1
00:00:01,110 --> 00:00:04,500
Thanks, Tom, and thank you, everyone in particular.

2
00:00:04,500 --> 00:00:13,350
Thanks, everyone for having me here, I'm having a great time in Oxford, so what I'm going to present here is a draft paper.

3
00:00:13,350 --> 00:00:26,730
So it's very much a work in progress on the topic of mental privacy and the ethics of mind reading 

4
00:00:26,730 --> 00:00:33,100
.....

5
00:00:33,100 --> 00:00:41,680
So it's really a work in progress, so feel free to provide any charitable or uncharitable feedback that's very much welcome.

6
00:00:41,680 --> 00:00:55,440
So

7
00:00:55,440 --> 00:01:06,950
 I'll start with two quotes.

8
00:01:06,950 --> 00:01:20,540
The first one is by Paracelsus, renaissance physician and natural philosopher, who wrote Thoughts are free and subject to no rule. On them

9
00:01:20,540 --> 00:01:27,920
Rest the freedom of men. And the second quote is from centuries later.

10
00:01:27,920 --> 00:01:33,800
Historian John Bagnall Bury the author of The Monumental History of Freedom of Thought,

11
00:01:33,800 --> 00:01:42,710
started this book by saying that it is a common saying that it is free and men can never be hindered from thinking whatever it chooses,

12
00:01:42,710 --> 00:01:53,450
so long as it conceals what it thinks. And this to two quotes, I think kind of provide the framework for the position that I'm going to give,

13
00:01:53,450 --> 00:02:03,050
because in this paper, I have two main objectives. One is to scrutinise the technological and conceptual premises of mind reading,

14
00:02:03,050 --> 00:02:16,890
and the normative part is trying to make an argument that privacy of mental information should be considered a fundamental right.

15
00:02:16,890 --> 00:02:21,180
So when it comes to mind reading

16
00:02:21,180 --> 00:02:26,640
many bioethicists and neuro ethicists perceive this as the kind of outdated

17
00:02:26,640 --> 00:02:31,980
debate which was very popular in the 1990s and at the beginning of this century,

18
00:02:31,980 --> 00:02:39,720
following recent developments in a functional magnetic resonance imaging technology.

19
00:02:39,720 --> 00:02:44,220
This is what we could refer to today as the classic mind reading debate.

20
00:02:44,220 --> 00:02:48,090
So for those who don't know

21
00:02:48,090 --> 00:02:56,160
It's a technique to measure brain activity indirectly, namely by detecting changes associated with blood flow.

22
00:02:56,160 --> 00:03:00,120
So when a certain brain region is in use.

23
00:03:00,120 --> 00:03:05,970
So there is an increased neuronal activation, then blood flow to that region also increases.

24
00:03:05,970 --> 00:03:15,990
So by detecting this increase or decrease in blood flow, we can also make indirect inferences about neuronal activation that particular area.

25
00:03:15,990 --> 00:03:22,800
And a lot of neuroscientists a few decades ago were very excited about the

26
00:03:22,800 --> 00:03:31,140
possibility of using functional magnetic resonance imaging to ensure mental states.

27
00:03:31,140 --> 00:03:36,780
This is a quote by John Dylan Haynes a famous cognitive neuroscientist,

28
00:03:36,780 --> 00:03:46,710
who wrote in the mind reading entry of the Oxford Handbook of Neuroethics that by combining fMRI with pattern recognition, we can.

29
00:03:46,710 --> 00:03:51,420
It is possible to reach out very detailed contents of a person's thoughts,

30
00:03:51,420 --> 00:03:57,900
including detailed visual process ideas, memories and even intentions and emotions.

31
00:03:57,900 --> 00:04:06,720
And it is possible to read both implicit and even unconscious mental states, such as unconscious persons and decisions.

32
00:04:06,720 --> 00:04:15,840
So this kind of very enthusiastic claims became controversial a few years ago for a variety of reasons.

33
00:04:15,840 --> 00:04:24,420
One reason was that after I started is had a hard time being replicated replicated.

34
00:04:24,420 --> 00:04:33,420
So a lot of amateur analysis in the subsequent decade showed that mania for large studies failed to record data analysis

35
00:04:33,420 --> 00:04:43,680
published in their publications and making an exact replication of those studies nearly impossible and following this.

