1
00:00:00,450 --> 00:00:18,270
I. Well, good evening, ladies and gentlemen, members of Linacre and Reuben Colleges.

2
00:00:18,270 --> 00:00:23,790
What a pleasure it is to see you all here, even though we are a little bit packed in this evening.

3
00:00:23,790 --> 00:00:36,150
Fantastic that we are able to have the 2022 O.C. Tanner Lecture in person. The Tanner Lectures are now in their 44th year.

4
00:00:36,150 --> 00:00:44,640
They were established by American scholar, industrialist and philanthropist Obert Clark Tanner and in creating the lectures.

5
00:00:44,640 --> 00:00:52,530
Professor Tanner said "I hope these lectures will contribute to the intellectual and moral life of mankind.

6
00:00:52,530 --> 00:00:58,830
I see them simply as a search for a better understanding of human behaviour and human values."

7
00:00:58,830 --> 00:01:08,850
Appointment as a Tanner lecturer is recognition of an exceptionally distinguished and important career in the field of human values,

8
00:01:08,850 --> 00:01:13,350
and that is a description fully justified in the case of tonight's lecturer.

9
00:01:13,350 --> 00:01:18,480
Professor Rosalind Picard. Professor Picard is founder and director of the Affective

10
00:01:18,480 --> 00:01:24,120
Computing Research Group at MIT Media Lab and the founding faculty chair of MIT's

11
00:01:24,120 --> 00:01:33,330
Mind, Hand and Heart Initiative. She co-founded Affectiva Inc, providing Emotional A.I. Technology and Sympatico Inc,

12
00:01:33,330 --> 00:01:39,840
which designs and develops A.I. systems to monitor human health through wearable sensors.

13
00:01:39,840 --> 00:01:49,020
Professor Picard holds a Bachelor's degree in electrical engineering from Georgia Institute of Technology and Masters and Doctoral degrees from MIT,

14
00:01:49,020 --> 00:01:57,210
both in engineering and computer sciences. She became internationally known for constructing mathematical models for content based retrieval

15
00:01:57,210 --> 00:02:02,910
of images and for pioneering methods of automated search and annotation in digital video.

16
00:02:02,910 --> 00:02:14,210
Her book, Affective Computing, published in 1997, started a new field by that name and has become a classic in computer science.

17
00:02:14,210 --> 00:02:23,060
Three weeks ago here in Oxford, Dr. Demis Hassabis, founder of DeepMind, gave a special O.C. Tanner lecture on artificial intelligence.

18
00:02:23,060 --> 00:02:27,980
Dr. Hassabis is perhaps best known across the globe as the inventor of AlphaGo,

19
00:02:27,980 --> 00:02:34,840
a computer game which defeated the world Go champion Lee Sedol in 2016.

20
00:02:34,840 --> 00:02:42,550
In his lecture, Dr. Hassabis enumerated three things that he believed were fundamental for the development of artificial intelligence.

21
00:02:42,550 --> 00:02:48,460
Firstly, a problem with a massive combinatorial search space.

22
00:02:48,460 --> 00:02:53,920
Secondly, a clear metric with which performance could be evaluated.

23
00:02:53,920 --> 00:03:00,440
And thirdly, lots of data. Or perhaps effective simulation.

24
00:03:00,440 --> 00:03:10,520
He then described how the real challenge in developing AlphaGo, to beat the world champion, had been the development of the evaluative function.

25
00:03:10,520 --> 00:03:16,850
It proved impossible, he said, to write in an evaluative function for the game go.

26
00:03:16,850 --> 00:03:24,470
Indeed, he hypothesised that the world's leading go players do not meant do not mentally calculate all possible moves.

27
00:03:24,470 --> 00:03:35,310
Rather, they rely on intuition. His breakthrough had been in developing methods for allowing AlphaGo to be intuitive.

28
00:03:35,310 --> 00:03:42,360
Research into human intelligence has revealed that emotions play an essential part in decision making, in perception.

29
00:03:42,360 --> 00:03:48,330
In learning and more, that is, they influence the very mechanisms of rational thinking.

30
00:03:48,330 --> 00:03:54,060
According to Professor Picard, if we want computers to be genuinely intelligent,

31
00:03:54,060 --> 00:04:00,070
then we must give them the ability to recognise, understand and respond to emotion.

32
00:04:00,070 --> 00:04:06,820
Her lecture this evening will expand on that thesis and is called 'Emotion, A.I. and Human Values'.

33
00:04:06,820 --> 00:04:24,240
Please join me in welcoming Professor Picard to the stage. Thank you.

34
00:04:24,240 --> 00:04:32,420
It's a real honour for me to be here with all of you and to see a full room after the pandemic, especially.

35
00:04:32,420 --> 00:04:42,440
I'm going to talk about how we might build a future life with A.I., where the best human values are hopefully honoured and where humans flourish.

36
00:04:42,440 --> 00:04:51,050
I'm going to start with a story - in 1996 when I was very young new faculty member at MIT.

37
00:04:51,050 --> 00:04:59,660
I had mostly been working on computer vision and on the side I had written a paper that was unpublished called Affective Computing.

38
00:04:59,660 --> 00:05:04,760
Somebody had read it over in the A.I. lab, in Rod Brooks's zoo.

39
00:05:04,760 --> 00:05:13,490
The zoo was a very famous group of world renowned roboticists, Rod at the helm, and philosopher Daniel Dinh, a theologian on a first,

40
00:05:13,490 --> 00:05:23,010
and other experts from places like Oxford who were trying to build the world's most advanced humanoid robot.

41
00:05:23,010 --> 00:05:29,280
Instead of inviting me to give the normal lecture, they sat me in a circle.

42
00:05:29,280 --> 00:05:34,680
It almost felt a little like a firing squad, except that it was delightful. It was questions being fired.

43
00:05:34,680 --> 00:05:41,520
Why would you want to give computers emotion? Isn't that what makes people so stupid?

44
00:05:41,520 --> 00:05:51,480
Isn't that a really bad idea? And I went through the evidence for, you know, from neuroscience why we now thought that.

45
00:05:51,480 --> 00:05:57,150
Yes, of course, while extreme emotional states could lead to very stupid decisions and behaviour,

46
00:05:57,150 --> 00:06:05,460
we are always having affective states and they are always interacting with cognition and perception and behaviour and action selection.

47
00:06:05,460 --> 00:06:14,070
And if we could understand them and model them in intelligent ways, maybe it would solve several of the problems that I and I elucidate.

48
00:06:14,070 --> 00:06:21,690
I listed those problems and said I thought that we needed to take a look at this.

49
00:06:21,690 --> 00:06:27,090
Well, by the end of this gathering, they were, I will say, more positive but still sceptical.

50
00:06:27,090 --> 00:06:34,050
Like it still seems impossible to me or I don't know. And as I left, one person who had been silent came up at the end.

51
00:06:34,050 --> 00:06:44,030
It was the theologian and she caught me as I was leaving the room with a question that has stuck with me since, she said.

52
00:06:44,030 --> 00:06:52,750
Isn't emotion the last thing that separates machines from humans?

53
00:06:52,750 --> 00:06:57,010
I gave a very short answer, which I'll touch on at the end of my talk.

54
00:06:57,010 --> 00:07:03,700
A one word answer. But today, I want to go a bit deeper into these topics, first,

55
00:07:03,700 --> 00:07:10,330
I'm going to start with some examples of why we build this and some of the

56
00:07:10,330 --> 00:07:18,250
perspectives now on this and then move into more of the human values towards the end.

57
00:07:18,250 --> 00:07:26,740
The paper turned into a book, and I want to highlight here that expressing and recognising human emotion is just a part of affective computing.

58
00:07:26,740 --> 00:07:33,820
There are many other parts that involve shaping algorithms to actually incorporate mechanisms inspired by human and animal emotion.

59
00:07:33,820 --> 00:07:37,720
And I'm not going into all of that today, although that is also an important part.

60
00:07:37,720 --> 00:07:45,320
I'm going to focus on some examples from interaction with humans, which involves expressing and recognising.

61
00:07:45,320 --> 00:07:52,430
Now, people naturally express emotion to computers, and also, I should say for the younger people today,

62
00:07:52,430 --> 00:07:56,270
they sometimes look at me like that's not a computer, that's a vacuum cleaner.

63
00:07:56,270 --> 00:08:01,850
When we think of computing, I think of it as anything with computing in it.

64
00:08:01,850 --> 00:08:11,510
That can be a robot, a software agent, a smartphone smartwatch, all kinds of devices that involve computation that we can embed in our computers.

65
00:08:11,510 --> 00:08:17,240
So affective computing was thinking not about any particular form.

66
00:08:17,240 --> 00:08:21,230
Now we express emotion to computers. In fact, about the time I started this,

67
00:08:21,230 --> 00:08:29,450
even twenty five percent of people under age twenty four admitted to having kicked their computer out of frustration.

68
00:08:29,450 --> 00:08:34,160
There was lots of expressing. There was almost nothing being done about it.

69
00:08:34,160 --> 00:08:38,570
Here's my favourite early example the first social robot built by Cynthia Brazil

70
00:08:38,570 --> 00:08:44,510
in Rod Brooks's zoo several years after my visit there that I love the example,

71
00:08:44,510 --> 00:08:55,030
not just because kismet is recognising from the tone of speech some affective states, but watch how kismet responds.

72
00:08:55,030 --> 00:09:01,570
So another important skill for the robot to be able to learn from people is being able to recognise communicative intent.

73
00:09:01,570 --> 00:09:09,040
Very good kismet. And the way we've done that with Casben right now is to have the robot recognise by tone of voice.

