57,900
And it's been what we see is that the procedure enables now large organisations

231
00:26:57,900 --> 00:27:04,080
to effect control rather more than individuals and democratic self-government.

232
00:27:04,080 --> 00:27:07,950
And I think that's where a source of a lot of people's unease arises.

233
00:27:07,950 --> 00:27:16,260
So I've got a very straightforward and I think much more instrumentalise view of what I is able to do at the moment.

234
00:27:16,260 --> 00:27:19,560
Great. So let me pick up on that point a little bit.

235
00:27:19,560 --> 00:27:28,380
So you're not seeing the action being in the development of artificial general intelligence, replicating human capacities across the board?

236
00:27:28,380 --> 00:27:36,900
It is about more narrow applications and the emphasis is on human choice about which values these more narrow applications will serve.

237
00:27:36,900 --> 00:27:41,670
So that's very much an image of a science, the technology for which we're responsible.

238
00:27:41,670 --> 00:27:48,960
I just want to ask a question to all of you and just with a brief answer, because the question is, is is rather imponderable.

239
00:27:48,960 --> 00:27:57,510
Is democracy a technology? Is democracy simply a mechanism for bringing up good results, like good decisions?

240
00:27:57,510 --> 00:28:02,070
Or is there inherent value in the democratic process?

241
00:28:02,070 --> 00:28:11,130
Or if both, where is where is its value mostly lie in the inherent value of everyone having an equal say as free and equal citizens,

242
00:28:11,130 --> 00:28:16,680
or in the fact that through this process we get better decisions than we would through other processes.

243
00:28:16,680 --> 00:28:20,620
Just a short answer from all of you then. Do you want to start? Yes.

244
00:28:20,620 --> 00:28:25,460
So I think I'll just answer with an anecdote by Robert Doll at the beginning of.

245
00:28:25,460 --> 00:28:31,820
Democracy, where you reports on the story of Danish Vikings driving up a river in France and from Normandy,

246
00:28:31,820 --> 00:28:38,360
some sticks to me on many levels and they were asked by a messenger coming out from the riverbanks, what's the name of your master?

247
00:28:38,360 --> 00:28:44,450
None. They replied, We are all equals. There's something intrinsically valuable about that.

248
00:28:44,450 --> 00:28:46,350
So they're.

249
00:28:46,350 --> 00:28:58,640
But you also think in terms of making better decisions that way as well, but I often get criticised for overemphasising the dimensions, so I thought.

250
00:28:58,640 --> 00:29:07,260
So you think that I care about the. So you've stressed the diversity, epistemic diversity conducive to better outcomes.

251
00:29:07,260 --> 00:29:15,550
Josh. Well, I think your muted mutagen.

252
00:29:15,550 --> 00:29:23,080
I am, but no longer. First on the issue of intrinsic value.

253
00:29:23,080 --> 00:29:38,530
Yes, and Ellen put it the powerfully on, it's a form of decision making amongst people conceived as equals with,

254
00:29:38,530 --> 00:29:53,590
as Rawls described, at certain moral powers. The in his last speech, Martin Luther in Memphis, Martin Luther King said, you know,

255
00:29:53,590 --> 00:30:02,600
the great glory of American democracy is the right to protest for right, not the right to grab, but the right to protest for right.

256
00:30:02,600 --> 00:30:12,070
So an exercise of that fundamental moral power. So I completely agree with you on the issue of at the same time,

257
00:30:12,070 --> 00:30:20,800
democracy is a form of political decision making and political decision making is a kind of collective problem solving.

258
00:30:20,800 --> 00:30:26,710
I, I know very little compelling, conclusive evidence that democracy there's some,

259
00:30:26,710 --> 00:30:32,980
but that democracy is consistently better at making collective decisions and alternatives.

260
00:30:32,980 --> 00:30:41,260
But it doesn't have to be better. It just has to be pretty good because there's a there's a limited willingness to pay

261
00:30:41,260 --> 00:30:46,330
the cost of a system that has intrinsic attractions but sucks in other respects.

262
00:30:46,330 --> 00:30:52,300
So but and there's no evidence that it is in way over the line.

263
00:30:52,300 --> 00:31:00,190
Well, good enough to make it worth having democracy for both of those reasons,

264
00:31:00,190 --> 00:31:08,990
but not as the system which somehow optimises on problem solving capacity to do it doesn't it doesn't need to do that.

265
00:31:08,990 --> 00:31:14,600
Nigel, I think we'll come back to that notion of sometimes setting up an expectation

266
00:31:14,600 --> 00:31:20,000
that I somehow has a more perfect solution it can struggle towards and or

267
00:31:20,000 --> 00:31:24,260
arrive at the point about this is there'll be there'll be fallibility all the

268
00:31:24,260 --> 00:31:28,840
way through these systems and they'll be educators we can't take account of.

269
00:31:28,840 --> 00:31:42,200
I mean, the way in which I can be at the service of the democratic process is around equity, availability of resource and insight,

270
00:31:42,200 --> 00:31:52,520
an explanation around facts of the matter where they can be adduced and and large participation of bringing groups together.

271
00:31:52,520 --> 00:32:00,410
But again, we've seen that experiments with the mobilisation at true scale of this falls into some really

272
00:32:00,410 --> 00:32:09,760
rather well understood problems now about giving people genuinely heterogeneous experience.

273
00:32:09,760 --> 00:32:17,980
OK, let's move on to some of the issues around the deployment of baby applications in decision making.

274
00:32:17,980 --> 00:32:23,920
I mean, there's the obvious allure in having an algorithm resolve important decisions.

275
00:32:23,920 --> 00:32:26,900
I mean, it's a mechanical judgement, free procedure.

276
00:32:26,900 --> 00:32:36,250
So you're not at the whim of someone's discretion or judgement, but promises to be faster, cheaper, perhaps more accurate, less biased than humans.

277
00:32:36,250 --> 00:32:40,720
I mean, there were these harrowing studies of judges engaged in sentencing and their

278
00:32:40,720 --> 00:32:44,800
sentences seem to be influenced by the time of day or whether they had lunch.

279
00:32:44,800 --> 00:32:52,100
And also just eliminating what Cass Sunstein calls noise, unwanted inconsistencies between various decision makers.

280
00:32:52,100 --> 00:32:54,640
So there is this tremendous promise to get.

281
00:32:54,640 --> 00:33:04,730
On the other hand, there's the worry that decision making is somehow through algorithms is somehow incompatible with democratic equality, for example.

282
00:33:04,730 --> 00:33:15,170
So there's been a lot of studies that talk about, for example, the ways in which algorithmic decision making to say facial recognition or in having

283
00:33:15,170 --> 00:33:19,370
to do with criminal justice amplifies and exacerbates existing inequalities,

284
00:33:19,370 --> 00:33:28,550
for example, not being able to recognise people with darker skin tones or discriminate against people from poorer neighbourhoods and so forth.

285
00:33:28,550 --> 00:33:39,350
Nigel, is this endemic to algorithmic decision making or can these discriminatory unfair effects be mitigated?

286
00:33:39,350 --> 00:33:46,970
Well, when we design and anticipate these systems, when we think about the goals that we're going to put them to work on,

287
00:33:46,970 --> 00:33:54,050
then we absolutely must reflect on on this broad range of potential and beddings.

288
00:33:54,050 --> 00:33:59,990
Otherwise, you will end up with the classic pratfalls that we've seen in some of these areas.

289
00:33:59,990 --> 00:34:00,410
Now, you know,

290
00:34:00,410 --> 00:34:10,100
if you were if you were well accustomed to the notion that this would be a classification system within these boundaries for these outcome types,

291
00:34:10,100 --> 00:34:17,810
then that's one thing. But when these things are deployed in very broad, open, textured contexts, you're really heading for trouble.

292
00:34:17,810 --> 00:34:21,140
And this is where I think the the kind of philosophical context,

293
00:34:21,140 --> 00:34:26,600
the one that you these are not entirely deterministic systems in, not in the sense that they run,

294
00:34:26,600 --> 00:34:33,350
but in the sense that was they're deployed and interpreted the open texture of the environment, society in which they are deployed.

295
00:34:33,350 --> 00:34:41,990
And of course, we can point to all sorts of very and we have done in the past in our discussions, very exciting examples of human bias.