36
00:04:43,680 --> 00:04:50,580
A lot of people started talking about the replication crisis in fMRI based neuroscience.

37
00:04:50,580 --> 00:04:58,260
And then there was also a conceptual problem with such enthusiastic claims by Haynes and colleagues.

38
00:04:58,260 --> 00:05:04,380
This is well put by Martha Farah who argued that due to interpersonal variability fMRI

39
00:05:04,380 --> 00:05:10,230
I can demonstrate correlations between brain activity and behaviour on a group level.

40
00:05:10,230 --> 00:05:19,530
But its ability to identify a single mental state in one person may never be totally tolerable level of accuracy.

41
00:05:19,530 --> 00:05:31,830
So she is sort of lifecycles of scientific claims such as the Gartner's hype cycle.

42
00:05:31,830 --> 00:05:37,890
We could probably argue that the end of the last century,

43
00:05:37,890 --> 00:05:46,530
the beginning of this early century was a sort of peak of inflated expectations regarding mind reading.

44
00:05:46,530 --> 00:05:53,580
Then we went through what we can be. We can call the disillusionment phase.

45
00:05:53,580 --> 00:06:02,300
But I would argue that now we are moving up again towards what we can call a plateau of productivity neuroscience.

46
00:06:02,300 --> 00:06:10,510
And in a second, I will explain why, 

47
00:06:10,510 --> 00:06:18,400
Problematic conceptual premises of the classic definitions of mind reading, I think a very valuable account is provided by the internet,

48
00:06:18,400 --> 00:06:29,470
and so kids born in the early 1990s works on mental representations as they distinguish

49
00:06:29,470 --> 00:06:35,680
what they call representing from representatives of mental representations.

50
00:06:35,680 --> 00:06:45,580
Also hold the distinction between the vehicle of the mental state and the content of the mental state, and Denver and Ginsberg argued.

51
00:06:45,580 --> 00:06:49,840
And I think this is still valid today, as summarised thought,

52
00:06:49,840 --> 00:06:59,050
is that what we can't grasp from fMRI signals or other indirect or direct recordings is the

53
00:06:59,050 --> 00:07:04,690
vehicle of the mental state and not the contents of the mental states to use an analogy.

54
00:07:04,690 --> 00:07:17,310
We can reveal something like a traffic road sign, but not the actual process of crossing the road on the zero.

55
00:07:17,310 --> 00:07:19,050
So what has changed since then,

56
00:07:19,050 --> 00:07:28,590
why the debate about my inbreeding that happened to be obsolete a few years ago is interesting again today for neuroscience,

57
00:07:28,590 --> 00:07:36,990
for computer science and for others. I would argue there are two main socio technical trends that are interesting here.

58
00:07:36,990 --> 00:07:42,040
The first one, we can call it a decoding and the other one, we could call it big data,

59
00:07:42,040 --> 00:07:48,570
as applied to mental states or big mind data in absence of a better word.

60
00:07:48,570 --> 00:08:00,420
So why is I changing the science of inferring mental states from brain activity?

61
00:08:00,420 --> 00:08:07,620
Mainly because the AI models, especially an analysis artificial neural networks,

62
00:08:07,620 --> 00:08:13,950
are providing the algorithmic tools to enable a new type of inference.

63
00:08:13,950 --> 00:08:22,920
So logical inference in neuroscience in imaging studies was what has been called the forward inference.

64
00:08:22,920 --> 00:08:33,330
So you manipulate the specific psychological function in order to identify, identify localised effects of that manipulation on brain activity.

65
00:08:33,330 --> 00:08:40,180
So let's say you manipulate attention and then you identified the localised effects that manipulating

66
00:08:40,180 --> 00:08:48,030
that tension has on brain activity in which brain regions are involved with AI based models.

67
00:08:48,030 --> 00:08:54,480
What we can do is also what has been called by people like Russ Poldrack and others regarding inference.

68
00:08:54,480 --> 00:09:03,330
So you sort of reverse engineer the process so you can reason backwards from patterns of data such as neural activity.

69
00:09:03,330 --> 00:09:13,710
But as we will see, it's not necessarily brain dead, it's not necessarily neuronal activation to infer the engagement of specific mental processes.