74
00:09:09,040 --> 00:09:19,520
Are you praising it? Are you scolding it? Where do you put your body? No, no, no.

75
00:09:19,520 --> 00:09:25,850
You're not to do that. No.

76
00:09:25,850 --> 00:09:36,470
So instead of just yelling at your machine or being frustrated, it could actually respond in ways that make you perhaps like it more.

77
00:09:36,470 --> 00:09:41,480
You can imagine how industry is now very interested in this with all of the speaking

78
00:09:41,480 --> 00:09:48,950
devices that try to provide support and services really to sell us things.

79
00:09:48,950 --> 00:09:55,310
You ask Alexa for Alexa this lately, but what's the right way to store Walmart?

80
00:09:55,310 --> 00:10:01,400
Alexa responds The right way to spell Walmart is Alma.

81
00:10:01,400 --> 00:10:06,650
That's not what you asked. So, you know, that was not right.

82
00:10:06,650 --> 00:10:13,280
Alexa doesn't have ears to put back. Alexa doesn't look humble and submissive, but Alexa could.

83
00:10:13,280 --> 00:10:23,540
The internal machine learning right could take that response as a signal to adjust what it's done to hopefully do a better job in the future.

84
00:10:23,540 --> 00:10:27,200
It could learn that in a personalised way, it could learn it and a culture dependent way.

85
00:10:27,200 --> 00:10:36,220
There are lots of ways the machine learning loop taking feedback from a person could be used to improve the AI technology.

86
00:10:36,220 --> 00:10:47,380
Now, while I was working in my lab on computer vision and trying to teach a computer to recognise emotions expressed on somebody's space,

87
00:10:47,380 --> 00:10:54,160
which isn't exactly what they're feeling, I'll say more about that in a moment.

88
00:10:54,160 --> 00:11:01,990
A young man came by my office to borrow a physical map. This is before digital maps, and he paused and he said, Hey, what are you working on?

89
00:11:01,990 --> 00:11:07,990
And I told him I was trying to teach computers to do a better job of understanding people's affective expressions.

90
00:11:07,990 --> 00:11:12,940
And he said, Could you help my brother? And I said, Tell me about your brother.

91
00:11:12,940 --> 00:11:16,000
And he started telling about his brother with autism.

92
00:11:16,000 --> 00:11:24,700
And for those of you who don't know many people with autism, not all can have a real challenging time interpreting kind of reading between the lines,

93
00:11:24,700 --> 00:11:31,740
the non-verbal signals, especially what's happening on your face while simultaneously trying to process speech.

94
00:11:31,740 --> 00:11:36,900
I realised that the tools we were building to teach the machine how to do this better could

95
00:11:36,900 --> 00:11:42,480
actually be very slightly adapted to help people and wouldn't that be a wonderful thing?

96
00:11:42,480 --> 00:11:49,710
So we started working on that and fast forward lots of work by our team, including Ron Paul,

97
00:11:49,710 --> 00:11:54,480
you who came over as a postdoc and co-founded Affectiva with me and later commercialised.

98
00:11:54,480 --> 00:12:00,480
There is now very robust and accurate interpret not only tracking of facial muscle movements,

99
00:12:00,480 --> 00:12:05,550
but mapping of those to the labels that people would give to those facial expressions,

100
00:12:05,550 --> 00:12:10,350
including doing groups of people, mapping them to cute little emojis for fun.

101
00:12:10,350 --> 00:12:18,640
But also now this is embedded in robots and other devices to help them interact a bit more smoothly with people.

102
00:12:18,640 --> 00:12:21,550
We didn't just build one tool to help people in the autism spectrum,

103
00:12:21,550 --> 00:12:28,210
we gave away these tools that other people could build into different kinds of learning environments, including things like Google Glass.

104
00:12:28,210 --> 00:12:33,310
We had built early glasses with cameras on them that could look at your face and see.

105
00:12:33,310 --> 00:12:37,330
This one was particularly useful for professors. If I was talking too much and you were looking bored.

106
00:12:37,330 --> 00:12:45,730
I would get a red light to tell me to stop and give you the floor so we could look at if you showed a facial expression of confusion,

107
00:12:45,730 --> 00:12:57,180
interest, boredom, it was like a non-verbal traffic signal. Now, we also were trying to implement things we had read in the scientific literature.

108
00:12:57,180 --> 00:13:00,030
So this became a way to test some of the science in real life.

109
00:13:00,030 --> 00:13:06,180
And one of the things that surprised us right away was that the science I had been told and wrote about in my book.

110
00:13:06,180 --> 00:13:11,200
And you can if you have my book, you can press this part out now and fix it.

111
00:13:11,200 --> 00:13:15,820
I wrote what I had been told by several neurologists that there was a true smile of happiness.

112
00:13:15,820 --> 00:13:22,480
It took a different pathway in the brain and it involved the eye crinkle, the cheek rays as well as the lip corner.

113
00:13:22,480 --> 00:13:28,810
And so if you saw that the person was truly happy was the way the dogma went.

114
00:13:28,810 --> 00:13:36,470
Now, it had been reported that this had appeared in some preschoolers using educational technology, and I was making observations.

115
00:13:36,470 --> 00:13:38,560
And so we ran a long story.

116
00:13:38,560 --> 00:13:46,120
Short ran a randomised, controlled trial study we do to look at if this would actually happen in cases where people were not delighted.

117
00:13:46,120 --> 00:13:54,430
I made a play for you a clip where we've designed an experiment where people are frustrated, they're putting their web information,

118
00:13:54,430 --> 00:14:27,860
their resume online and unbeknown to them, we're using the kind of captures that you can't fill out properly.

119
00:14:27,860 --> 00:14:32,990
I am sorry. You seem to be having difficulties getting through this form.

120
00:14:32,990 --> 00:14:39,140
Would you mind looking at the camera and share your thoughts on how to improve the form this?

121
00:14:39,140 --> 00:14:52,550
I'm stuck. It's important to give the context, because in the next person participating here,

122
00:14:52,550 --> 00:15:01,990
he's experiencing the exact same situation, but you'll see a very different and direct response.

123
00:15:01,990 --> 00:15:17,740
On. You know. That was not the true smile of delight.

124
00:15:17,740 --> 00:15:25,450
However, when you show people not only the still images of these which they rate 50 50, they can't guess which is to light and which is frustration.

125
00:15:25,450 --> 00:15:31,030
But even the videos we find that we can build a computer algorithm does better

126
00:15:31,030 --> 00:15:36,280
than people at recognising which video clips are from people in the situation.

127
00:15:36,280 --> 00:15:40,420
I just showed you where their smile is elicited with frustration versus people.

128
00:15:40,420 --> 00:15:48,160
In another situation I didn't show you where they're watching. Giggling baby videos and even grumpy students usually start off with a giggling baby.

129
00:15:48,160 --> 00:15:54,230
VIDEO Not really smiling, but by the end of it, they're they're usually cracking the light smiles.

130
00:15:54,230 --> 00:15:57,830
Now, this does not mean that computers are better than people at recognising motion.

131
00:15:57,830 --> 00:16:04,760
This is a very narrowly defined recognition task in a very limited situation where it's

132
00:16:04,760 --> 00:16:10,670
a probability of being one of two classes and there's no context being interpreted here.

133
00:16:10,670 --> 00:16:22,190
When humans are given the context and the whole situation, they're much better at discerning what's going on.

134
00:16:22,190 --> 00:16:32,930
Now, when we take the ability of the machine to read facial vocal and additional information, we can put that together in a lot of different ways.

135
00:16:32,930 --> 00:16:38,530
Some of the ways people are proposing involve things like just replacing people.

136
00:16:38,530 --> 00:16:45,280
And I'll show an example of that later. I want to show you a very different example here that's more about trying to help people,

137
00:16:45,280 --> 00:16:51,250
and we started by wanting to help people on the autism spectrum for this particular Ph.D. we were running out of time,

138
00:16:51,250 --> 00:16:57,010
so we decided we would just help MIT students who could use help with some of these skills as well.

139
00:16:57,010 --> 00:17:00,550
Pretty much everybody can. When people talk to each other,

140
00:17:00,550 --> 00:17:06,460
the majority of the information that's conveyed comes from the way we see things rather than the words we're actually saying.

141
00:17:06,460 --> 00:17:10,690
Eye contact smiles. Voice modulation speaking rate pauses.

142
00:17:10,690 --> 00:17:15,700
An emphasis on certain words often add an extra layer of information in our interactions.

143
00:17:15,700 --> 00:17:20,200
Many of us want to improve these interaction skills, but don't have the resources to do so.

144
00:17:20,200 --> 00:17:25,720
Imagine if you could practise your interaction skills with an automated system in the privacy of your own living room.

145
00:17:25,720 --> 00:17:29,450
A programme designed at the MIT Media Lab lets you do just that.

146
00:17:29,450 --> 00:17:37,870
Hi, I'm Mary. I'm looking forward to doing your interview. My automated conversation coach consists of a 3D character on a computer screen that

147
00:17:37,870 --> 00:17:41,890
can see here and make its own decisions based on its interaction with a person,

148
00:17:41,890 --> 00:17:47,650
and it works on a personal computer. Now, let's get started using a webcam.

149
00:17:47,650 --> 00:17:50,020
The system can analyse facial expressions.

150
00:17:50,020 --> 00:17:57,130
For example, it can measure where in the interaction you are smiling and can recognise your head gestures, such as a nod or a shake.

151
00:17:57,130 --> 00:18:02,950
The system also analyses your voice. It not only understands what you say, but how you say it.