296
00:34:41,990 --> 00:34:50,720
The column in the Tversky study is this extraordinary evidence that we're very susceptible to very simple distortions and

297
00:34:50,720 --> 00:34:58,370
effects in our decision making and perhaps our algorithms can help us edges towards better decisions that one thing,

298
00:34:58,370 --> 00:35:01,130
rather than putting the system entirely in charge,

299
00:35:01,130 --> 00:35:09,140
an earlier generation of A.I. systems was seen as a decision support what they did and in some modalities,

300
00:35:09,140 --> 00:35:14,480
explicitly it was critique the human decision making, trying to get them out of a local minima,

301
00:35:14,480 --> 00:35:19,430
some kind of local preference or bias zone they were in.

302
00:35:19,430 --> 00:35:25,190
So I think you can play this to support human decision making.

303
00:35:25,190 --> 00:35:32,540
But when you allow the system to simply be deployed without a broad range of appreciation

304
00:35:32,540 --> 00:35:38,480
of the cases of the task to which is being put to critically reflect on that,

305
00:35:38,480 --> 00:35:43,520
you will end up with the rather poor examples we've seen over time recently,

306
00:35:43,520 --> 00:35:49,250
I think in this area with facial recognition or bots online or whatever that might be.

307
00:35:49,250 --> 00:35:55,080
Anyone else want to come in on that? I was just thinking that, you know,

308
00:35:55,080 --> 00:36:02,370
there are different situations if you take a plane and you don't really care whether you are driven by a robot or a human beings,

309
00:36:02,370 --> 00:36:09,420
if you're handed a sentence for a crime you've committed, you absolutely care about the reasons behind the sentence.

310
00:36:09,420 --> 00:36:21,480
So I think it comes down to the reason this is justifiability, the kind of reason you're owed as human beings and citizens in some contexts.

311
00:36:21,480 --> 00:36:27,490
So I suppose even the use of loss that I'm very much in favour of, in fact,

312
00:36:27,490 --> 00:36:34,080
was toned down or avoided by 18th century philosophers because they had no rhyme or reason.

313
00:36:34,080 --> 00:36:41,520
It was not easy to justify. And so they preferred elections because it's based on choice and consent and full rationality.

314
00:36:41,520 --> 00:36:48,750
And I do understand the worry about non self-liquidating or non self-explanatory AI,

315
00:36:48,750 --> 00:36:55,980
where you're just given results with no understanding of how you got to those results.

316
00:36:55,980 --> 00:37:03,900
So this is an interesting point that it's not simply that we're interested in a certain kind of decision being within a frame of accuracy or fairness.

317
00:37:03,900 --> 00:37:11,550
It's also the reasons for the decision. I think this ties in with Joshes notion of the duty of stability, which is not simply about politeness,

318
00:37:11,550 --> 00:37:18,490
but about giving our fellow citizens the appropriate sorts of reasons, the decisions that affect them, especially for officials.

319
00:37:18,490 --> 00:37:23,460
Now, there is a kind of major issue here around the explained ability of AI.

320
00:37:23,460 --> 00:37:29,970
So I just want to quote something from a scientist, Dudayev Tul, on the deep learning style of machine learning.

321
00:37:29,970 --> 00:37:36,480
He says, I find many users who say that it works well and we don't know why.

322
00:37:36,480 --> 00:37:40,890
Once you unleash it on large data, deep learning has its own dynamics.

323
00:37:40,890 --> 00:37:47,640
It does its own repair and its own optimisation. And it gives you the right results most of the time.

324
00:37:47,640 --> 00:37:54,060
But when it doesn't, you don't have a clue about what went wrong and what should be fixed.

325
00:37:54,060 --> 00:38:00,600
In particular, you do not know if the fault is in the programme, in the method or because things have changed in the environment.

326
00:38:00,600 --> 00:38:06,570
We should be aiming, he says, a different kinds of transparency. So you don't know why things went wrong.

327
00:38:06,570 --> 00:38:13,200
But presumably that must also apply to the case where things go right? So there's a kind of divorce between outcome and process.

328
00:38:13,200 --> 00:38:16,170
What do you say to someone who says, look.

329
00:38:16,170 --> 00:38:24,080
I just want to have the right decision that really care about the nature of the process, this is a kind of abstract intellectual concern.

330
00:38:24,080 --> 00:38:27,330
Just, you know, so a few things on that.

331
00:38:27,330 --> 00:38:36,510
I want to go back to the previous point about algorithms and just and I think in the similar spirit to what was said,

332
00:38:36,510 --> 00:38:41,910
just one thing to bear in mind is a tour down historical lane.

333
00:38:41,910 --> 00:38:51,750
I mean, the development of a theory of algorithms in the 1930s is one of the most extraordinary intellectual achievements of the 20th century.

334
00:38:51,750 --> 00:38:56,790
And there's an important point about it that I just want to emphasise, which is, I mean,

335
00:38:56,790 --> 00:39:02,160
basically what happened was Hilbert has a programme in the foundations of mathematics.

336
00:39:02,160 --> 00:39:06,510
He's concerned about completeness, consistency and decide ability.

337
00:39:06,510 --> 00:39:15,570
Girdle takes care of the first two problems. There's things, incompleteness, theorems, the problem of decide ability is left.

338
00:39:15,570 --> 00:39:23,310
And in 36, 37, church interring solve the incidence problem.

339
00:39:23,310 --> 00:39:32,580
Another negative result, and this is one of these cases where you have a you couldn't solve the decision

340
00:39:32,580 --> 00:39:36,990
problem until you had a precise characterisation of what an algorithm is.

341
00:39:36,990 --> 00:39:45,540
And that's what Turing proposes, is I mean, church had a different one, but Turing's becomes the dominant one, computable by a Turing machine.

342
00:39:45,540 --> 00:39:52,160
And now you could prove you could conclusively resolved the decision.

343
00:39:52,160 --> 00:40:05,050
Problem, a pure problem in the most abstract, rarified areas of philosophy and the foundations of mathematics, that's the basis of computer science.

344
00:40:05,050 --> 00:40:12,080
I mean, it's an extraordinary case of following out a kind of intellectually pure problem and doing something that has,

345
00:40:12,080 --> 00:40:15,620
you know, obviously incredibly profound consequences.

346
00:40:15,620 --> 00:40:25,850
Now, one of the troubles is that I think people don't don't have a kind of grip on what an algorithm is.

347
00:40:25,850 --> 00:40:35,780
And so what you get are statements about how, you know, in allocation decisions of the kind that you describe,

348
00:40:35,780 --> 00:40:42,690
the algorithm did it OK, this is like literally nonsense, literally nonsense.

349
00:40:42,690 --> 00:40:50,710
In the following way, there are accountable infinity of algorithms.

350
00:40:50,710 --> 00:41:00,490
So if you don't like the one that you're using, there are plenty more where that one came from and it's a kind of fetishism

351
00:41:00,490 --> 00:41:04,990
attributing a magical property to the algorithm that you're blaming the algorithm.

352
00:41:04,990 --> 00:41:08,500
It's your fault. Look in the mirror. Don't look at the algorithm.

353
00:41:08,500 --> 00:41:15,970
I mean, I'm here very much agreeing with what Nigel has been saying about having a kind of clarity about what the values are that you're pursuing,

354
00:41:15,970 --> 00:41:19,780
what the sense of purpose is and how things can go. That's all on you.

355
00:41:19,780 --> 00:41:23,710
It's on your data. It's also on your model. It's not just a data driven model.

356
00:41:23,710 --> 00:41:39,550
So this kind of blaming algorithms, I think, is letting the letting people off the hook in in a way that they shouldn't be let off the hook.

357
00:41:39,550 --> 00:41:48,820
Now, the particular issue that you raised about explained ability that you day a pearl raises about people,

358
00:41:48,820 --> 00:41:54,550
that's really specifically an issue about deep learning. I mean, it's just a characteristic of it.

359
00:41:54,550 --> 00:42:03,250
It's not an A.I. issue generically. It's a deep learning issue because of the way that deep learning models work.

360
00:42:03,250 --> 00:42:10,180
And it's easy to overexaggerate the uninterpretable ability of deep learning mahsa lots of people who are spending lots

361
00:42:10,180 --> 00:42:21,820
of time trying to figure out how you can make sense of what's happening at a node in a in a neural in a neural net.

362
00:42:21,820 --> 00:42:28,960
But I think. The reason that.

363
00:42:28,960 --> 00:42:36,550
There's a demand for explained ability appropriately is not because there's

364
00:42:36,550 --> 00:42:41,860
an intellectual curiosity about what's happening in the guts of the system,

365
00:42:41,860 --> 00:42:46,450
that intellectual curiosity can't be answered any more than you can.