70
00:09:13,710 --> 00:09:23,190
And this is a new thing, and it's quite exciting. What new approaches to competition neuroscience can do?

71
00:09:23,190 --> 00:09:33,330
I think this is one of the most interesting cases the work of Jack Gallant and his lab at the University of California, Berkeley.

72
00:09:33,330 --> 00:09:45,030
I think they are probably the research group that worldwide is coming closer to ensuring not only the vehicle of mental states,

73
00:09:45,030 --> 00:09:55,410
but also the content to understand its terminology. These are two studies that they published three years away from each other,

74
00:09:55,410 --> 00:10:01,260
the first ones about reconstructing the visual content of mental states from brain activity.

75
00:10:01,260 --> 00:10:09,150
So basically, they presented participants with visual stimuli, and this piece was a clip of a short movie.

76
00:10:09,150 --> 00:10:22,770
And then they train an artificial neural network to reconstruct the visual content of these mental states from using only brain activity.

77
00:10:22,770 --> 00:10:34,140
And as you can see, the level of detail is not grandiose, so you cannot determine, for example, the identity of the person involved.

78
00:10:34,140 --> 00:10:39,870
So you cannot use it for evidential testimony in court,

79
00:10:39,870 --> 00:10:45,510
but it's really quite advanced and sufficiently sophisticated to determine whether

80
00:10:45,510 --> 00:10:50,520
you are having a mental state related to a person as opposed to an unknown person,

81
00:10:50,520 --> 00:10:55,110
as opposed to an object and even as opposed to a non-human animal.

82
00:10:55,110 --> 00:10:57,720
So I think this is really quite exciting.

83
00:10:57,720 --> 00:11:05,960
In this other study, what they were able to do was to infer the content of the semantic content of mental states.

84
00:11:05,960 --> 00:11:10,860
Some of the first study was about visual content. The other one is about semantic content.

85
00:11:10,860 --> 00:11:16,140
So they train using the same methodological approach that you train an artificial

86
00:11:16,140 --> 00:11:24,210
neural network to decode the semantic content of mental states for productivity.

87
00:11:24,210 --> 00:11:31,740
So this map identifies with which content for could be inferred from brain activity.

88
00:11:31,740 --> 00:11:37,860
The concepts implied are those that have a higher statistical accuracy.

89
00:11:37,860 --> 00:11:50,520
So those that are more reliably the ones in in blue or purple are the ones that could be inferred with less statistical accuracy.

90
00:11:50,520 --> 00:11:56,080
And if you go close, you could see that all large taxonomic concepts.

91
00:11:56,080 --> 00:12:08,200
So, for example, being a German or even being a mammal could be inferred quite reliably by the artificial neural.

92
00:12:08,200 --> 00:12:17,350
The second socio technical trend, which I think is revamping interest on my reading, is what we can call a big mandate.

93
00:12:17,350 --> 00:12:22,870
As I said, if you have ideas about a better concept, this is very welcome.

94
00:12:22,870 --> 00:12:29,950
So that approaches that basically involve data related to mental states,

95
00:12:29,950 --> 00:12:37,690
so we can call this a large scale, the availability of neural and neural data resources,

96
00:12:37,690 --> 00:12:44,980
which is enabling processing activities that are aimed at inserting mentioned information by reversing friends.

97
00:12:44,980 --> 00:12:48,430
Why am I saying neuro and non-euro?

98
00:12:48,430 --> 00:12:57,910
Because this process of rigorous inference can allow us to reconstruct mental states not necessarily only from brain activity,

99
00:12:57,910 --> 00:13:07,840
but also from data such as voice recordings, text and so on, are several trends in this field.

100
00:13:07,840 --> 00:13:10,330
One is consumer neuro technology,

101
00:13:10,330 --> 00:13:18,790
which is basically new technology used outside of the clinical setting or outside of the biomedical research setting.

102
00:13:18,790 --> 00:13:29,230
As a personal technology, more and more people worldwide are using devices such as wearable neuro headsets like this chew that you see here.

103
00:13:29,230 --> 00:13:38,200
They're usually based on electroencephalography, so they collect electrical activity from the brain in a non-invasive manner,

104
00:13:38,200 --> 00:13:44,800
and this can be used for a variety of purposes, such as cognitive training, neurofeedback and so on.