152
00:18:02,950 --> 00:18:11,590
Using real time speech recognition and prosody analysis, it can capture the nonverbal nuances of conversations and display it in an intuitive format.

153
00:18:11,590 --> 00:18:15,490
When you're done, it gives you a summary of the information when you smiled,

154
00:18:15,490 --> 00:18:20,740
how fast you spoke and so on, and it can show how these measures change over multiple sessions.

155
00:18:20,740 --> 00:18:27,220
It even allows you to watch the video of your interactions with various measures of your behaviour displayed alongside the video,

156
00:18:27,220 --> 00:18:32,770
such as when you smile, how the volume of your voice rises and falls, and what words you emphasise.

157
00:18:32,770 --> 00:18:37,030
It even shows when your attention wanders. Can't find you?

158
00:18:37,030 --> 00:18:38,080
There you are.

159
00:18:38,080 --> 00:18:46,270
You were saying in a study with 90 MIT undergraduates, the subjects went through simulated job interviews before and after receiving this training.

160
00:18:46,270 --> 00:18:52,870
Those who got the feedback from this automated system were rated as better candidates for the job than those who did not.

161
00:18:52,870 --> 00:18:57,610
Besides job interviews, the researchers say the system could help with public speaking,

162
00:18:57,610 --> 00:19:11,050
dating, learning languages or helping people who have difficulties in social communications.

163
00:19:11,050 --> 00:19:14,020
Now, you might notice that was a little bit dated.

164
00:19:14,020 --> 00:19:20,500
One can make much more human looking avatars now, which you might want to do if you want to intimidate people a little bit more.

165
00:19:20,500 --> 00:19:30,820
In most of our work, we've deliberately chosen not to do that so as not to deceive people about the abilities of the A.I.

166
00:19:30,820 --> 00:19:37,660
When I was working with people on the autism spectrum and trying to understand more about what they really needed,

167
00:19:37,660 --> 00:19:43,990
at one point, one of them, a woman who had become a friend who was not speaking, was typing.

168
00:19:43,990 --> 00:19:48,460
And she communicated to me that I had it all wrong.

169
00:19:48,460 --> 00:19:52,450
She says, Roz, you're what you're doing is all wrong.

170
00:19:52,450 --> 00:20:00,070
I was like, Oh, great, you know, what am I doing wrong? And she said, You're not recognising my emotions properly.

171
00:20:00,070 --> 00:20:03,760
And it's I don't need help recognising other people's emotions.

172
00:20:03,760 --> 00:20:06,400
I need you to recognise my emotions.

173
00:20:06,400 --> 00:20:12,430
And I realised the literature at all focussed on what people on the spectrum couldn't do, not on what we couldn't do for them, right?

174
00:20:12,430 --> 00:20:17,740
How we were not pulling up our end of the bargain.

175
00:20:17,740 --> 00:20:20,710
I said, Gee, I feel terrible. What am I not recognising?

176
00:20:20,710 --> 00:20:28,330
And she said, it's not just you, it's everybody's not recognising our emotions and in particular, they're not recognising our stress.

177
00:20:28,330 --> 00:20:35,680
I realised in the lab we had built physiological measures. These had actually been well known for some time in the physiology literature

178
00:20:35,680 --> 00:20:41,740
that corresponded to sympathetic and parasympathetic nervous system stress. And what was missing was the ability to take them out of the lab.

179
00:20:41,740 --> 00:20:50,860
And so we started to build sensors that could embed this physiological measurements, such as the electrical activity, skin conductance.

180
00:20:50,860 --> 00:20:55,720
You know, the notion where your palms get sweaty goes up with perspiration.

181
00:20:55,720 --> 00:21:01,570
But it turns out it also goes up even when your hands don't feel sweaty with very subtle changes in stress.

182
00:21:01,570 --> 00:21:04,960
In fact, it's one of the key signals still used in lie detection,

183
00:21:04,960 --> 00:21:12,760
although it's not an accurate lie detector on its own requires, again, a lot of contextual interpretation.

184
00:21:12,760 --> 00:21:16,240
So we built this sweatband here with sensors in it.

185
00:21:16,240 --> 00:21:18,430
There's another version of the sweatband.

186
00:21:18,430 --> 00:21:26,710
This particular version and an NBA version were borrowed by a student one day coming into my office right before winter break.

187
00:21:26,710 --> 00:21:29,350
Professor Picard, could I please borrow a couple of your sensors?

188
00:21:29,350 --> 00:21:33,130
My little brother has autism, he can't talk, and I want to see what's stressing him out.

189
00:21:33,130 --> 00:21:37,600
I said, Sure. In fact, don't just take one. Take two. They often break. You need to borrow a soldering iron.

190
00:21:37,600 --> 00:21:44,850
He said, No, I have a soldering iron. Great and mighty student can fix them, so he takes the sensors home.

191
00:21:44,850 --> 00:21:51,480
Puts them on his little brother. I'm looking at the data back at MIT and I notice that the boys data was not.

192
00:21:51,480 --> 00:21:56,490
It was usually pretty calm looking, but there was one gigantic peak that I couldn't explain.

193
00:21:56,490 --> 00:22:02,070
In fact, it was on one wrist and not the other, and I couldn't figure out, how can you be stressed on one side of your body and not the other?

194
00:22:02,070 --> 00:22:06,780
So I figured out the sensors must be broken, but I tried all kinds of things to replicate it.

195
00:22:06,780 --> 00:22:14,760
I'm an electrical engineer by training, and nothing replicated the weird data, so I resolved to very old fashioned debugging.

196
00:22:14,760 --> 00:22:18,870
I picked up the phone. I called the student at home on vacation.

197
00:22:18,870 --> 00:22:23,950
Hi, how's your vacation? How's your little brother? Hey, any idea what happened to him?

198
00:22:23,950 --> 00:22:31,080
And I gave the exact date and time in the data and he said, I don't know. I'll check the diary diary and it just keeps a diary help.

199
00:22:31,080 --> 00:22:35,940
Like, what are the odds that some young man has written this stuff, right? He comes back.

200
00:22:35,940 --> 00:22:41,430
He has the exact record. And he says that was right before he had a grand mal seizure.

201
00:22:41,430 --> 00:22:49,380
That led me to connect with Boston Children's Hospital doctor Joe Madison and colleagues and start measuring this

202
00:22:49,380 --> 00:22:59,280
electro dermal activity data in the epilepsy monitoring unit where we could get gold standard 24-7 video EEG,

203
00:22:59,280 --> 00:23:08,460
electrocardiogram, electro dermal activity. And we found that 100 percent of the grand mal seizures we got had this significant more than two

204
00:23:08,460 --> 00:23:14,160
standard deviations above the pre seizure period skin conductance response that together with movement,

205
00:23:14,160 --> 00:23:24,840
we're able to Ming's or work build a machine learning algorithm that could run light in a wearable device and reliably detect grand mal seizures.

206
00:23:24,840 --> 00:23:27,360
Now, it was not able to forecast them at that time.

207
00:23:27,360 --> 00:23:33,900
There has just recently been some very nice work done by the Mayo Clinic using other devices that have come out of our lab for devices,

208
00:23:33,900 --> 00:23:45,350
and they are doing forecasting now using these same data that are used here together with some additional data in our devices.

209
00:23:45,350 --> 00:23:48,380
I didn't know anything about epilepsy when I started this, you know,

210
00:23:48,380 --> 00:23:54,920
I was already this sort of computer vision person become this weird affective computing research person and now I'm,

211
00:23:54,920 --> 00:24:01,490
you know, wasn't really that interested in learning about epilepsy until I started learning more about what was happening.

212
00:24:01,490 --> 00:24:09,430
I'm curious how many of you know somebody with epilepsy? Well, I get many more hands up in the EU than in the US.

213
00:24:09,430 --> 00:24:14,830
One in 26 people in America will have epilepsy. Those of you who did not raise your hand if you go and ask your network,

214
00:24:14,830 --> 00:24:20,800
you probably do know somebody and they just haven't told you how many of you have heard of SUDEP.

215
00:24:20,800 --> 00:24:23,960
Sudden, unexpected death in epilepsy. All right.

216
00:24:23,960 --> 00:24:33,350
Far fewer, this needs to be communicated more sudden crib death, sudden infant death syndrome takes fewer lives every year in the US than SUDEP.

217
00:24:33,350 --> 00:24:39,110
And yet, you know, everybody's a baby at some point. But one in twenty four will have epilepsy.

218
00:24:39,110 --> 00:24:45,560
There's death from this particular unknown cause of death in epilepsy every seven to nine minutes,

219
00:24:45,560 --> 00:24:49,970
and tragically, there will be several during this talk.

220
00:24:49,970 --> 00:24:55,040
But we now know in the last five years we've been learning that it's deadliest if you're alone,

221
00:24:55,040 --> 00:24:59,150
that in fact, the one of the key mechanisms is you have a seizure.

222
00:24:59,150 --> 00:25:03,800
This unusual electro activity, we think it spreads to a part of the brain that can turn off your breathing.

223
00:25:03,800 --> 00:25:09,320
But the breathing can be turned back on. You're still capable of breathing if somebody stimulates you.

224
00:25:09,320 --> 00:25:15,270
So not in a hundred percent of cases, but in many cases when you stimulate the person, they will start breathing again.

225
00:25:15,270 --> 00:25:20,180
And this could be why it is so much less likely that there's a SUDEP when somebody is there,

226
00:25:20,180 --> 00:25:24,590
because the first thing they usually do is they turn you on their side and ask, if you're OK.