366
00:42:46,450 --> 00:42:52,690
Answer the question of why, you know, the photon went through one hole rather than the other hole.

367
00:42:52,690 --> 00:42:56,650
That's the way I don't think that can be answered.

368
00:42:56,650 --> 00:43:04,480
The reason that people are concerned about that is because systems are not, as Nigel said, these systems are not perfect and reliable.

369
00:43:04,480 --> 00:43:09,160
And you want to know about the you may want to know why it succeeds when it succeeds,

370
00:43:09,160 --> 00:43:12,250
but you want to know that partly because you know it's going to fail.

371
00:43:12,250 --> 00:43:19,930
It's going to say that your dog is a dishtowel and you want to know how the [INAUDIBLE] did it come up, you know, with a classification like that.

372
00:43:19,930 --> 00:43:31,780
And so I think if I were 25 years old and starting out in philosophy and looking for a really fantastic problem to explore,

373
00:43:31,780 --> 00:43:38,620
I would start writing about explained ability because it's an incredibly rich, complicated,

374
00:43:38,620 --> 00:43:44,260
philosophically, technically complicated area with huge practical importance.

375
00:43:44,260 --> 00:43:53,410
And I think there were about 18 different concerns that are being wrapped together in under the label of explain ability.

376
00:43:53,410 --> 00:44:00,890
And there's enormous benefit that could come from somebody doing some untangling around those.

377
00:44:00,890 --> 00:44:07,310
That is a fantastic recommendation that hopefully some of our listeners will take up.

378
00:44:07,310 --> 00:44:13,670
Let me move to another issue. So we've talked a little bit about issues about fairness and discrimination with respect,

379
00:44:13,670 --> 00:44:18,980
algorithms that address, albeit briefly, this issue of explain ability.

380
00:44:18,980 --> 00:44:26,000
But of course, another worry that people have about the operation of algorithms is that many of them are fed on big data.

381
00:44:26,000 --> 00:44:30,020
And there are issues about the data and in particular issues around privacy.

382
00:44:30,020 --> 00:44:35,420
And of course, a lot of this is the focus, the criticism of what's known as surveillance,

383
00:44:35,420 --> 00:44:41,120
capitalism and the use of big data in order to push people consumer goods.

384
00:44:41,120 --> 00:44:49,310
But you might think that there is a more serious issue when there's a prospect to advance public health to the use of big data.

385
00:44:49,310 --> 00:44:57,170
So the question I'd like perhaps to address briefly here is whether I in areas like advancing public health, for example,

386
00:44:57,170 --> 00:45:06,200
contact tracing and dealing with the pandemic forces us to reconfigure how we think of the balance between individual rights,

387
00:45:06,200 --> 00:45:16,430
like privacy and issues like the furtherance of public goods, like controlling pandemics and eliminating them.

388
00:45:16,430 --> 00:45:23,510
And I just want to quote something that you mentioned, Nigel, in your book.

389
00:45:23,510 --> 00:45:34,900
You say. The single thing that every citizen and every corporate decision maker needs to understand is that the enormous data stores that government,

390
00:45:34,900 --> 00:45:37,480
government agencies, corporations,

391
00:45:37,480 --> 00:45:47,110
trusts and individuals hold are as much a key part of national and international infrastructure as the road network.

392
00:45:47,110 --> 00:45:54,650
So this idea that these massive stores of data are, in a sense, public goods that we can debate about, exclude ability and so forth,

393
00:45:54,650 --> 00:45:58,390
that they're public goods in the sense that they stand to benefit everyone and

394
00:45:58,390 --> 00:46:03,200
it's benefiting one person is not at the expense of benefiting anyone else.

395
00:46:03,200 --> 00:46:10,040
But you claim that we need to open up these sources of data in a way, however, that respects privacy.

396
00:46:10,040 --> 00:46:17,170
How do we how do we strike that balance, that the data can be both open yet at the same time respecting privacy?

397
00:46:17,170 --> 00:46:23,530
Well, I mean, this isn't a single dimension of variation on this,

398
00:46:23,530 --> 00:46:30,880
we often talk about data as being on a spectrum from open to closed in private, but in fact, the conditions under which you vary that.

399
00:46:30,880 --> 00:46:36,370
I mean, I think what we appeal for in the open data movement is that a broader foundation of

400
00:46:36,370 --> 00:46:42,430
possible of data is made available on the open and publicly open licence conditions.

401
00:46:42,430 --> 00:46:44,620
And much of that is not about personal data.

402
00:46:44,620 --> 00:46:54,310
It's everything from when the trains run to where the roads actually join one another or the legal addresses in a country, you know,

403
00:46:54,310 --> 00:47:04,420
you can argue about what you like as your as your as your non personal public data infrastructure, the stuff that relates to individuals.

404
00:47:04,420 --> 00:47:10,270
Now we're seeing an interesting change in perceptions at the moment of a public health crisis.

405
00:47:10,270 --> 00:47:12,940
There is a sense of a collective good. I mean,

406
00:47:12,940 --> 00:47:22,330
there are already laws and regulations which allow various intrusions into your liberties and your your kind

407
00:47:22,330 --> 00:47:27,160
of rights of movement in the case of certain infectious diseases where we're living under them right now.

408
00:47:27,160 --> 00:47:30,280
The issue about how your data is used now,

409
00:47:30,280 --> 00:47:37,090
I think we often get to heads up with this because we think it's bound to be about dereference identifiable data.

410
00:47:37,090 --> 00:47:44,620
But we've got lots of examples now where the data can remain in highly secure enclaves behind all sorts of levels.

411
00:47:44,620 --> 00:47:55,930
Security are my models that seek to understand the transmission of the virus at this moment or where the susceptible groups are in the demographic.

412
00:47:55,930 --> 00:48:04,810
Why models can interrogate that data and the data never has to leave a particular set of secure data centres, for example.

413
00:48:04,810 --> 00:48:08,410
And where and we know that because our patient records are held in that way

414
00:48:08,410 --> 00:48:16,060
and they're currently linked behind various forms of of of secure firewall.

415
00:48:16,060 --> 00:48:24,280
So the question is, the danger comes when that data is just kind of shifted around the place with very little attention and thought to the security.

416
00:48:24,280 --> 00:48:30,850
There's a different aspect to this which says, well, maybe that data will be better held close to the individuals whose data it is.

417
00:48:30,850 --> 00:48:34,600
And so you use different sorts of rights and interests over that data.

418
00:48:34,600 --> 00:48:39,070
And then the decision is where and how you can aggregate and collect that data together.

419
00:48:39,070 --> 00:48:41,290
So we're seeing a very interesting discussion at the moment,

420
00:48:41,290 --> 00:48:49,000
not just around the technology for securing and making data available as and when different groups and interests need it.

421
00:48:49,000 --> 00:48:52,270
But around the need for new kinds of data institution,

422
00:48:52,270 --> 00:49:00,780
new kinds of possibly co-operatives and mutuals where the data is stewarded for public or other collective interests.

423
00:49:00,780 --> 00:49:05,230
So I think that there's not a there's not a single answer to this.

424
00:49:05,230 --> 00:49:09,970
It might appear like that when you look at the kind of throw around contact tracing,

425
00:49:09,970 --> 00:49:15,640
but even then, there are good designs and much worse designs about how you do that effectively.

426
00:49:15,640 --> 00:49:24,790
Presumably, you're not saying that all this goes through some detail of consent by particular individuals with respect to their data?

427
00:49:24,790 --> 00:49:35,200
It can't. In all cases, you've got to kind of understand where the urgency of some of the use to which you want to understand that.

428
00:49:35,200 --> 00:49:42,910
I mean, again, patient records are good examples. There's a general sense that, you know, that within a trusted health care delivery system,

429
00:49:42,910 --> 00:49:48,040
there are limits and purposes to which that data can and should or might be caught.

430
00:49:48,040 --> 00:49:58,600
And I think that we're becoming more sophisticated about our understandings around how you permit and vary that level of of linkage.

431
00:49:58,600 --> 00:50:05,350
But there has to be very careful characterisation of the rules and governance that sits over the top of that.

432
00:50:05,350 --> 00:50:08,650
So we don't see mass exfiltration, for example,

433
00:50:08,650 --> 00:50:19,960
of important information and insights to two organisations that may or may not have a general broader public interest.

434
00:50:19,960 --> 00:50:27,280
Josh, I know you want to come in on this. Yeah, just you know, I agree very much with what Nigel is saying.