105
00:13:44,800 --> 00:13:51,340
Other technologies such as digital phenotyping can use a similar approach,

106
00:13:51,340 --> 00:14:01,870
but instead of collecting direct brain activity directly, they mine proxy information.

107
00:14:01,870 --> 00:14:06,280
So, for example, approaches, which has affective computing.

108
00:14:06,280 --> 00:14:14,590
Also for the motion, I can use face recognition technology to create statistical correlations

109
00:14:14,590 --> 00:14:21,700
between certain face expressions and certain mental states with an increasing,

110
00:14:21,700 --> 00:14:28,090
increasing degree of accuracy and social media platforms, as you all know, are attempting to do the same.

111
00:14:28,090 --> 00:14:39,250
So there are studies that, for example, use Instagram pictures as a predictive biomarkers of mood disorders such as depression.

112
00:14:39,250 --> 00:14:49,780
And the Cambridge Analytica scandal from 2018 was a very famous example of abusive data processing

113
00:14:49,780 --> 00:14:57,510
actors using social media data to ensure mental states for purposes such as psychographic profiling,

114
00:14:57,510 --> 00:15:03,030
so clustering people according to their psychological traits.

115
00:15:03,030 --> 00:15:09,900
So based on these trends, we'd probably argue that mind reading in the public mind,

116
00:15:09,900 --> 00:15:17,190
reading is more complex than we originally thought, and we can distinguish probably two different problems of mind reading,

117
00:15:17,190 --> 00:15:23,370
which using borrowing the terminology from Noam Chomsky stewardship of language,

118
00:15:23,370 --> 00:15:27,840
we would probably call my reading in the narrow sense, and I'm reading in the broad sense.

119
00:15:27,840 --> 00:15:29,440
So when reading a narrow sense,

120
00:15:29,440 --> 00:15:35,980
there's the possibility that at some point we will be able to decode the content of mental states from print reporting.

121
00:15:35,980 --> 00:15:47,910
So this is possible from my reading. And as we said, with studies like the ones and I mentioned before, we are slowly embarking on this project,

122
00:15:47,910 --> 00:15:57,030
but we are quite far from having a like a rich and complete understanding of the content of mental states.

123
00:15:57,030 --> 00:16:03,030
But what we are already experiencing today and what is already relevant from an ethical perspective,

124
00:16:03,030 --> 00:16:10,170
is the possibility of revealing privacy sensitive statistical correlations between certain obligations.

125
00:16:10,170 --> 00:16:21,790
European Union Road data. I think we will need access to personal.

126
00:16:21,790 --> 00:16:28,940
What's going to happen up there?

127
00:16:28,940 --> 00:16:38,720
Let's see. The laptop would survive and answer the mental states so.

128
00:16:38,720 --> 00:16:47,480
And the second problem is already very relevant today. So this two problems can also be worded by distinguishing truth.

129
00:16:47,480 --> 00:16:52,880
Typekit practises are now shifting from from that perspective to the normative domain.

130
00:16:52,880 --> 00:17:04,070
We can distinguish two problems of privacy the first one, we could label it as neutral privacy and the second one mental privacy.

131
00:17:04,070 --> 00:17:09,410
So what what? What's the difference? Let's try to frame this broadly.

132
00:17:09,410 --> 00:17:20,810
So privacy, as you all know, what is conceptualising in many ways about the most common and influential approach to

133
00:17:20,810 --> 00:17:27,200
conceptualising privacy is the one provided by Warren and Brandeis at the beginning of the year,

134
00:17:27,200 --> 00:17:33,410
at the end of the 19th century, who interpreted as a right to be alone.

135
00:17:33,410 --> 00:17:46,100
And a lot of research in legal philosophy as tried to determine what the actual meaning of this act of being, let alone.

136
00:17:46,100 --> 00:17:53,840
And one influential account is the one provided by both in the 1980s,

137
00:17:53,840 --> 00:18:01,280
according to which privacy is the condition of being protected from unwanted access by others,

138
00:18:01,280 --> 00:18:06,710
either physical access through personal information or attention.