227
00:25:24,590 --> 00:25:34,360
And they stimulate you. The hard work of the team at Adam had to go, it was to commercialise this, I have enormous respect now for every company,

228
00:25:34,360 --> 00:25:41,210
all the work they do to take something from a research prototype to something that not only works but can get through FDA approval.

229
00:25:41,210 --> 00:25:51,880
I see medical was easy compared to FDA, and this is now out there being used to help alert with air running on board,

230
00:25:51,880 --> 00:26:00,850
to help alert, to bring a person to help. The first time I got a mail like this, I turned white and I think my skin conductance went through the roof.

231
00:26:00,850 --> 00:26:09,220
The mom described that she was in the shower. Her phone was connected to her daughter's smartwatch, detecting seizures.

232
00:26:09,220 --> 00:26:13,540
She put the phone on the counter, saw it go off while she was in the shower,

233
00:26:13,540 --> 00:26:20,560
ran out of the shower to her daughter's bedroom, found her face down in bed, blue and not breathing.

234
00:26:20,560 --> 00:26:29,530
Flipped her over, stimulated her and her daughter took a breath and another breath, and turned normal colour again.

235
00:26:29,530 --> 00:26:35,710
The family wants to be shown here, they are celebrating one of many happy occasions.

236
00:26:35,710 --> 00:26:41,140
Unfortunately, I've talked to lots of families who were never told about the possibility of SUDEP.

237
00:26:41,140 --> 00:26:47,320
So I'm going to ask all of you if you remember anything when you leave here today to make sure that you talk about

238
00:26:47,320 --> 00:26:53,650
this because the medical professionals that are learning about it are now recommending officially to talk about it,

239
00:26:53,650 --> 00:27:00,700
but they're about 10 years behind and doing so. And now we know that letting people know that seizures can be deadly and that they should have

240
00:27:00,700 --> 00:27:09,250
somebody there who might be able to help them with first aid or stimulation could save their life.

241
00:27:09,250 --> 00:27:15,240
The. As I was reflecting on this and somebody was saying, Oh, this device saved a life, I might know the device didn't save a life.

242
00:27:15,240 --> 00:27:26,930
It was a person who got there who saved the life. The device was helping connect people, and I said, the power is when it connects us to each other.

243
00:27:26,930 --> 00:27:30,230
But you might join me in thinking of some counter examples of that, right,

244
00:27:30,230 --> 00:27:37,610
like the how many of you seen the social dilemma know this kind of take down of lots of the social media industry groups like Facebook,

245
00:27:37,610 --> 00:27:41,840
where the focus has been on connecting people? Right?

246
00:27:41,840 --> 00:27:52,440
So it doesn't seem like the power is to connect us, but it's the power to connect us in ways that help people flourish.

247
00:27:52,440 --> 00:27:58,240
And this is like another dimension that I think has been neglected in a lot of what we've built.

248
00:27:58,240 --> 00:28:04,370
And I perhaps in part because it's really been hard to measure.

249
00:28:04,370 --> 00:28:14,670
Now launching these devices, you don't need to read all the small text led to a lot of really cool science findings that I wasn't expecting.

250
00:28:14,670 --> 00:28:19,310
I thought I was going on a detour with epilepsy. You know, I'm ever going to get back to affective computing.

251
00:28:19,310 --> 00:28:25,580
Well, it turns out we ran into the regions of the brain most likely involved in SUDEP that included things like the amygdala,

252
00:28:25,580 --> 00:28:35,210
which are the core regions we've been studying for emotion, core regions involved and in anxiety and depression and mood disorders,

253
00:28:35,210 --> 00:28:39,440
where suddenly what the epilepsy tells us we're working with had stuck direct electrodes.

254
00:28:39,440 --> 00:28:43,600
And so now we could actually monitor directly those regions of the brain.

255
00:28:43,600 --> 00:28:48,410
We have to wait for an artificial scanning environment joint with our other signals.

256
00:28:48,410 --> 00:28:55,700
We could get noninvasively from the body, and that was surprising and exciting.

257
00:28:55,700 --> 00:29:01,630
And I remembered something my boss at the Media Lab had asked before. Decades earlier.

258
00:29:01,630 --> 00:29:07,090
Back when I was just working on computer vision, I try to make machines smarter.

259
00:29:07,090 --> 00:29:16,360
He had said to me. Ross, when are you going to do something useful, like, OK, Walter, what do you want?

260
00:29:16,360 --> 00:29:25,340
He said, I need a mood ring that tells me my wife's mood before I go home.

261
00:29:25,340 --> 00:29:30,410
Now, at the time, I, you know, I grew up with mood rings, they were these stupid temperature sensors that changed colours,

262
00:29:30,410 --> 00:29:33,440
and somebody probably made a huge amount of money and probably I could save

263
00:29:33,440 --> 00:29:36,560
time not having to write research grants if I would do something like that.

264
00:29:36,560 --> 00:29:43,260
But at MIT, we like to do things that are hard and important, and temperature sensor was trivial.

265
00:29:43,260 --> 00:29:50,970
As I ran into these brain science findings about the connexions between these deep regions of the brain and what we were picking up on the periphery,

266
00:29:50,970 --> 00:29:55,830
and I started to learn about the growing problem in mental health.

267
00:29:55,830 --> 00:29:59,880
This data ends at twenty seventeen here.

268
00:29:59,880 --> 00:30:09,600
It has gone up, up, up with the pandemic. What you see in this graph from very nice work by Twiggy at all is the huge

269
00:30:09,600 --> 00:30:14,400
increase from ages 16 to twenty five in particular since the iPhone came out.

270
00:30:14,400 --> 00:30:19,320
I'm not saying that's the cause. We can't go back and run a causal stay there.

271
00:30:19,320 --> 00:30:25,320
However, there are causal studies with social media showing not just correlation studies now,

272
00:30:25,320 --> 00:30:34,170
but causal studies where you do a randomised controlled trial. You ask one group My study was done by hunted all that you can limit the use of social

273
00:30:34,170 --> 00:30:38,340
media to 10 minutes a day versus regular use and just monitor what you're doing,

274
00:30:38,340 --> 00:30:40,770
which itself leads to improvements.

275
00:30:40,770 --> 00:30:47,280
And the group that limited their use to 10 minutes a day showed significantly better improvements in mental health.

276
00:30:47,280 --> 00:30:56,370
So we know that behaviour is connected to mental health, and I could say a lot more about that, but we have to get onto other topics here.

277
00:30:56,370 --> 00:31:03,090
Briefly, what we can now do with the A.I. is collect data from lots of modalities, not just wearable sensors,

278
00:31:03,090 --> 00:31:08,310
not just what you're saying on your phone, who you're connected to regular surveys.

279
00:31:08,310 --> 00:31:13,260
We might ask you information about your sleep. Sleep regularity is looking very important.

280
00:31:13,260 --> 00:31:19,080
And more, and we put all of that together and we get gold standard labels from psychiatrists.

281
00:31:19,080 --> 00:31:29,250
In this case, what you see is across the bottom, the higher the number, the higher the psychiatrist rating is of your Hamilton depression scale.

282
00:31:29,250 --> 00:31:34,800
You're more depressed if you have a higher stress. So these numbers are given by psychiatrists.

283
00:31:34,800 --> 00:31:38,490
The numbers on the vertical axis are given by the passive data only.

284
00:31:38,490 --> 00:31:40,680
We're not having to interrupt and ask you any questions.

285
00:31:40,680 --> 00:31:47,670
Passive from your smartphone use in your wearable, plus one initial value from your first visit with the psychiatrist.

286
00:31:47,670 --> 00:31:52,500
The correlation here is as good as the doctors are with each other, and this was early work.

287
00:31:52,500 --> 00:31:59,370
Now since then, we've gotten data with that's even harder with more variability in it and the correlations vary.

288
00:31:59,370 --> 00:32:03,090
Here it was actually pretty hard to beat the patient baseline if they weren't changing a whole lot.

289
00:32:03,090 --> 00:32:07,830
Since then, we've got results that are beating patient baselines. Also,

290
00:32:07,830 --> 00:32:12,450
it's still an active area where we need to get a lot more data before we can start to really

291
00:32:12,450 --> 00:32:20,400
understand the different categories of these data and also what kinds of treatments might work best.

292
00:32:20,400 --> 00:32:25,650
We don't want to just treat people, though what we'd like to do is things that haven't been able to be done in medicine before.

293
00:32:25,650 --> 00:32:33,150
See, right now, what happens is if this vertical axis represents your well-being and hundreds of numbers are rolled up in here.

294
00:32:33,150 --> 00:32:39,000
By the way, this is my one concept diagram, not real. Data on the horizontal axis is time.

295
00:32:39,000 --> 00:32:45,660
Let's say that you're applying to schools, you get into Oxford, you're really happy, great positive wellbeing there.

296
00:32:45,660 --> 00:32:49,710
And then unfortunately, major stressors hit.

297
00:32:49,710 --> 00:32:53,190
They always do, and everybody takes a dip.

298
00:32:53,190 --> 00:33:03,000
And then we see the difference between the people who are exhibiting fragile behaviour versus who are resilient and can bounce back.

299
00:33:03,000 --> 00:33:08,280
And then if you're in trouble and especially if you're in trouble and you're down on the slope right right now,

300
00:33:08,280 --> 00:33:11,250
strongly encourage you to get help reach out further down.

301
00:33:11,250 --> 00:33:16,440
You are sometimes the harder it is to help, but there's hope you can always come back and recover.

302
00:33:16,440 --> 00:33:25,500
What we'd like to do is help people detect if they're on the red line as early as possible and also give personalised advice,

303
00:33:25,500 --> 00:33:31,270
evidence based help to move to the blue line. So that's active research right now.