435
00:50:27,280 --> 00:50:38,980
I mean, there's real. I mean, there's interesting philosophical work on privacy, but the technologies of privacy protection are important,

436
00:50:38,980 --> 00:50:43,720
sophisticated, secure enclaves, differential privacy to a whole bunch of things.

437
00:50:43,720 --> 00:50:57,040
I want to come, though, to the issue about contact tracing, which this was a big topic around the tensions that you were describing,

438
00:50:57,040 --> 00:51:02,080
potential tensions between protecting privacy and solving a big public health problem.

439
00:51:02,080 --> 00:51:06,220
And I think the issue there are some very instructive issues here.

440
00:51:06,220 --> 00:51:14,020
And I'll start with the terminological point, but I think it has very large, substantive implications.

441
00:51:14,020 --> 00:51:19,130
There is no such thing as digital contact tracing.

442
00:51:19,130 --> 00:51:22,430
There's digital exposure notification.

443
00:51:22,430 --> 00:51:29,120
Contact tracing is different from exposure notification, exposure notification is you got a piece of information.

444
00:51:29,120 --> 00:51:33,650
The information is that you are within a certain relevant distance,

445
00:51:33,650 --> 00:51:43,790
10 or 15 feet for a certain relevant period of time, 15 minutes to somebody who received a positive test for covid.

446
00:51:43,790 --> 00:51:52,670
Now, I've taken several contact tracing courses. I have certification from Johns Hopkins and from the WHL as I never did any because

447
00:51:52,670 --> 00:51:57,110
the programmes were so badly organised that my volunteering didn't get me anywhere.

448
00:51:57,110 --> 00:52:00,620
Or maybe they were well enough organised that they didn't want me and they knew that.

449
00:52:00,620 --> 00:52:03,680
But anyway, I got I took the courses.

450
00:52:03,680 --> 00:52:10,700
And what you learn in a contact tracing course is the contact tracing is basically a human public health activity.

451
00:52:10,700 --> 00:52:15,860
Step one is letting somebody know about an exposure it.

452
00:52:15,860 --> 00:52:25,850
But contact tracing is what you do after that. It's telling somebody how they can be tested, telling somebody how they can inform their employer,

453
00:52:25,850 --> 00:52:31,010
helping them to get some support while they're in isolation.

454
00:52:31,010 --> 00:52:37,040
It's not it's not exposure notification.

455
00:52:37,040 --> 00:52:39,620
Not simply exposure notifications,

456
00:52:39,620 --> 00:52:55,610
and I think the fact that so much intellectual energy went into the arguments about privacy protection in contact tracing,

457
00:52:55,610 --> 00:53:01,010
is it revealing of a kind of if I can use this,

458
00:53:01,010 --> 00:53:04,580
you know, piece of [INAUDIBLE] terminology over indexing, I mean,

459
00:53:04,580 --> 00:53:14,570
over indexing on the significance of information as opposed to a human public health process.

460
00:53:14,570 --> 00:53:27,620
Now. That said, it's also true, so far as I know that, you know, you look in the United States now, case rates are down and mortality is down.

461
00:53:27,620 --> 00:53:37,520
It's a very good. Things have improved a bunch. And there's a discussion about why there is literally no one, as far as I've read,

462
00:53:37,520 --> 00:53:44,270
literally no one who thinks that the rates are down because of successful exposure notification.

463
00:53:44,270 --> 00:53:50,040
Perhaps a lot of people spent a lot of time developing very privacy, protecting exposure notification apps.

464
00:53:50,040 --> 00:53:52,940
I was involved in one of those projects. It's great.

465
00:53:52,940 --> 00:54:02,780
But there if you look at the literature on success stories and less successful stories about dealing with covid Cross nationally,

466
00:54:02,780 --> 00:54:10,130
it is really not a story about some places that had fantastic exposure notification apps.

467
00:54:10,130 --> 00:54:15,980
The most popular exposure notification app is this Indian app.

468
00:54:15,980 --> 00:54:20,360
I've mispronounce it Oroya Saturnus, millions and millions and millions.

469
00:54:20,360 --> 00:54:25,490
But I don't think anybody thinks it's really done very much good.

470
00:54:25,490 --> 00:54:31,950
So there's, I think the public discussion around.

471
00:54:31,950 --> 00:54:39,750
Contact tracing apps, actually, exposure notification apps and protecting people's privacy and dealing with the public health issue and how to

472
00:54:39,750 --> 00:54:49,440
navigate these troubled waters is an expression of a kind of distortion in public discussion of the kind that I know.

473
00:54:49,440 --> 00:55:00,340
As you know, this is the kind of thing that you think about dealing with covid from the contact tracing point of view is a human public health.

474
00:55:00,340 --> 00:55:06,910
Stand up, large numbers of people didn't really happen anywhere to deal with these human needs.

475
00:55:06,910 --> 00:55:16,350
It's not about providing somebody with a piece of information that they don't know what to do with and probably won't do anything with.

476
00:55:16,350 --> 00:55:21,720
Helen, do you want to come into this and just said, I was thinking about this connexion between,

477
00:55:21,720 --> 00:55:28,560
again, democracy and its privacy rights and property rights of our own deadline?

478
00:55:28,560 --> 00:55:35,740
The problem is that with the current sort of structure of our so-called democratic government, a lot of these decisions,

479
00:55:35,740 --> 00:55:42,840
these trade-offs that you think is societies are entitled to make between more privacy or more collective safety.

480
00:55:42,840 --> 00:55:52,130
They are actually made by. Groups of people that are not entirely representative of the larger interests of the population,

481
00:55:52,130 --> 00:55:57,320
sometimes they are delegated to a technocratic sort of structure or agency or

482
00:55:57,320 --> 00:56:05,630
administrative groups that that that that make decisions on on behalf of the rest of us.

483
00:56:05,630 --> 00:56:19,460
And so my own take is that I would like to see something like a citizen's convention on I gotta digital technology to talk about these issues from it,

484
00:56:19,460 --> 00:56:28,040
from an authentically. Democratic representative perspective, if you will, and it could happen at the global level, I know that,

485
00:56:28,040 --> 00:56:35,570
for example, people in Australia don't like Australia is organising something like a citizen's assembly.

486
00:56:35,570 --> 00:56:44,660
You randomly selected on genome editing, and I'm sure we could do something similar on questions of artificial intelligence.

487
00:56:44,660 --> 00:56:46,020
That's a really interesting point.

488
00:56:46,020 --> 00:56:52,580
Is this the kind of lazy assumption that writes accounting, majoritarian norms and therefore not the business of majority,

489
00:56:52,580 --> 00:56:58,100
but rather of technocratic experts, judges and others itself is a mistake?

490
00:56:58,100 --> 00:56:59,330
That was a good thing to actually.

491
00:56:59,330 --> 00:57:05,840
You know, when I read your first question and answer your first question, I was thinking, well, actually, our privacy rights democratic to begin with.

492
00:57:05,840 --> 00:57:09,680
I mean, it's the liberal notion. Did the Greeks have privacy rights?

493
00:57:09,680 --> 00:57:15,080
I don't think so. The business of the private citizen with the business of of the city.

494
00:57:15,080 --> 00:57:25,880
So it's actually interesting to see how stretched, to use your terms on the concept of democracy has become.

495
00:57:25,880 --> 00:57:32,690
It encompasses all the things we deem desirable, including liberal rights, including human rights and all these things.

496
00:57:32,690 --> 00:57:41,330
And I just said before that I do think of democracy as a way of life in society, but it doesn't have to be a liberal society necessarily.

497
00:57:41,330 --> 00:57:41,990
It doesn't have to be.

498
00:57:41,990 --> 00:57:49,430
I mean, we could be a little bit more we could be a little bit more parsimonious in the way we think about a democratic society.

499
00:57:49,430 --> 00:57:53,250
OK, so if I could just say yes, I would just say a word about that.

500
00:57:53,250 --> 00:57:58,370
I just in support of thinking of privacy as having a value in a democracy,

501
00:57:58,370 --> 00:58:05,390
which is if you're constantly under threat of having everything about your life exposed,

502
00:58:05,390 --> 00:58:16,670
the idea that that is going to be empowering to you as a vigilant member of the polity feels a little bit like a stretch.

503
00:58:16,670 --> 00:58:21,410
I mean, I understand there are a whole bunch of other foundations for the idea of a right of privacy,

504
00:58:21,410 --> 00:58:25,880
but I think there is a specific case that I'm sure you were.