139
00:18:06,710 --> 00:18:24,110
So and when personal information is invoked, so disturbing in in definition, then we have a sub type of subset of the subset of privacy,

140
00:18:24,110 --> 00:18:30,320
which is called information private, so that the privacy of private personal information would,

141
00:18:30,320 --> 00:18:39,770
you know, privacy and mental privacy fit in this picture? There are two types of information privacy, but they are not at this.

142
00:18:39,770 --> 00:18:47,900
Two subtypes have a significant area of overlap, but they are not separate entities.

143
00:18:47,900 --> 00:18:53,180
So it is possible to have mental privacy within your privacy.

144
00:18:53,180 --> 00:18:58,490
It is possible to have new privacy with government comprising both, and I would expand on this in your.

145
00:18:58,490 --> 00:19:07,730
Privacy can be defined as the protection of brain data also called neuro data, whereas mental privacy concerns the protection of mental information.

146
00:19:07,730 --> 00:19:15,710
Now distinguishing these two categories, I think, is very important because data and information are not the same thing,

147
00:19:15,710 --> 00:19:22,190
even though they're these two terms are usually are often used interchangeably in ordinary language.

148
00:19:22,190 --> 00:19:32,210
So brain data are the individual role measurements of human brain structure, activity or function that still need to be processed.

149
00:19:32,210 --> 00:19:39,440
Whereas mental information is the process an organised set of data that collectively carry

150
00:19:39,440 --> 00:19:46,260
a logical meaning so they have trust values and accurately reveal aspects of mental state,

151
00:19:46,260 --> 00:19:58,880
such as thoughts, memories and emotions. As I said earlier, these two sets are not super premium, so that's why they're presented here in a diagram.

152
00:19:58,880 --> 00:20:02,690
So not all your privacy challenges have to do with mental information,

153
00:20:02,690 --> 00:20:08,150
and not all mental privacy challenges have to do with brain data, so it should give you an example.

154
00:20:08,150 --> 00:20:16,760
There is plenty of potential studies about new imaging biomarkers of neurological neuropsychiatric conditions as

155
00:20:16,760 --> 00:20:28,100
schizophrenia happens to be one of those conditions for which we have quite a large number of imaging biomarker.

156
00:20:28,100 --> 00:20:37,970
But by decoding this biomarkers, we are not able to make any inference about the mental states of the people who have those biomarkers.

157
00:20:37,970 --> 00:20:46,820
So we definitely have our health privacy problem because we are revealing sensitive information about the person's health status.

158
00:20:46,820 --> 00:20:54,740
But you are not improving anything related to mental states and to the content of mental states.

159
00:20:54,740 --> 00:21:00,290
On the other hand, it is possible to have mental privacy without your privacy being invoked.

160
00:21:00,290 --> 00:21:08,270
This is the case, for example, of emotionally I would use our data sources such as natural language,

161
00:21:08,270 --> 00:21:18,500
eye tracking or facial expressions to detect and analyse affective mental states or emotions.

162
00:21:18,500 --> 00:21:26,880
So we can define your privacy as the right and ability to exercise control over their brain data.

163
00:21:26,880 --> 00:21:33,930
And as such, this qualifies as a subset of medical privacy, which is the privacy of medical data.

164
00:21:33,930 --> 00:21:39,390
No way did I have specific features that make them particularly interesting from a privacy perspective.

165
00:21:39,390 --> 00:21:45,210
One is that they are predictive of present and future health studies and behaviour.

166
00:21:45,210 --> 00:21:55,800
So if you have access to, for example, my neurological measurements that show a predisposition to develop early onset dementia.

167
00:21:55,800 --> 00:22:05,720
This is an information that can be quite problematic for me if it falls in the wrong hands, such as my health insurance provider or my employer.

168
00:22:05,720 --> 00:22:11,210
The high temperature resolution, meaning that certain techniques, such as elections,

169
00:22:11,210 --> 00:22:19,400
can monitor brain activity, real time, which is not common to all physiological measurements,

170
00:22:19,400 --> 00:22:26,750
and they are not read only for much brain data, always in a read and write format, so you can read brain information,

171
00:22:26,750 --> 00:22:33,980
but also right into the brain using techniques such as neurostimulation and neuromodulation.