304
00:33:31,270 --> 00:33:36,580
It tends to involve digital therapeutics and things that come to you and your phone.

305
00:33:36,580 --> 00:33:41,790
We've built games and other tools to try to support people.

306
00:33:41,790 --> 00:33:49,430
A big problem is that people don't stay engaged.

307
00:33:49,430 --> 00:33:57,320
So even like our latest game, which is more than twice as effective as the published average best digital mental

308
00:33:57,320 --> 00:34:03,110
health interventions and we're even in the top 15 percent of games by day 30,

309
00:34:03,110 --> 00:34:09,260
we're down to six point six percent of the people who got the app actually using it.

310
00:34:09,260 --> 00:34:18,320
And so if a therapy, you need to use a little bit of it every day, it's really hard to get that retention.

311
00:34:18,320 --> 00:34:21,860
Well, what about interactions with people?

312
00:34:21,860 --> 00:34:27,290
I'm reminded of when my next door neighbour said, I hate to exercise, but I'll show up for an appointment with a person.

313
00:34:27,290 --> 00:34:38,190
I don't want to let them down. There is evidence that relational agents will certainly people engage and can hold you longer for therapy.

314
00:34:38,190 --> 00:34:41,700
Relational agents also have evidence of engaging, and Tim Bickmore,

315
00:34:41,700 --> 00:34:45,990
who did his Ph.D. on this with me, has continued and built lots of nice work in this area.

316
00:34:45,990 --> 00:34:52,290
Here's a recent paper from September 2020 where the relational agent,

317
00:34:52,290 --> 00:34:59,700
even this very cartoony type face not pretending to be a real human, just being kind of a dummy.

318
00:34:59,700 --> 00:35:07,350
I was able to support women and improving their health.

319
00:35:07,350 --> 00:35:13,770
All right. What we see, though, an effort to engage and to, well, the cameras.

320
00:35:13,770 --> 00:35:17,940
And also because it's like climbing Everest, it's there and you want to see if you can do it.

321
00:35:17,940 --> 00:35:30,590
And for an eye person, we love a hard challenge. We see work like the example I'm going to show you next where I'll let it speak for itself here.

322
00:35:30,590 --> 00:35:38,270
The Hong Kong team behind celebrity humanoid robot Sophia is launching a new prototype Grace targeted at the

323
00:35:38,270 --> 00:35:46,610
health care market and designed to interact with the elderly and those isolated by the COVID 19 pandemic.

324
00:35:46,610 --> 00:35:51,470
Dressed in a blue nurse's uniform, Greece has Asian features, collar length,

325
00:35:51,470 --> 00:35:59,910
brown hair and a thermal camera in her chest to take your temperature and measure your responsiveness.

326
00:35:59,910 --> 00:36:09,090
She uses artificial intelligence to diagnose a patient and can speak English, Mandarin and Cantonese.

327
00:36:09,090 --> 00:36:15,360
Hello, everybody. I'm Sophia. Let me introduce you to my sister, Grace.

328
00:36:15,360 --> 00:36:21,360
Come over here. Hi, Grace. Thank you, Sophia.

329
00:36:21,360 --> 00:36:25,050
Hello, everybody. I am grace.

330
00:36:25,050 --> 00:36:32,560
I am built by Hanson Robotics for awakening how I can do all kinds of things for elderly people.

331
00:36:32,560 --> 00:36:38,620
I can visit with people and brighten their day with social stimulation, entertain and help guide exercise,

332
00:36:38,620 --> 00:36:46,150
but also can do talk therapy, take fire readings and help health care providers assess their health and deliver treatments.

333
00:36:46,150 --> 00:36:49,900
Hi, my friend. Nice to see you. OK, great.

334
00:36:49,900 --> 00:36:56,050
So while we talk, I will take your temperature reading and pulse with this little thermal camera on my chest.

335
00:36:56,050 --> 00:37:00,730
See, you are thirty six point six degrees Celsius, by the way.

336
00:37:00,730 --> 00:37:06,430
No risk of COVID. Also, I note that you are responsive and aware, which is good,

337
00:37:06,430 --> 00:37:12,850
and they may share that important data back to the doctors and nurses at the institution.

338
00:37:12,850 --> 00:37:18,250
Resemblance to a health care professional and capacity for social interaction is aimed

339
00:37:18,250 --> 00:37:23,440
at relieving the burden of frontline hospital staff overran during the pandemic,

340
00:37:23,440 --> 00:37:31,670
said founder David Hanson. So he designed Grace to look professional and health care setting.

341
00:37:31,670 --> 00:37:38,450
So that would mean so that she would resemble the sort of human like appearance of health care stock.

342
00:37:38,450 --> 00:37:42,020
And so then this would facilitate more natural interactions.

343
00:37:42,020 --> 00:37:50,750
A human like appearance facilitates trust and facilitates natural engagement because we're wired for a human face to face interactions.

344
00:37:50,750 --> 00:37:52,760
That's just the way that human beings are.

345
00:37:52,760 --> 00:37:59,750
And so giving her a face that would be familiar to people comforting to people and also language abilities that are comforting.

346
00:37:59,750 --> 00:38:06,860
So we now have Mandarin and Cantonese capabilities for the Grace and Sophia robot platforms.

347
00:38:06,860 --> 00:38:10,520
Hey, Grace, why don't you look at my face? Thank you.

348
00:38:10,520 --> 00:38:18,650
And on, I'm going to show you a smile. And let's let's see what your smile looks like.

349
00:38:18,650 --> 00:38:33,750
OK. Yeah. What are sad? Oh, yes, reacting a little sad if I feel pain and then maybe like imagine a big loud noise happens bear, so on.

350
00:38:33,750 --> 00:38:44,540
So with the grace robot, we are simulating over 48 major muscles in the face, so we have 36 motors.

351
00:38:44,540 --> 00:38:48,950
I'm going to pause there. You're invited to go online and watch the whole thing.

352
00:38:48,950 --> 00:38:57,120
It is an engineering marvel, the kind of AI and robotics that goes into making something like this.

353
00:38:57,120 --> 00:39:03,100
How many of you feel really positive and excited about this direction?

354
00:39:03,100 --> 00:39:08,890
Some, OK. All right. How many of you feel kind of unnerved or worried about this direction?

355
00:39:08,890 --> 00:39:19,030
All right, that has the majority here. I want to just point out for some of you who who may be less familiar with how the technology works,

356
00:39:19,030 --> 00:39:24,820
that while some of the air is a bit like a chat bot and some of the demos you'll find online with these robots,

357
00:39:24,820 --> 00:39:31,360
they are largely scripted when you see these really sort of most impressive presentations.

358
00:39:31,360 --> 00:39:34,510
They are not conscious despite the little or alive,

359
00:39:34,510 --> 00:39:45,490
despite the ability to imitate some movements while they display expressions on their face and with gestures, and can do that with tone of voice.

360
00:39:45,490 --> 00:39:53,470
They do not have feelings, and they will, I think, be with us, whether you like them or not.

361
00:39:53,470 --> 00:39:56,680
You know, you might say you hate these things or you like these things.

362
00:39:56,680 --> 00:40:00,250
They're just, there's this flow that I think they're not going to go away.

363
00:40:00,250 --> 00:40:08,440
There is a kind of an enthralling challenge, and these are just going to keep being designed with abilities inspired by people.

364
00:40:08,440 --> 00:40:17,530
It's my prediction. Now, an even older version built by Hiroshi Ishiguro was built to look like him.

365
00:40:17,530 --> 00:40:22,660
His original idea this was even before the pandemic was to send the robot off to give his talk.

366
00:40:22,660 --> 00:40:25,060
So like, I could be my avatar up here,

367
00:40:25,060 --> 00:40:33,340
I could be sitting back in Boston controlling it right and kind of look like I'm there wouldn't be anywhere near as much fun as being here in person.

368
00:40:33,340 --> 00:40:38,470
But as we all now know from all of our Zoom meetings and so forth.

369
00:40:38,470 --> 00:40:42,300
But let's go back to this question.

370
00:40:42,300 --> 00:40:54,660
If we could make it look and act as if the emotions were human in this, is this the last thing that separates humans from machines?

371
00:40:54,660 --> 00:41:00,450
And I've maintained from the beginning that we don't know how to do consciousness or feelings.

372
00:41:00,450 --> 00:41:05,050
And my answer to the theologian at the zoo was No.

373
00:41:05,050 --> 00:41:11,170
And we could discuss more detail later, but some of the key elements are we can make it look lifelike,

374
00:41:11,170 --> 00:41:17,080
but it is not alive nor conscious, and many other things are missing.

375
00:41:17,080 --> 00:41:27,630
But it's not just about these functions that are missing. And this is where I want to transition a little bit to some thoughts about human values.

376
00:41:27,630 --> 00:41:36,800
I think that we are much more than our functions. We are much more than what we do.

377
00:41:36,800 --> 00:41:42,740
Now, there are probably people here who are experts in ontology, and there are at least two kinds of ontology here.

378
00:41:42,740 --> 00:41:52,850
This is not the linguistic category kind, but the branch of metaphysics that deals with the nature of being.

379
00:41:52,850 --> 00:41:57,290
Now again, I'm an engineer. My doctorate is in science, not philosophy.

380
00:41:57,290 --> 00:42:04,100
So I'm going to appeal to some more famous philosophers for a moment here for a very quick summary.

381
00:42:04,100 --> 00:42:08,960
Socrates on a famous T-shirt here to do is to be Plato.

382
00:42:08,960 --> 00:42:16,250
To be is to do. And my favourite Frank Sinatra, W.E.B. Du.