505
00:58:25,880 --> 00:58:33,020
I, of course, agree. I do think that if you want citizens who are capable of thinking, they have to have the sort of like a freedom,

506
00:58:33,020 --> 00:58:39,890
internal space to develop ideas that are not that everyone who I would second that,

507
00:58:39,890 --> 00:58:43,970
but still, I think that this triggers the thought that we may want to be, again,

508
00:58:43,970 --> 00:58:52,100
a bit more parsimonious about where those concepts are conscious of, say, and defensive.

509
00:58:52,100 --> 00:58:54,380
By the way, exposure tracing.

510
00:58:54,380 --> 00:59:04,500
There is research recently for such a group in Oxford who claim that they've seen a two point three percent decrease in the in income and vice.

511
00:59:04,500 --> 00:59:16,880
So, I mean, it is hard to do that. Absolutely. And it's a theme that is emerging is that democracy requires genuine deliberation.

512
00:59:16,880 --> 00:59:26,240
That's informed by the respect for truth, that offers reasons to our fellow citizens for decision making,

513
00:59:26,240 --> 00:59:35,030
and that is oriented towards the common good. A lot of people now think digital technology is subverting these processes of public reason,

514
00:59:35,030 --> 00:59:38,990
and they think very often that there is a kind of unholy alliance between,

515
00:59:38,990 --> 00:59:39,860
on the one hand,

516
00:59:39,860 --> 00:59:47,480
big tech that is in the business of sucking up as much of people's attention as possible in order then to manipulate them in various ways.

517
00:59:47,480 --> 00:59:54,980
And on the other hand, exclusionary forms of populism that use social media in order to push out their message to get

518
00:59:54,980 --> 01:00:02,960
people more and more sort of caught up with extremist views and in filter bubbles and so forth.

519
01:00:02,960 --> 01:00:12,800
And so there's this kind of very negative take. And I just wanted to read to you a passage in an op ed recently published by The New York Times.

520
01:00:12,800 --> 01:00:16,880
The author is very distinguished scholar Sesana Zubov,

521
01:00:16,880 --> 01:00:25,790
and she talks about the epistemic coup that is surveillance capitalism, an epistemic coup comparable well,

522
01:00:25,790 --> 01:00:34,880
also at a different level, but comparable in some ways with the coup that she thinks took place or was attempted on January 6th in the US capital.

523
01:00:34,880 --> 01:00:42,890
She says the following is an information. Civilisation societies are defined by questions of knowledge.

524
01:00:42,890 --> 01:00:49,280
How was it distributed? The authority that governs its distribution and the power that protects that authority?

525
01:00:49,280 --> 01:00:53,180
Who knows? Who decides? Who knows? Who decides? Who decides?

526
01:00:53,180 --> 01:01:01,640
Who knows? Surveillance capitalist's. Now hold the answers to each question, though we never elected them to govern.

527
01:01:01,640 --> 01:01:11,240
This is the essence of the epistemic coup. They claimed the authority to decide who knows by asserting ownership rights over our personal information

528
01:01:11,240 --> 01:01:17,660
and defend that authority with the power to control critical information systems and infrastructures.

529
01:01:17,660 --> 01:01:21,650
So beyond any issue of algorithms, there is this deeper structural.

530
01:01:21,650 --> 01:01:28,910
Issue that she's talking about, this epistemically being in a sense perpetrated by big tech,

531
01:01:28,910 --> 01:01:35,860
in some sense in cahoots with government, can I get any responses to that claim?

532
01:01:35,860 --> 01:01:48,020
Yes. Just OK, so, you know, Shoshana Zubov has been writing about these issues for a long time.

533
01:01:48,020 --> 01:01:55,250
She's a fantastic sociologist with great insight, surveillance, capitalism.

534
01:01:55,250 --> 01:02:10,430
Very interesting book. I read a statement like that and it reminds me of something that I often see in amongst people who have,

535
01:02:10,430 --> 01:02:19,070
you know, deep association with technology companies, which is there's a kind of weird, so great chest beating.

536
01:02:19,070 --> 01:02:23,150
And then a few years later, we are so horrible.

537
01:02:23,150 --> 01:02:29,340
We're the most horrible. And it's all the most whether it's the the best or the worst.

538
01:02:29,340 --> 01:02:39,770
There's this I think what she says in that statement is incredibly exaggerated and not about the concern,

539
01:02:39,770 --> 01:02:47,540
the kind of antitrust concern about dominance over communication infrastructure.

540
01:02:47,540 --> 01:02:51,230
No, but the control of what people.

541
01:02:51,230 --> 01:03:00,640
So let me just say a few things about American politics, which she's writing about despite its parlous state.

542
01:03:00,640 --> 01:03:05,980
First of all, it's true that Donald Trump got 74 million votes.

543
01:03:05,980 --> 01:03:13,480
It's also true that Joe Biden got six million more than that, the most votes anybody has ever gotten running for president.

544
01:03:13,480 --> 01:03:19,780
It's also true that there are a bunch of states that were central to Biden's victory,

545
01:03:19,780 --> 01:03:30,070
including Georgia and Wisconsin, where there was a huge long play in state politics over a very long period of time.

546
01:03:30,070 --> 01:03:35,880
That Stacey Abrams in Georgia.

547
01:03:35,880 --> 01:03:46,440
Ben Wikler, amongst others, in in Wisconsin, years of work, years of work, hard work, political work that produced that change.

548
01:03:46,440 --> 01:03:51,910
Thirdly, there is an absolutely decisive break in the Democratic Party from the milk,

549
01:03:51,910 --> 01:03:56,400
the market fundamentalism that dominated the party really since the 1970s.

550
01:03:56,400 --> 01:04:05,850
This is not about Clinton. This is about Carter. Gary Hart, it's goes back a long time to you.

551
01:04:05,850 --> 01:04:12,660
If you look at the debates, about the current debates, about the stimulus's, you see that there's a real willingness to go big,

552
01:04:12,660 --> 01:04:18,240
to do something about climate and a real willingness to go big and to do something about this,

553
01:04:18,240 --> 01:04:27,390
you know, deep, persistent, profound set of issues about racial justice when I lay that parlous state.

554
01:04:27,390 --> 01:04:33,300
Yes, absolutely. But I lay those facts alongside what Shoshana Zubov says.

555
01:04:33,300 --> 01:04:49,250
I just feel like she's. You know, it's a 30000 foot look at something where things look a little different when you get down to 29000 feet even.

556
01:04:49,250 --> 01:04:55,480
Yeah. You're on view, John. I'm sorry, can I get reactions from the audience?

557
01:04:55,480 --> 01:04:59,920
So I don't know, I always feel like this is really new and provocative,

558
01:04:59,920 --> 01:05:06,070
but an epistemic coup, I all I'm thinking of is the actual political failure behind it.

559
01:05:06,070 --> 01:05:11,560
I remember Zuckerberg testifying in front of Congress and what what a what a catastrophe that was,

560
01:05:11,560 --> 01:05:17,980
how incompetent, ignorant those senators in particular, where, I mean, a disgrace.

561
01:05:17,980 --> 01:05:25,540
So, yeah, then surprise, surprise. You know, then these companies take over and do what they want.

562
01:05:25,540 --> 01:05:28,900
And we we're left in our filter bubbles and echo chambers.

563
01:05:28,900 --> 01:05:35,110
And no one is really any way to figure out whether truth life, I think, is the basic political problem,

564
01:05:35,110 --> 01:05:40,270
lack of real lack of regulations and over representation of corporations in Congress.

565
01:05:40,270 --> 01:05:45,740
The American Congress, 84 percent belong to the ten wealthiest percent of the population.

566
01:05:45,740 --> 01:05:51,640
Hello. The working class completely ignored. Then you wonder why you know the rise in torture.

567
01:05:51,640 --> 01:06:00,550
So so you know, and we know for a fact that so basically the overrepresentation of economic interests is the problem,

568
01:06:00,550 --> 01:06:06,640
not the epistemically by some companies. That's that's pointing the finger in the wrong way.

569
01:06:06,640 --> 01:06:08,760
This is a really good point.

570
01:06:08,760 --> 01:06:19,180
And I think in your book, you mention empirical research that says that if you control for the political views of the 10 percent wealthiest in the US,

571
01:06:19,180 --> 01:06:23,380
then there is no real impact of Democratic decision.

572
01:06:23,380 --> 01:06:29,950
Majorities are not causally efficient once you control for the economic preferences of the richest people.

573
01:06:29,950 --> 01:06:30,910
I think this is my way.

574
01:06:30,910 --> 01:06:40,990
And I kind of agree with Josh that ACORN and Bartel's in a very famous 2014 article called Testing its Theories of Democracy Now.