172
00:22:33,980 --> 00:22:41,120
So potential risks for new privacy include the collection of redundant data during brain recordings,

173
00:22:41,120 --> 00:22:46,970
which is particularly challenging when it occurs outside of the clinical setting,

174
00:22:46,970 --> 00:22:57,290
such as neuro marketing studies, the repurposing and contextualisation of brain data collection of brain data on the weak and central genes.

175
00:22:57,290 --> 00:23:06,860
This is a very big problem with consuming newer technology applications, which are used typically without medical supervision and only consent.

176
00:23:06,860 --> 00:23:17,180
Regime is basically accepting the device terms of use just like any other software or hardware for a use,

177
00:23:17,180 --> 00:23:21,830
and then also the problem of malicious hacking of neurotechnology.

178
00:23:21,830 --> 00:23:29,030
So when someone's x a device to extract information or to sabotage the function of the device.

179
00:23:29,030 --> 00:23:33,710
This will also fall under this category.

180
00:23:33,710 --> 00:23:45,230
Regulating your privacy is quite a topical area now for new ethics and new law because of the and features of data that I've mentioned earlier.

181
00:23:45,230 --> 00:23:57,140
There are several approaches, so a lot of people are arguing that we should regulate brain data in order to enhance our bar of your privacy.

182
00:23:57,140 --> 00:24:02,990
One approach that I found interesting and I support to a large extent as being provided by

183
00:24:02,990 --> 00:24:09,470
IS the brand new ad blitz who argued that we should treat all brain data as health data.

184
00:24:09,470 --> 00:24:17,120
So basically expanding the category of brain data and even when brain data are collected outside of the clinical setting,

185
00:24:17,120 --> 00:24:20,870
such as with direct to consumer applications, they should.

186
00:24:20,870 --> 00:24:26,780
Still, the data collected in this way should still be considered health data and other approach,

187
00:24:26,780 --> 00:24:30,800
which is actually not necessarily in opposition to this.

188
00:24:30,800 --> 00:24:37,670
Actually, there are two things can be harmonised is treating branded as a special category of health data.

189
00:24:37,670 --> 00:24:43,100
So all the way to our health data, but with enough data, you can have multiple categories.

190
00:24:43,100 --> 00:24:44,870
As some of you probably already know,

191
00:24:44,870 --> 00:24:54,200
genetic data are considered special category of health data under the General Data Protection Regulation of the EU.

192
00:24:54,200 --> 00:25:00,140
And so brain data could become a similar special category of branding.

193
00:25:00,140 --> 00:25:01,960
Then there are also more radical approaches.

194
00:25:01,960 --> 00:25:12,500
So interestingly enough, the Chilean Senate has passed Neuroprotection Bill, which regulates brain data in the same way as human organs,

195
00:25:12,500 --> 00:25:19,070
which is an interesting innovation because it's the first time that data are regulated like tissue.

196
00:25:19,070 --> 00:25:26,060
So their approach is aimed at basically shutting down the implant market of consumer technology,

197
00:25:26,060 --> 00:25:31,310
so you can only donate your brain data voluntarily or you cannot monetise.

198
00:25:31,310 --> 00:25:38,410
That can only be an act of non monetizable donation.

199
00:25:38,410 --> 00:25:44,020
The positive news here is that regulating neuroplasticity,

200
00:25:44,020 --> 00:25:54,280
she can benefit a lot from technological solutions, so techniques to ensure data security and privacy,

201
00:25:54,280 --> 00:25:55,990
such as encryption,

202
00:25:55,990 --> 00:26:06,000
differential privacy and a secure multiparty computation and others are proving quite are proving to be quite successful in achieving this goal.

203
00:26:06,000 --> 00:26:10,710
Now, let's look at Edmonton privacy, because here things get a little trickier.

204
00:26:10,710 --> 00:26:19,740
So mental privacy can be defined as the right and ability of people to exercise control over their mental information.

205
00:26:19,740 --> 00:26:25,470
And as such, it can be defined as a right to mental self determination.

206
00:26:25,470 --> 00:26:32,520
And if we really look at the fundamental conceptual foundations of mission privacy,

207
00:26:32,520 --> 00:26:43,800
we see that this cognitive dimensions of exercising control about information about ourselves was actually at the centre of the legal,

208
00:26:43,800 --> 00:26:50,490
philosophical discourse we, before then, Europe acknowledges, became widely available.