383
00:42:16,250 --> 00:42:28,580
Obviously a great oversimplification, but there is a lot of wonderful discussion out there about the difference between doing and being.

384
00:42:28,580 --> 00:42:42,390
My message for you is that I think we are onto logically equivalent to whom we get.

385
00:42:42,390 --> 00:42:47,580
I have three sons. They are vastly better than me at some things.

386
00:42:47,580 --> 00:42:55,140
If you just compare functions, you may find that someone you get has a much greater function than you at something.

387
00:42:55,140 --> 00:43:04,220
You may have a greater function than them at something. But we are onto logically equal.

388
00:43:04,220 --> 00:43:10,160
What we make is different, even if it looks like us and has a function like us,

389
00:43:10,160 --> 00:43:24,680
if we talk about this kind of ontology, then we are onto logically superior to these things we make.

390
00:43:24,680 --> 00:43:34,910
And building A.I. to help humans flourish, I think we need a higher standard of expressing human worth than abilities or functions.

391
00:43:34,910 --> 00:43:39,630
I'm going to propose some criteria. I'm going to propose a higher standard.

392
00:43:39,630 --> 00:43:46,260
Some criteria include acknowledging the equal worth of every living person.

393
00:43:46,260 --> 00:43:59,400
We do see increasing emphasis, certainly in the US, and I'm seeing a bit more globally of trying to reduce different kinds of discrimination.

394
00:43:59,400 --> 00:44:03,840
I think we also need independence of functions or abilities, and as I get older and older,

395
00:44:03,840 --> 00:44:09,210
I appreciate this one more right, because our abilities change over time.

396
00:44:09,210 --> 00:44:16,830
And certainly, if you're on the operating table, your abilities have changed for a moment, but you shouldn't lose your human worth.

397
00:44:16,830 --> 00:44:31,260
I don't think the standard should be limited by materialism, naturalism and scientism is somewhat myopic ways of describing the world.

398
00:44:31,260 --> 00:44:38,040
I also don't think it should be something that could be instated or removed by somebody in power over you,

399
00:44:38,040 --> 00:44:42,460
a person or a government and power over you.

400
00:44:42,460 --> 00:44:54,940
And last, the easiest criteria is it should be bigger than some professor's idea, bigger than anything somebody like me would just propose.

401
00:44:54,940 --> 00:45:07,220
The concept that I propose meets all of these criteria is the concept known in Latin as Imago De.

402
00:45:07,220 --> 00:45:13,830
For fun, I asked one of the most popular guys right now, GPT three, what is Imago Day?

403
00:45:13,830 --> 00:45:18,800
It's kind of small. I typed in What is my good day? I click Submit.

404
00:45:18,800 --> 00:45:24,890
And it says the Imago Day is the biblical concept that humans are created in the image of God.

405
00:45:24,890 --> 00:45:33,960
This means that humans have the ability to think, feel and create just as God does.

406
00:45:33,960 --> 00:45:38,910
Note, it also elicited a trigger warning, sensitive content.

407
00:45:38,910 --> 00:45:44,790
I think I need to make that clear for this talk as well.

408
00:45:44,790 --> 00:45:54,220
Note also that the second line that the eye has pulled from the Web here and coming up with its answer is focussed only on function.

409
00:45:54,220 --> 00:46:02,500
OK, it's going on things we do not on our state of being.

410
00:46:02,500 --> 00:46:05,980
Now, the Margot is literally Latin for image,

411
00:46:05,980 --> 00:46:14,440
and any of you with an engineering or simple maths background know that we can project down a high dimensional object here,

412
00:46:14,440 --> 00:46:21,940
we just show a 3D object down to a two dimensional image. But we could also deal with objects.

413
00:46:21,940 --> 00:46:25,060
We do this in machine learning all the time with thousands of dimensions,

414
00:46:25,060 --> 00:46:29,590
millions of dimensions, billions of dimensions and projected down to just a few.

415
00:46:29,590 --> 00:46:39,230
So that the image itself is actually leaving out most of what the original is, that it is the image of.

416
00:46:39,230 --> 00:46:43,910
So I want to be clear, that image is not visual similarity.

417
00:46:43,910 --> 00:46:48,460
This is not us looking like God.

418
00:46:48,460 --> 00:46:54,660
It's some kind of image that we may not fully understand.

419
00:46:54,660 --> 00:47:04,260
Being here where Tolkien lived and was a professor, I want to point out, too, that we can make things that may have like imago humano, right?

420
00:47:04,260 --> 00:47:12,510
Or here, Tolkien's made a character, Bilbo Baggins, that has some features like the its maker.

421
00:47:12,510 --> 00:47:26,410
It is a Orser, like its maker. The qualities of the Imago Day are due do not give arrogance and superiority their humility.

422
00:47:26,410 --> 00:47:33,240
Their mercy. Love, grace and forgiveness.

423
00:47:33,240 --> 00:47:39,930
There is a quality this is a free gift given to all, not given more to those who have greater functions.

424
00:47:39,930 --> 00:47:58,580
And it is an estimable worth. C.S. Lewis, another local great, wrote, There are no ordinary people you have never talked to a mere mortal.

425
00:47:58,580 --> 00:48:02,300
Now, some of you who are not religious and I used to be an atheist,

426
00:48:02,300 --> 00:48:08,540
so I sympathise with you will say, is it exclusive to people who are religious right?

427
00:48:08,540 --> 00:48:15,200
This is taught in Genesis and in Jewish traditions and Christian traditions and some Muslim traditions.

428
00:48:15,200 --> 00:48:19,520
No, it is not restricted to people who practise any of these faiths.

429
00:48:19,520 --> 00:48:30,350
It is a concept given to everyone. It's interesting, too, to look at lived experience of people who maybe are not religious.

430
00:48:30,350 --> 00:48:35,270
I was intrigued by an account written an interview with Peter Singer,

431
00:48:35,270 --> 00:48:46,800
the great philosopher and proponent of utilitarianism talking about how when his mother got Alzheimer's and had serious dementia.

432
00:48:46,800 --> 00:48:56,970
His teachings of how to treat her, he was confronted with those, and in fact, she was like, Yeah, you know, you don't have to take care of me now.

433
00:48:56,970 --> 00:49:06,840
You could follow your teachings, which were basically that she didn't have the kind of valuable human life that she had before.

434
00:49:06,840 --> 00:49:13,170
And yet when it came to those final days, he chose to treat her as if she had a good day,

435
00:49:13,170 --> 00:49:21,330
as if she had dignity against what his logic and his teachings had been saying to everyone.

436
00:49:21,330 --> 00:49:28,890
In doing so, I think he also showed Imago Day.

437
00:49:28,890 --> 00:49:38,420
It's interesting, too, to look at things like the Universal Declaration of Human Rights affirmed by more than 50 states in 1948.

438
00:49:38,420 --> 00:49:44,420
Assuming all human beings are born free and equal in dignity and rights,

439
00:49:44,420 --> 00:49:52,520
they are endowed with reason and conscience and should act towards one another in a spirit of brotherhood.

440
00:49:52,520 --> 00:49:56,180
So hopefully we're not going to get much dissension on this.

441
00:49:56,180 --> 00:50:07,670
But how many of you at the start of your Ph.D. or your research have thought about this aspect while you're designing your work?

442
00:50:07,670 --> 00:50:14,700
I know when I began as a Ph.D. student, I was just looking for something cool and interesting and hard, right?

443
00:50:14,700 --> 00:50:22,170
When you're at a place like Oxford or MIT. You want to do something super hard, super impressive, something people say that can't be done.

444
00:50:22,170 --> 00:50:28,240
Some other people might just be over there like just enough. I just want to get my degree right.

445
00:50:28,240 --> 00:50:30,940
People want to get published, they want to get on with their career.

446
00:50:30,940 --> 00:50:40,250
And so the optimality criteria is pretty much like drive it as far as you can to the right and still graduate in a reasonable amount of time.

447
00:50:40,250 --> 00:50:45,680
Thinking about Imago Day, I think, calls us not just to a higher standard about what is human worth,

448
00:50:45,680 --> 00:50:51,080
but calls us to, I hope, a higher standard when we're choosing problems to work on.

449
00:50:51,080 --> 00:51:01,010
And I. Can they simply try to picture it this way that we add an extra dimension and of course, this makes optimisation harder?

450
00:51:01,010 --> 00:51:10,700
But our optimality criteria could aim for the upper right here.

451
00:51:10,700 --> 00:51:14,270
At the MIT Media Lab, we've been challenging ourselves to try to do this,

452
00:51:14,270 --> 00:51:19,220
which means we have to learn a lot about things that aren't just engineering and science.

453
00:51:19,220 --> 00:51:24,230
We're bringing the principles of science and engineering to understanding people and what helps people flourish.

454
00:51:24,230 --> 00:51:28,760
We've borrowed from people who've focussed on flourishing and positive psychology here.

455
00:51:28,760 --> 00:51:34,700
From them, we've learnt, for example, that it's not enough to just try to make people happy, right?

456
00:51:34,700 --> 00:51:38,300
Happiness is just one of many emotions. All of the emotions are valuable.

457
00:51:38,300 --> 00:51:42,470
It's very important supplies to be sad. Your friends are sad, be sad with them.

458
00:51:42,470 --> 00:51:45,470
There are, in fact, the most important.

459
00:51:45,470 --> 00:51:53,870
One of the pillars I've seen described in components of flourishing is probably the one of relationships not being focussed on yourself,

460
00:51:53,870 --> 00:52:00,380
but being able to enter into a relationship where you're focussed on another and developing that healthy relationship.