575
01:06:40,990 --> 01:06:45,610
I mean, of course, but it sounds awfully plausible to me.

576
01:06:45,610 --> 01:06:51,430
I just sort of wonder whether this sort of very vehement attack on big tech actually

577
01:06:51,430 --> 01:06:56,740
shrouds bigger problems about capitalism and its relationship with democracy.

578
01:06:56,740 --> 01:06:58,030
So let me just put that out there.

579
01:06:58,030 --> 01:07:07,000
I mean, John Rawls, your teacher and mentor, Josh, famously thought that his theory of justice was incompatible with capitalism,

580
01:07:07,000 --> 01:07:15,710
that you had to have a property owning democracy or that you had to have some sort of socialist system that respected liberal rights.

581
01:07:15,710 --> 01:07:23,030
Is someone like Zubov actually papering over the deeper issue, because, you know, if you think about even on epistemic basis,

582
01:07:23,030 --> 01:07:27,710
prior to the advent of social media having sort of great impact,

583
01:07:27,710 --> 01:07:32,930
how did the mainstream media deal with the Iraq war pretty much lined up in favour of it.

584
01:07:32,930 --> 01:07:38,450
How did it deal with the Vietnam War? So it's not as if there's this great history of successes.

585
01:07:38,450 --> 01:07:57,200
Well, you know, I'm mindful of the fact that, you know, I'm 69 years old, and if I say I've heard this music before, this is nothing new.

586
01:07:57,200 --> 01:08:04,310
It's like the geezers, you know, talking. That's what you do. You say, oh, nothing new here, but I don't know.

587
01:08:04,310 --> 01:08:23,930
There there we are. I mean, I think it's all of the you know, consider just the case of political fragmentation.

588
01:08:23,930 --> 01:08:33,610
More or less than this is not a nice example, but I understand, but more or less than in Germany, 1928.

589
01:08:33,610 --> 01:08:39,520
Fragmentation of the media in Germany, 1928, same deal.

590
01:08:39,520 --> 01:08:50,860
Now, that's as I say, it's not a very good but it had zero like obviously zero to do with digital digital technology.

591
01:08:50,860 --> 01:09:04,640
If you look at the outsized growth in inequality, whether wealth inequality or income inequality.

592
01:09:04,640 --> 01:09:11,900
Not so clear that it's got I mean, there may be something to do with the success of tech firms,

593
01:09:11,900 --> 01:09:17,330
but that's not the I don't think on anybody's story, that's the biggest driver.

594
01:09:17,330 --> 01:09:24,000
And I think outsized growth of inequality has something to do with the nature of our politics.

595
01:09:24,000 --> 01:09:30,830
You look at the failures to go back to the previous point about, you know, the break from market fundamentalism.

596
01:09:30,830 --> 01:09:36,460
You look at the failure, politically consequential failures around trade policy.

597
01:09:36,460 --> 01:09:45,370
I had nothing to do with digital technology and all of this is consistent with saying

598
01:09:45,370 --> 01:09:55,420
that the there are real serious antitrust issues about the dominance of some players.

599
01:09:55,420 --> 01:09:59,600
And and there's there's a lot of action on that.

600
01:09:59,600 --> 01:10:05,560
What it took to get the action in the United States was for people to reject the Chicago

601
01:10:05,560 --> 01:10:13,870
school antitrust view that antitrust is all about consumer welfare as registered in price,

602
01:10:13,870 --> 01:10:18,340
even if you go to consumer welfare is registered and product quality, not price.

603
01:10:18,340 --> 01:10:24,820
Odina Srinivasan has written fantastically on how Facebook competed on product quality.

604
01:10:24,820 --> 01:10:31,270
In particular, Facebook competed against MySpace on privacy protections.

605
01:10:31,270 --> 01:10:42,300
And I think MySpace was gone, so the I mean, this is the idea of attributing all of this stuff to the latest piece of technology.

606
01:10:42,300 --> 01:10:46,420
I think it's it's a little much.

607
01:10:46,420 --> 01:10:56,020
I want to let Nigel come in before we go to questions from the audience, just because I think I think that's pretty much what just to say.

608
01:10:56,020 --> 01:10:59,800
But I would just say that one thing that is available to us,

609
01:10:59,800 --> 01:11:07,570
I think in an age of huge amounts of data traffic, is to apply the same technology that we're using now,

610
01:11:07,570 --> 01:11:14,720
technology, much of it, to shape the traffic, to allow these video conferences to happen, to help us see where the information is flowing.

611
01:11:14,720 --> 01:11:19,240
Now, that may sound like the reason that matters is we have very little visibility,

612
01:11:19,240 --> 01:11:26,980
either as individual citizens or as regulators about what the market concentrations are actually like.

613
01:11:26,980 --> 01:11:30,040
And this isn't just around the commercial context.

614
01:11:30,040 --> 01:11:37,630
It could just as well do with the facts of where concentration of data is at everything from safety data to scientific data.

615
01:11:37,630 --> 01:11:44,200
There is a real need to get much more transparency about the actual flows of data to the destinations.

616
01:11:44,200 --> 01:11:52,960
And actually, unfortunately, what we have is an arms race where everybody is trying to obscure and variously protect that information.

617
01:11:52,960 --> 01:12:02,740
I think if open data, which I see as a jewel to much of the prospect of algorithmic algorithmic decision making, is to offer a hope.

618
01:12:02,740 --> 01:12:08,680
It's to say certain classes of information we could recognise as being a good one is things like beneficial

619
01:12:08,680 --> 01:12:16,060
ownership on issues like where the data flows actually are in this so-called surveillance of his populace,

620
01:12:16,060 --> 01:12:18,130
then we can address some of these alleged concerns.

621
01:12:18,130 --> 01:12:27,400
Are they real or they imagined all that actually concentrations that are invidious to the conditions, the democratic conditions we want to foster.

622
01:12:27,400 --> 01:12:32,860
OK, I'm going to move now to some questions from the audience. So the first one is from Nicholas Hincker.

623
01:12:32,860 --> 01:12:39,970
And he says there's so much progress on recommending systems, learning so much about individuals from data.

624
01:12:39,970 --> 01:12:44,270
Are there ways for this to help democracy and not harm democracy?

625
01:12:44,270 --> 01:12:54,720
So what about the positive side? So then you think that perhaps these developments can actually help have a more radical form of democracy?

626
01:12:54,720 --> 01:13:02,890
I tried to imagine what it could do to a completely reinvented form of democracy where a lot

627
01:13:02,890 --> 01:13:11,140
of it takes place online and you'd be registered on it on a platform called Citizen Book,

628
01:13:11,140 --> 01:13:14,440
you know, like a sort of democratised non-profit version of Facebook.

629
01:13:14,440 --> 01:13:21,520
And and you'd have artificial intelligence at your fingertips to help you synthesise knowledge about, I don't know,

630
01:13:21,520 --> 01:13:33,310
various candidates at the global level or the European level or similar large scale political level to help you sift through various profiles

631
01:13:33,310 --> 01:13:44,050
and then choose the right candidate for an electoral mood for some issues or and sort of question when there's a sort of electronic referendum.

632
01:13:44,050 --> 01:13:54,580
So I could see these these technologies as augmenting and speeding up the process of making collective decisions.

633
01:13:54,580 --> 01:13:58,570
So that's one way in which I was trying to imagine that these tools could be useful.

634
01:13:58,570 --> 01:14:08,650
You could envision I know Jim Fishkin, I'd send for his artificial intelligence, artificial intelligence facilitator for Notkin deliberation,

635
01:14:08,650 --> 01:14:17,290
so that that also cheapens the cost of large scale democratic innovations like deliberative polls, like citizens assembly.

636
01:14:17,290 --> 01:14:21,280
So in the end, you know, I already had a chance to argue that point before.

637
01:14:21,280 --> 01:14:31,870
But I do think that we the danger, the generosity of our A.I. depends in part on the quality of our democracy.

638
01:14:31,870 --> 01:14:41,650
If we are in control, if we know how to run, instrumentalise technology to good ends, then then we'll be fine.

639
01:14:41,650 --> 01:14:47,770
But our democracies are in terrible shape. In fact, they are hardly democracy at all.

640
01:14:47,770 --> 01:14:52,270
And so and just to go back to the point about democracy is a way of life.