209
00:26:50,490 --> 00:26:55,690
This is a quote by Freed's anatomy of values.

210
00:26:55,690 --> 00:27:06,040
And I think there is the argument that you make in this paper is that mental privacy is a fundamental prerequisite

211
00:27:06,040 --> 00:27:15,820
of information privacy because only by exercising control over one's own information about their mental states,

212
00:27:15,820 --> 00:27:22,510
it is possible in principle to exercise control over any other form of information.

213
00:27:22,510 --> 00:27:30,180
The inherent features of governmental information are different than enduring features of data.

214
00:27:30,180 --> 00:27:39,760
But what is, I think, very relevant from an ethical perspective is that mental information includes not only observable behaviour,

215
00:27:39,760 --> 00:27:48,700
but also on an action and unspoken speech. The second feature is that mental information may elude conscious control.

216
00:27:48,700 --> 00:27:56,650
So a lot of brain imaging techniques operate below the threshold of conscious awareness of the person.

217
00:27:56,650 --> 00:28:06,440
So it's very hard for the data subjects to determine for himself or herself which information should be a secret or not.

218
00:28:06,440 --> 00:28:12,910
And the same goes with the analogy base or other known neural approaches.

219
00:28:12,910 --> 00:28:19,860
And then again, that this information is not read only for much, but it can also be modified.

220
00:28:19,860 --> 00:28:32,640
Regulating mental privacy due to disappearing features is trickier, because it's very unlikely that data type based approaches can be effective.

221
00:28:32,640 --> 00:28:36,900
So let's assume we regulate all brain data as health data.

222
00:28:36,900 --> 00:28:43,050
There are still a plethora of other data sources that can be used to make inferences about mental states.

223
00:28:43,050 --> 00:28:55,530
So this will not really solve the problem. So it seems that it is advisable to shift from data based approach to inference based

224
00:28:55,530 --> 00:29:03,480
approach so that the normative criteria and here should not be the class of data that I use,

225
00:29:03,480 --> 00:29:08,400
but the kind of inferences that are derived from a certain data.

226
00:29:08,400 --> 00:29:17,940
So regardless of whether you are collecting voice recording or electroencephalogram data or facial images,

227
00:29:17,940 --> 00:29:21,900
the important thing is the kind of inferences that your model throws.

228
00:29:21,900 --> 00:29:26,760
So if your model is making predictive inferences, abutments state mental states,

229
00:29:26,760 --> 00:29:34,230
it is at a higher risk of violating mental privacy and has to be regulated accordingly.

230
00:29:34,230 --> 00:29:39,420
So the question should be not which data will be processed, but how.

231
00:29:39,420 --> 00:29:51,360
Examples of mental privacy, relevant cases again involve in sharing emotions from face recognition, in varying moods,

232
00:29:51,360 --> 00:30:01,280
from natural language processing and inserting cognitive function for neurofeedback or cognitive training apps.

233
00:30:01,280 --> 00:30:04,160
So concluding,

234
00:30:04,160 --> 00:30:17,510
why is mental privacy so based on the news and features and this normative prerequisites can or should be considered a fundamental right.

235
00:30:17,510 --> 00:30:21,500
I want to really stayed here.

236
00:30:21,500 --> 00:30:32,930
The previous quote by Burri, many of which highlights this intimate connexion between mental privacy and freedom of thought.

237
00:30:32,930 --> 00:30:40,810
So people in this account can exercise their freedom of thought only to the extent that they can conceal their thoughts,

238
00:30:40,810 --> 00:30:46,540
so only to the extent that they can exercise their mental privacy.

239
00:30:46,540 --> 00:30:58,640
And another I think you quote here is by Michael Lynch, who wrote a very important words on privacy and the threat to the self.

240
00:30:58,640 --> 00:31:06,980
And Lynch argued that the concept of privacy is intimately connected to what it is to be an anonymous person.

241
00:31:06,980 --> 00:31:15,470
And I find it funny that you mentioned this never makes examples related to neuroscience or A.I. in this

242
00:31:15,470 --> 00:31:23,060
book erodes that if I could telepathically read all your conscious and unconscious thoughts and feelings,

243
00:31:23,060 --> 00:31:27,140
then your existence as a distinct person would begin to shrink,

244
00:31:27,140 --> 00:31:36,860
and our relationship would be so lopsided that there might cease to be, at least to me, anything subjective about you.