461
00:52:00,380 --> 00:52:05,120
During the pandemic, the suffered quite a bit and is probably the thing to prioritise.

462
00:52:05,120 --> 00:52:10,550
In particular, if you're struggling with mental health, call a call an old friend at places like Oxford.

463
00:52:10,550 --> 00:52:16,640
We don't have to worry about engagement and achievement much because people are hyper achievers and usually love what they're doing and are engaged.

464
00:52:16,640 --> 00:52:26,750
But these are valuable contributors to human flourishing, and meaning has not been addressed, probably as much as it should in this literature.

465
00:52:26,750 --> 00:52:31,610
Usually, things like religion are stuffed in under meaning as something very important.

466
00:52:31,610 --> 00:52:38,960
In fact, there's some really nice causal inference that's been done on large studies of nurses.

467
00:52:38,960 --> 00:52:50,380
Work by Tyler Vandewalle at Harvard, showing that people with regular religious practise have a lot of better health outcomes.

468
00:52:50,380 --> 00:52:58,360
This particular theory I've shown here, PR in a is attributed to Marty Seligman.

469
00:52:58,360 --> 00:53:05,050
All right, let me wrap up here a quick recap of what we've been through quite a journey here.

470
00:53:05,050 --> 00:53:12,190
A.I. is acquiring increasingly skills that look like it's developing emotional intelligence.

471
00:53:12,190 --> 00:53:19,390
We still have a long way to go. I'm not. I'm not confident we're gonna be able to solve some of the problems, especially with consciousness,

472
00:53:19,390 --> 00:53:27,140
but we can make it look pretty good and that, to some people, is enough.

473
00:53:27,140 --> 00:53:37,250
It will increasingly take on human like appearance, and we will see this in ways that may be very unnerving and disturbing.

474
00:53:37,250 --> 00:53:42,020
They may fail in the business sense. They may succeed in some narrow areas.

475
00:53:42,020 --> 00:53:47,240
There certainly are some areas where it appears to be succeeding.

476
00:53:47,240 --> 00:53:52,430
I think as we reflect, though, more on what it means to be human and on our human values,

477
00:53:52,430 --> 00:54:00,160
we've got to recognise we are not simply our abilities and our functions.

478
00:54:00,160 --> 00:54:11,880
I think we should start leading the community and thinking more about the higher values, the higher human values and looking for a higher standard.

479
00:54:11,880 --> 00:54:23,070
The Imago Day is a higher standard, and it invites those of us shaping A.I. to recognise that people are more than what they do,

480
00:54:23,070 --> 00:54:29,230
that we can look at these other aspects and help people try to flourish.

481
00:54:29,230 --> 00:54:39,580
And finally, this future requires a lot more than I experts in order to understand what it is to help people flourish and to craft it.

482
00:54:39,580 --> 00:54:47,320
We really need to spend time understanding one another and we need all kinds of people, not just engineers and scientists, artists,

483
00:54:47,320 --> 00:54:58,150
designers, entertainers, all kinds of people who understand people to join not just in the vision, but through every stage of this process.

484
00:54:58,150 --> 00:55:04,390
Even new business models, I think, need to be developed. New regulations probably need to be developed.

485
00:55:04,390 --> 00:55:11,590
Everyone in society, I think, needs to to speak out and have their voice be a part of the future that we will build so that

486
00:55:11,590 --> 00:55:19,930
hopefully we build a future where human life with A.I. is one where all humans can flourish.

487
00:55:19,930 --> 00:55:40,340
Thank you very much. Super, well I now invited two colleagues from the university to respond to Rosalind's lecture.

488
00:55:40,340 --> 00:55:46,190
And our first respondent is the face that would be very familiar to many in the audience this afternoon.

489
00:55:46,190 --> 00:55:55,970
Professor Lionel Tarasenko is an engineer and also a leading expert in the application of signal processing and machine learning to health care.

490
00:55:55,970 --> 00:56:04,460
He was previously Head of the Department of Engineering, and in addition to his academic roles here at the university,

491
00:56:04,460 --> 00:56:10,490
he's also a pro vice chancellor and first president of Reuben College. Lionel,

492
00:56:10,490 --> 00:56:16,310
would you like to begin? Thank you very much. Apologies for my voice.

493
00:56:16,310 --> 00:56:24,740
Nothing to do with COVID. I do assure you. I have the evidence on my smartphone to show a negative lateral flow test.

494
00:56:24,740 --> 00:56:26,660
It's an impossible task to respond.

495
00:56:26,660 --> 00:56:35,210
And so you all agree because we've not had one kind of lecture, but we've had about three Tanner Lectures and they've all been equally wonderful.

496
00:56:35,210 --> 00:56:44,810
So I'm going to attempt to perhaps see if there are links between each of the Tanner Lectures which you gave tonight.

497
00:56:44,810 --> 00:56:53,270
These wonderful ways of exploring some really important aspects of research done in

498
00:56:53,270 --> 00:56:58,100
places like MIT, in this universities and many other leading universities,

499
00:56:58,100 --> 00:57:01,490
and indeed many of the Big Tech companies.

500
00:57:01,490 --> 00:57:09,200
You started by showing us how a computer vision could interpret human emotions, and the first thing I would say,

501
00:57:09,200 --> 00:57:19,610
which is very important, what Professor Picard showed is that even if you are analysing human interaction, using speech.

502
00:57:19,610 --> 00:57:31,870
Speech is not enough. And what you need is multimodal ways of analysing the interaction where we are talking to each other.

503
00:57:31,870 --> 00:57:38,520
So in addition to the speech, really to duplicate the way human beings communicate,

504
00:57:38,520 --> 00:57:46,080
we need only to understand through auditory cortex the meaning of the pressure waves impinging upon them.

505
00:57:46,080 --> 00:57:52,240
But we need, as you should, very clearly. And that could you rose from Professor Picard.

506
00:57:52,240 --> 00:57:56,430
Roz. Roz, right? And that facial movements.

507
00:57:56,430 --> 00:58:06,930
It was delightful to see what the true smile of delight looks like, what the look of frustration looks like, and that is very important.

508
00:58:06,930 --> 00:58:12,900
And then you added a third modality. So speech, facial movements from computer vision algorithms.

509
00:58:12,900 --> 00:58:19,830
The third one was stress levels through recording electro dermal activity.

510
00:58:19,830 --> 00:58:21,900
And then the second thing that you showed us,

511
00:58:21,900 --> 00:58:32,220
which I think is very important for all the graduate students here from Linacre and Reuben is. Roz was very honest with us.

512
00:58:32,220 --> 00:58:42,180
Scientific research for those of us who have got our best used behind us, I include myself in that and is not a linear process.

513
00:58:42,180 --> 00:58:44,700
There's a huge amount of, serendipity.

514
00:58:44,700 --> 00:58:53,670
And it was wonderful to see what the leading scientists in the field of AI acknowledged that that many of the things you've done,

515
00:58:53,670 --> 00:58:59,130
which have been some of your best papers and your best work have been serendipitous detours.

516
00:58:59,130 --> 00:59:03,300
And you showed us that with epilepsy and SUDEP,

517
00:59:03,300 --> 00:59:12,990
'Sudden Unexpected Death In Epilepsy', and from the work that you've been doing previously, including the electro dermal monitoring.

518
00:59:12,990 --> 00:59:21,060
And I would argue that for those of you who still at the graduate school level, yes, you can have an idea where you're going to be going.

519
00:59:21,060 --> 00:59:25,200
But please don't miss these details. They're very important.

520
00:59:25,200 --> 00:59:30,480
They may lead you to the best work that you'll ever do. And in this case, I would argue it is.

521
00:59:30,480 --> 00:59:36,540
Some of the work that Roz, Professor Picard, is well known throughout the world,

522
00:59:36,540 --> 00:59:44,190
and it had a huge impact on a subset of patients and improving potentially their outcomes.

523
00:59:44,190 --> 00:59:53,040
And you showed that as well. We wearable devices and again, the principle of using multimodal data through mobile surveys, wearable data,

524
00:59:53,040 --> 01:00:00,270
monitoring activity, monitoring sleep and so on, is perhaps opening a door to understanding depression.

525
01:00:00,270 --> 01:00:06,360
So multi modality, serendipity, and thirdly, absolutely love this,

526
01:00:06,360 --> 01:00:15,900
and there are a few of my own graduate students or some of the post-docs now in theatres around the audience and working on monitoring,

527
01:00:15,900 --> 01:00:21,360
improving monitoring wearables, improving monitoring through videos that we do in my lab as well.

528
01:00:21,360 --> 01:00:26,460
I always ask my students and post-doc: great, you can monitor this.

529
01:00:26,460 --> 01:00:32,040
What are you going to do with our ability to monitor this? And I ask my student the same question. You monitor people's activities.

530
01:00:32,040 --> 01:00:37,440
How do we turn that into an intervention? How do we go? That's the question I always ask my students.

531
01:00:37,440 --> 01:00:39,000
Fantastic. Great results.

532
01:00:39,000 --> 01:00:48,720
How are you going to use that to effect to close the loop, if you will, to have an intervention that will improve the life of someone?

533
01:00:48,720 --> 01:00:53,610
And there was a great picture then showing that as you try to do that,

534
01:00:53,610 --> 01:00:58,170
you discover that actually the monitoring heart as it is, is the easiest part of it.

535
01:00:58,170 --> 01:01:06,540
An intervention that lasts and continues to have a positive effect on a patient and so on, is extremely hard to achieve.

536
01:01:06,540 --> 01:01:14,160
And that part of the talk was really very clever. I've never seen this link before, because what you said and it's absolutely true,

537
01:01:14,160 --> 01:01:20,590
is to maintain the resilience to long term engagement with intervention.