641
01:14:52,270 --> 01:14:53,620
And, you know,

642
01:14:53,620 --> 01:15:01,780
I also think the question of the workplace is really important in the sense that I wish we could all say we have new masters like those Vikings,

643
01:15:01,780 --> 01:15:09,730
you know, but if you if you're in a corporation and your health care depends on your boss and, you know, and in this pandemic,

644
01:15:09,730 --> 01:15:17,260
you could lose your job and you get depressed and, you know, you have a master and can you be free?

645
01:15:17,260 --> 01:15:22,700
Can you. So all these things are connected to my life.

646
01:15:22,700 --> 01:15:26,300
Another question from Lloyd Rushton regarding personal data,

647
01:15:26,300 --> 01:15:33,920
should we impose fiduciary duties on those who seek to utilise such data in addition to empowering data subjects through collective initiatives?

648
01:15:33,920 --> 01:15:36,110
Nigel.

649
01:15:36,110 --> 01:15:47,970
I think that's certainly something where that exists in certain cases of sensitive data, you're not simply free to do what you want with this stuff.

650
01:15:47,970 --> 01:15:57,830
I think I think evolving a regulatory landscape and a legal landscape where there are duties of care over how data is actually

651
01:15:57,830 --> 01:16:07,100
deployed is something that we will see emerge as we start to understand the scope and range of it will become very acutely aware.

652
01:16:07,100 --> 01:16:11,930
And you see this particularly, for example, in areas where age appropriate design.

653
01:16:11,930 --> 01:16:14,750
So if you want to look at some of the most harrowing examples,

654
01:16:14,750 --> 01:16:22,830
then you look into examples of where a persuasive design is used with respect to interacting with children on children's apps.

655
01:16:22,830 --> 01:16:39,780
So I think there is every prospect of needing to think about a balance of interests that place fiduciary actual constraints on use of data.

656
01:16:39,780 --> 01:16:48,320
If I may say, I do think also, though, and I don't think we're in disagreement here, fiduciary responsibilities are fine.

657
01:16:48,320 --> 01:16:55,070
But you have you've got a bunch of data that's a threat surface that somebody is going to go after.

658
01:16:55,070 --> 01:17:00,800
And so you I think it's very important to think on the technology side of privacy protection.

659
01:17:00,800 --> 01:17:04,670
So and there are a whole bunch of tools. There's differential privacy.

660
01:17:04,670 --> 01:17:12,770
You do own device processing. You have a principle of data minimisation so that you only collect the data that are needed.

661
01:17:12,770 --> 01:17:15,980
You have some sunset law that requires that the data go.

662
01:17:15,980 --> 01:17:26,030
You have secure enclave's because you know, what you're what you want is both to have the data protected against threats,

663
01:17:26,030 --> 01:17:33,140
but also, frankly, to protected to be protected against your own temptations to do the wrong thing.

664
01:17:33,140 --> 01:17:38,870
But if you want to protect it against yourself, so fiduciary obligations.

665
01:17:38,870 --> 01:17:42,680
But I think discharging those in a way that gives a kind of technolog,

666
01:17:42,680 --> 01:17:50,840
a set of that wraps the protections in a pretty secure technology is an important strategy.

667
01:17:50,840 --> 01:17:57,740
We've talked about some of the power structures and power imbalances that into the operation of democracy.

668
01:17:57,740 --> 01:18:03,260
But Carrisi and asks a question which I'm going to summarise now, that is, how can we reach a consensus,

669
01:18:03,260 --> 01:18:08,810
a democratic consensus about fairness with respect to the operation of algorithms when we know that

670
01:18:08,810 --> 01:18:14,390
people's conception of fairness is so heavily shaped by cultural norms and they differ so greatly.

671
01:18:14,390 --> 01:18:22,430
So there's so much cultural fragmentation within our democracies, who gets whose conception of fairness gets to win out?

672
01:18:22,430 --> 01:18:28,940
Who overrides whom? She asks. So I think that that's a wrong way to ask the question.

673
01:18:28,940 --> 01:18:32,120
It's not about who overrides to it's it's to convince who.

674
01:18:32,120 --> 01:18:38,420
And again, I am just going to bring it back to my favourite thing, which is the Citizens Conventions.

675
01:18:38,420 --> 01:18:42,590
One took place in France recently on climate change, very controversial issue.

676
01:18:42,590 --> 01:18:50,030
You had renters against landlords about mandatory housing renovations, you know, to limit greenhouse emissions.

677
01:18:50,030 --> 01:18:57,110
And so if you look at it as a as a as a negative sum game where some people are going to lose, some people are going to win,

678
01:18:57,110 --> 01:19:03,800
and some people's views of fairness are going to override other people's views of fairness if it's not never going to work.

679
01:19:03,800 --> 01:19:12,260
But if you look at it as a as an orientation toward the common good, we all reason as citizens to go into enterprise.

680
01:19:12,260 --> 01:19:20,930
We are trying to achieve a common goal. In that case was kerbing greenhouse emissions by 40 percent of 1990 levels in a state of social justice.

681
01:19:20,930 --> 01:19:28,850
It actually works. If you institutionalised property, you give it time, you facility well and you structure it well.

682
01:19:28,850 --> 01:19:37,400
It's a success story from my perspective, in terms of what they managed to produce as consensual solutions to very,

683
01:19:37,400 --> 01:19:41,780
very thorny value disagreements and practical disagreements.

684
01:19:41,780 --> 01:19:46,190
They would sit at the end during the seven session about those proposals,

685
01:19:46,190 --> 01:19:53,240
some of them likely to be very difficult to pass through the regular legislative system.

686
01:19:53,240 --> 01:19:59,360
And they voted with like high percentage of of approval, like eighty five percent they convinced each other.

687
01:19:59,360 --> 01:20:04,920
So I believe in that. That's the whole idea of deliberative democracy.

688
01:20:04,920 --> 01:20:11,160
John, if I could add a point on that, I agree with what Alan is saying,

689
01:20:11,160 --> 01:20:18,690
but I also want to pick up on another way that the question is put, which makes me a little uncomfortable.

690
01:20:18,690 --> 01:20:20,220
I mean, it's a great question.

691
01:20:20,220 --> 01:20:34,350
And I can tell you, it comes up all the time and you won't be surprised to hear that if you start by asking what's your conception of fairness?

692
01:20:34,350 --> 01:20:40,980
And you go around the room, you know, honestly, I don't care if it's collective deliberation or individual reflection.

693
01:20:40,980 --> 01:20:51,010
It's going to be all over the place. If you say here's a more concrete problem, is this fair?

694
01:20:51,010 --> 01:20:52,190
Let me give an example.

695
01:20:52,190 --> 01:21:00,730
There was a story in the Times a year ago about speech recognition systems, all the speech recognition systems have have higher word error.

696
01:21:00,730 --> 01:21:06,930
Existing speech recognition systems have a higher word error rate on African-American vernacular English.

697
01:21:06,930 --> 01:21:15,320
I don't know anybody maybe I don't know enough people, but I don't know anybody who thinks, oh, that's that's fine, there's no fairness issue there.

698
01:21:15,320 --> 01:21:18,900
And if you say, well, you know,

699
01:21:18,900 --> 01:21:26,190
if you use African-American vernacular English or you have to do is code switch and don't speak in African-American vernacular English,

700
01:21:26,190 --> 01:21:31,050
it'll be understood fine. No one thinks that that's a fair solution to the problem.

701
01:21:31,050 --> 01:21:38,610
So if you take concrete's rather than abstractions about what's your theoretical definition of fairness,

702
01:21:38,610 --> 01:21:46,710
I think you find a little considerably less of the kind of unending disputation.

703
01:21:46,710 --> 01:21:48,960
I don't mean it goes away because there are hard cases,

704
01:21:48,960 --> 01:21:57,010
but a considerably less of it than you find if you start with high level abstractions about what fairness is.

705
01:21:57,010 --> 01:22:03,090
That's resolving ethical disagreements doesn't mean having to penetrate down to philosophical foundations of those.

706
01:22:03,090 --> 01:22:08,280
No, that's right. And what you what I find is and I think this is very important and it goes back to something

707
01:22:08,280 --> 01:22:13,290
that you said at the beginning about the importance of philosophy in this enterprise.

708
01:22:13,290 --> 01:22:16,440
I agree with you. Philosophy is important in the enterprise,

709
01:22:16,440 --> 01:22:25,500
but it would be hopeless for philosophers to participate in the enterprise were it not for the fact that most people are morally intact.

710
01:22:25,500 --> 01:22:30,210
Philosophers don't teach people morality.

711
01:22:30,210 --> 01:22:35,130
They help to clarify moral thinking. They help to make it more salient.