245
00:31:36,860 --> 00:31:45,680
So trying to translate this into a set of normative requirements meant some

246
00:31:45,680 --> 00:31:52,640
privacy could be construed as a poor fundamental right for five main reasons.

247
00:31:52,640 --> 00:31:57,680
The first one is that it is a practise for which information privacy to work.

248
00:31:57,680 --> 00:32:00,440
So only by exercising control over your mental states,

249
00:32:00,440 --> 00:32:08,030
you can exercise control over the information that you externalise from your mental states onto the world.

250
00:32:08,030 --> 00:32:17,730
It is considered a subjective experience. So mental information is the only piece of information that we process that is entirely subjective.

251
00:32:17,730 --> 00:32:25,380
It is a precondition for the development and participation of person and personal autonomy as to the charges.

252
00:32:25,380 --> 00:32:29,640
And it is a prerequisite for freedom of thought because of course,

253
00:32:29,640 --> 00:32:36,990
unauthorised access to mental information increases the risk of covert a regional influence or manipulation.

254
00:32:36,990 --> 00:32:44,520
So the higher the chances of violation of mental privacy, the higher the chances of violations of mental integrity.

255
00:32:44,520 --> 00:32:56,720
And finally, as Reitman and other Abhijit mental privacy is necessary for the recognition of one's ownership of their physical and mental reality.

256
00:32:56,720 --> 00:33:03,290
I want to conclude with an open question, because this is something I do not have a fixed opinion about.

257
00:33:03,290 --> 00:33:06,860
I would be really curious to know about your thoughts on this.

258
00:33:06,860 --> 00:33:12,290
So should mental privacy considered a relative or an absolute rights?

259
00:33:12,290 --> 00:33:21,290
As you know, most rights are relative. They're not absolute. Very few rights are absolute, like the right not to be tortured.

260
00:33:21,290 --> 00:33:30,830
They do not admit any exception. Pope is actually argued that mental privacy should be an absolute right.

261
00:33:30,830 --> 00:33:35,510
So you argued that the skull should be designated as a domain of absolute privacy.

262
00:33:35,510 --> 00:33:43,250
No, no one should be able to probe an individual's mind against their will, which is not permitted with the court order,

263
00:33:43,250 --> 00:33:51,340
which not permitted for religious or national security purposes and no coercive use should be accepted.

264
00:33:51,340 --> 00:34:01,810
An interesting alternative is provided by science fiction in 1990 episodes of Star Trek,

265
00:34:01,810 --> 00:34:08,860
Captain Kirk is being informed that a dangerous fly as a fisherman who joined one of the groups that are

266
00:34:08,860 --> 00:34:15,910
using the enterprise and desperately wants to identify the intruder and to know more about his plans.

267
00:34:15,910 --> 00:34:22,690
And he appeals to one of the staff members who happen to have telepaths telepathic abilities.

268
00:34:22,690 --> 00:34:30,190
And so Kirk wants to read the minds of all the visitors in order to understand which one is the dangerous type.

269
00:34:30,190 --> 00:34:35,260
However, the captain is reminded by one of his assistants that, according to the law,

270
00:34:35,260 --> 00:34:43,690
the right to mention privacy is an inalienable right of all information citizens and should not be abrogated.

271
00:34:43,690 --> 00:34:48,880
But they also mention, of course, should not be abrogated without due process of law.

272
00:34:48,880 --> 00:35:00,670
So it seems that in in court proceedings, exceptions to the absolute nature of mental privacy can be admitted.

273
00:35:00,670 --> 00:35:10,000
And this is something I think that is very pertinent to our current scenario, especially new rollout.

274
00:35:10,000 --> 00:35:23,320
So should, for example, mental privacy violations be admitted under risks for national security or for public health risk and so on.

275
00:35:23,320 --> 00:35:28,090
And I think this is an open question to which I don't have an answer yet.

276
00:35:28,090 --> 00:35:32,850
We can discuss the pros and cons of each approach, but I haven't balanced them yet.

277
00:35:32,850 --> 00:35:40,477
So thank you very much.