712
01:22:35,130 --> 01:22:39,120
But if you weren't in a universe of people who,

713
01:22:39,120 --> 01:22:47,700
you know on the whole and for the most part had a kind of facility with moral distinctions and ideas, it would be the enterprise would be hopeless.

714
01:22:47,700 --> 01:22:56,670
That's just to amplify that. The fact is that this area has become practical by the operationalisation of of of context.

715
01:22:56,670 --> 01:23:03,210
So we're both into contact with really interesting classes of moral questions in particular cases.

716
01:23:03,210 --> 01:23:06,570
And I think that is why the discussion between computer science,

717
01:23:06,570 --> 01:23:13,860
application builders and and and ethical philosophical is just such a rich conversation

718
01:23:13,860 --> 01:23:19,560
because you can actually ground in just the way just as difficult questions,

719
01:23:19,560 --> 01:23:25,650
in particular implemented realities. And this has a powerful pedagogic aspect.

720
01:23:25,650 --> 01:23:33,090
So one of my colleagues, Michael Phillips Brown, teaches ethics to engineers, but not by saying, well, his concern is Aristotle in his mill,

721
01:23:33,090 --> 01:23:41,490
not go ahead and apply that to your problems, but by working within the issues that arise in their research is kind of bespoke kind of ethics.

722
01:23:41,490 --> 01:23:50,040
And therefore, even if you have deep of theoretical disagreements, that doesn't prevent being able to resolve those more concrete problems.

723
01:23:50,040 --> 01:23:54,060
As Josh said, you know, I'm going to ask two questions.

724
01:23:54,060 --> 01:24:01,380
I think it's likely the last I'm going to run them together. So one is a question from Jose Marti, who's asking?

725
01:24:01,380 --> 01:24:08,880
Is there any specific contribution that I might make, in particular to strengthen the quality and extension of Democratic deliberation?

726
01:24:08,880 --> 01:24:13,410
So that's about improving Democratic deliberation by us.

727
01:24:13,410 --> 01:24:17,310
But there's another question from my colleague, Ted Lieutenant, who says,

728
01:24:17,310 --> 01:24:21,840
suppose artificial general intelligence can someday generate legislation that is

729
01:24:21,840 --> 01:24:27,600
maximally just political equality is preserved because no one has any influence at all.

730
01:24:27,600 --> 01:24:38,330
Does this reveal democracy's value? I mean, I guess the point would be at that point, Democratic Decision-Making would become redundant.

731
01:24:38,330 --> 01:24:43,700
So enhancement redundancy, pick up either of those and hensman,

732
01:24:43,700 --> 01:24:48,230
I feel like I gave some examples of what I had in mind, but I just had another thought,

733
01:24:48,230 --> 01:24:54,560
actually, which is about mass deliberation, because the big puzzle for me is whether we can have quality deliberation,

734
01:24:54,560 --> 01:24:58,640
the kind that would be legitimising of laws and policies at the large scale,

735
01:24:58,640 --> 01:25:06,110
or whether we always have to go through a representative moment where we did the task of the aversion to a subset with a random or to.

736
01:25:06,110 --> 01:25:13,810
And from what I understand, maybe artificial intelligence could again, because it helps speed up and clarify.

737
01:25:13,810 --> 01:25:18,040
Issues, maybe, I'm not entirely sure,

738
01:25:18,040 --> 01:25:25,210
but I could envisage a role there for artificial intelligence in connecting a humongous amount

739
01:25:25,210 --> 01:25:31,720
of people and to get to that ideal of like all the brains connected together via technology.

740
01:25:31,720 --> 01:25:36,550
I'm not sure it's remotely possible, but that's actually at least an idea I have.

741
01:25:36,550 --> 01:25:47,620
And then what was the second question? How do we know what would happen if we had a perfectly just taking over it was God, right?

742
01:25:47,620 --> 01:25:53,800
I mean, that's pretty much who that would be.

743
01:25:53,800 --> 01:26:00,460
I guess it depends on how much we care about having reasons for what we do and and whether we want to have a master or not.

744
01:26:00,460 --> 01:26:06,550
Because if it's gone to the OR in some respects and I go back to the Vikings, I think we don't want to live with the master.

745
01:26:06,550 --> 01:26:10,930
I think there's a value in that that trumps even getting the right outcomes.

746
01:26:10,930 --> 01:26:15,820
Is it always the case that there are people who could make better decisions than us, like, for example,

747
01:26:15,820 --> 01:26:22,210
someone could make far better sartorial decisions on my behalf, but I don't surrender is something.

748
01:26:22,210 --> 01:26:28,510
I mean, I'm not saying that I'd want to only I think there's a threshold of mistakes that that can go below.

749
01:26:28,510 --> 01:26:34,180
But I think there's a value in learning from your own mistakes and not getting to the right.

750
01:26:34,180 --> 01:26:41,250
And so but sort of deploying yourself as a human being in life through your own agenda.

751
01:26:41,250 --> 01:26:50,010
I guess. Josh, maybe what would happen and no one could make better sartorial decisions for me than me.

752
01:26:50,010 --> 01:26:57,480
I'll just I'll put a stake in the ground there. And if they tried to, I would make their life miserable.

753
01:26:57,480 --> 01:26:59,100
Well, let's see. Nigel's taking his tie off.

754
01:26:59,100 --> 01:27:09,220
So he's coming back to you in one respect, you know, with with respect to the you know, the redundancy alliance as well.

755
01:27:09,220 --> 01:27:18,250
Wouldn't that be like God? And I guess what I would hope is that it would be a Assab abscond dittos.

756
01:27:18,250 --> 01:27:23,560
Yes, I did have the good grace to let us get on with a good grace to let us make continue

757
01:27:23,560 --> 01:27:34,480
to make a mess and sometimes a blessing out of our crookedly timbered lives.

758
01:27:34,480 --> 01:27:40,450
I do wonder whether one of the you know, there have been various attempts at liquid democracy,

759
01:27:40,450 --> 01:27:47,140
liquid democracy meets the block, all sorts of ideas about how you might change the democratic process or the or that.

760
01:27:47,140 --> 01:27:56,410
One of the things that's been striking about recent developments in large scale platforms has been the emergence of citizen science,

761
01:27:56,410 --> 01:28:00,850
a participatory engagement that hasn't been about doing all of the science,

762
01:28:00,850 --> 01:28:06,040
but has been about really important aspects of helping the science succeed, you know,

763
01:28:06,040 --> 01:28:12,070
or tapping into people's enthusiasm to either collect the primary data or to clean data up.

764
01:28:12,070 --> 01:28:16,630
And I think that sometimes we we kind of elevate the decision making to the point where

765
01:28:16,630 --> 01:28:21,430
I think there's lots of ways of participating that these platforms can facilitate.

766
01:28:21,430 --> 01:28:27,430
And we found some fascinating data. In fact, the Zooniverse group here in Oxford, Chris Lento's group,

767
01:28:27,430 --> 01:28:34,450
did some fascinating work where they could actually look at large scale participation and run machine learning across those participations

768
01:28:34,450 --> 01:28:46,680
and find the sub groups within that large participation who were particularly effective for certain sorts of of task or contribution.

769
01:28:46,680 --> 01:28:53,440
And I think there's a lot to be had in reframing this in a slightly broader sense of what it is to participate.

770
01:28:53,440 --> 01:28:59,560
Absolutely. People often forget the Universal Declaration includes the right to participate in science and culture,

771
01:28:59,560 --> 01:29:03,070
and arguably that is a dimension of the sort of broader,

772
01:29:03,070 --> 01:29:09,610
informal public sphere that is relevant, highly relevant to democratic participation on that optimistic note.

773
01:29:09,610 --> 01:29:14,380
Nigel, I think we're going to have to draw to a close. Thank you so much to our three wonderful speakers.

774
01:29:14,380 --> 01:29:20,560
I really enjoyed this invigorating discussion. We've had to touch over lightly some really important themes,

775
01:29:20,560 --> 01:29:24,910
but I think it's been incredibly stimulating and hopefully given our audience a

776
01:29:24,910 --> 01:29:29,920
taste for the sorts of discussions we're hoping to have in future at the institute.

777
01:29:29,920 --> 01:29:35,650
Thank you to the three speakers that really appreciate you taking the time to be part of the launch.

778
01:29:35,650 --> 01:29:41,480
And thank you to everyone for tuning in and listening to us tonight.

779
01:29:41,480 --> 01:29:46,060
Thanks very much. Thank you. All right. Thank you, everyone.

780
01:29:46,060 --> 01:30:20,121
Thank you. Thank you.