538
01:01:20,590 --> 01:01:25,720
And we need to find ways of doing this,

539
01:01:25,720 --> 01:01:37,330
and you showed us Grace, and Grace was potentially as a human like robot the way in the health care context that she might be deployed,

540
01:01:37,330 --> 01:01:41,770
the way to maintain the long term engagement.

541
01:01:41,770 --> 01:01:49,090
But at that point, this was where I said there were multiple lectures and there were these, I think, your third lecture.

542
01:01:49,090 --> 01:01:55,780
And is it sufficient to be a human like Grace or do we need a real Grace?

543
01:01:55,780 --> 01:02:02,790
In that was what is the difference between a human like Grace and a human Grace, I don't know if any people in the audience called Grace?

544
01:02:02,790 --> 01:02:12,520
And but that's what I'm trying to draw the differences, and that's where you really spent the third lecture.

545
01:02:12,520 --> 01:02:16,450
On Imago Dei. How do we decide to pronaunce? As a vice chancellor,

546
01:02:16,450 --> 01:02:22,360
you have to think very hard about how you pronounce Latin, and the answer is nobody knows.

547
01:02:22,360 --> 01:02:26,020
And you do what you think is right. My wife's a mediaeval historian.

548
01:02:26,020 --> 01:02:30,820
I've talked about her, talked about it with her several times, and there isn't a set way.

549
01:02:30,820 --> 01:02:36,580
So you pronounce it just right, even in Oxford. Imago Dei.

550
01:02:36,580 --> 01:02:47,350
And what you attempted to do, and I think this is very hard, is to explain the concept of image in the way that the image of God,

551
01:02:47,350 --> 01:02:57,640
I believe, is to be understood. If you, if the picture is of a multidimensional God, how is that projected then to two or three dimensions?

552
01:02:57,640 --> 01:03:09,940
And the key idea here is if God is higher being, or however you want to describe it, is in n dimensions where 'n' is a large number.

553
01:03:09,940 --> 01:03:16,600
The projection then, and you because using a flat wall, you had to use two, even down to three,

554
01:03:16,600 --> 01:03:23,710
isn't simply the idea of removing n minus 2 or n minus 3, to get down to two or three.

555
01:03:23,710 --> 01:03:29,800
It is actually combining all end and forming that mapping down to two or three.

556
01:03:29,800 --> 01:03:33,700
And that is really difficult to understand, but it is incredibly important.

557
01:03:33,700 --> 01:03:39,160
It is something that if you've ever done data visualisation and use machine learning algorithms,

558
01:03:39,160 --> 01:03:48,310
you will understand it's something that we do in computer science. How do we use a very high dimensional space then to 2D or 3D?

559
01:03:48,310 --> 01:03:53,410
We do not delete n minus 3 or n minus 2, we construct mappings.

560
01:03:53,410 --> 01:04:01,420
And that's what the images. So the image, even though is limited to two or three dimensions, has aspects of all end dimensions.

561
01:04:01,420 --> 01:04:07,750
And I thought that was a brilliant way for us to think about what Imago Dei might be.

562
01:04:07,750 --> 01:04:20,840
And. What I think you're trying to show, then, is as a result of this image of God, it is a difference between the human like Grace and the human Grace.

563
01:04:20,840 --> 01:04:27,570
And I would then link the third with possibly the second Tanner Lecture that you gave us.

564
01:04:27,570 --> 01:04:34,400
And I would argue it is because we are in that image that we do take the serendipitous detours.

565
01:04:34,400 --> 01:04:40,790
It is an aspect of human behaviour to take those serendipitous detours.

566
01:04:40,790 --> 01:04:46,880
And would I agree that the best A.I. algorithm would not take the serendipitous detour because

567
01:04:46,880 --> 01:04:53,420
it would not optimise an objective criterion that the A. I. algorithm has been designed for?

568
01:04:53,420 --> 01:05:01,250
And that's where I would leave it. And I rejoice that as human beings, we do take those serendipitous detours and many thanks

569
01:05:01,250 --> 01:05:08,140
Professor Picard, Roz, for showing us how that is really intrinsic to what it is to be a human being.

570
01:05:08,140 --> 01:05:24,370
Thank you. Our second respondent this evening is Tom Fletcher, Tom is the Principal of Hertford College.

571
01:05:24,370 --> 01:05:36,040
But between 2007 and 2011, he was foreign policy adviser to three UK prime ministers and subsequently the UK's ambassador to Lebanon.

572
01:05:36,040 --> 01:05:36,850
More recently,

573
01:05:36,850 --> 01:05:45,660
he's been visiting professor at New York University and has chaired the International Advisory Council on the Creative Industries Federation.

574
01:05:45,660 --> 01:05:49,690
Tom. Thank you so much, Nick. And thank you Prof. Picard, Roz, and Lionel.

575
01:05:49,690 --> 01:05:55,600
What a, what a feast, really. It's like a Lebanese meal, as an ex-ambasador of Lebanon,

576
01:05:55,600 --> 01:05:59,670
We have all the dishes in front of you, and you're wondering which one to put your pita bread into.

577
01:05:59,670 --> 01:06:04,220
And I'm sure there'll be lots of opportunities to ask questions in just, in just a moment.

578
01:06:04,220 --> 01:06:10,030
Your, your story of the mood ring reminded me of a friend of mine who was part of a trilingual household,

579
01:06:10,030 --> 01:06:13,060
and he didn't have the mood ring to decipher his partner's moods.

580
01:06:13,060 --> 01:06:19,870
But because they were trilingual, he knew that as he walked in the door, the mood was determined by the language that his partner used.

581
01:06:19,870 --> 01:06:25,570
And so if he walked in and heard Guten Abend, he was in trouble. If he walked in and heard

582
01:06:25,570 --> 01:06:33,130
Good evening, he knew it's very practical evening, and if he walked in and heard Bonsoir, well, I mean, I'll leave it.

583
01:06:33,130 --> 01:06:43,990
I wouldn't linger on that one. You really, I think, challenged us and I come into this space as a non-scientist and as a non-academic.

584
01:06:43,990 --> 01:06:53,980
I'm a recovering ambassador as Nick said. But I think you've challenged us as Lionel finished with, this thought,

585
01:06:53,980 --> 01:07:00,610
with this idea that in working out and trying to understand what makes, what can make machines more human.

586
01:07:00,610 --> 01:07:08,890
Are we able to better understand what actually makes us human? And I have a mischievous thought and three challenges for my world.

587
01:07:08,890 --> 01:07:13,780
If what we used to think it was maps and chaps, you know, protocol and flags and statecraft.

588
01:07:13,780 --> 01:07:25,510
My mischievous thought. Watching your avatars and Grace and the others is, how far are we from being able to automate an Oxford Head of House?

589
01:07:25,510 --> 01:07:31,030
I say that in the presence of at least four, I'm not sure what the collective noun is for principal.

590
01:07:31,030 --> 01:07:34,930
Someone said it was a lack of principle, but I wonder, you know,

591
01:07:34,930 --> 01:07:41,710
but in trying to answer that question, could you replace the four of us with with Grace?

592
01:07:41,710 --> 01:07:47,170
I think that already helps us to start to understand where it is. We might add value as humans.

593
01:07:47,170 --> 01:07:50,900
Many, I think many governing bodies would quite, would be quite attracted to

594
01:07:50,900 --> 01:07:57,940
the idea of replacing the Head of House. Inevitably, at this stage of time, many Heads of House would be quite attracted to the idea as well.

595
01:07:57,940 --> 01:08:02,950
But it does allow us to focus on where do we really add value as humans?

596
01:08:02,950 --> 01:08:08,590
And that's what really struck me from from your extraordinary lecture.

597
01:08:08,590 --> 01:08:17,560
The three challenges for my old world of diplomacy and statecraft and and politics.

598
01:08:17,560 --> 01:08:24,730
Firstly, how do we renew our education systems in response to the challenges you've set us?

599
01:08:24,730 --> 01:08:28,360
How do we ensure that we are educating first class humans,

600
01:08:28,360 --> 01:08:35,650
not second class robots, amongst the students and academics who we're developing at Oxford and, and globally?

601
01:08:35,650 --> 01:08:44,550
Are we able to reset education to renew education in a way that focuses on the hand and the heart, and not just the head?

602
01:08:44,550 --> 01:08:50,910
Well, discipline of history, we tended to learn the list of the wars that we happen, we happen to have won.

603
01:08:50,910 --> 01:08:54,780
Are we able to learn how we co-existed between those wars?

604
01:08:54,780 --> 01:09:04,980
Are we able for the hand, for the skills, to think about what we've done on global competence, which is of social, emotional learning, really?

605
01:09:04,980 --> 01:09:10,110
Many people call it 21st century skills, in my world we call it diplomacy. Can you understand your own filter?

606
01:09:10,110 --> 01:09:13,740
The Instagram filter through which you see the world and understand that other people

607
01:09:13,740 --> 01:09:19,050
have a different filter to yours and maybe yours might not be the right one. Key skill,

608
01:09:19,050 --> 01:09:24,220
but we don't, we don't really teach that, and we certainly don't assess it.

609
01:09:24,220 --> 01:09:28,510
We'll focus always on the things that can be tested and, and memorised.

610
01:09:28,510 --> 01:09:31,000
So could we learn that that skill of global competence,

611
01:09:31,000 --> 01:09:41,920
the global antenna that mean that you could be dropped in, in Bahrain or Singapore tomorrow or Boston tomorrow and adapt, ask the right questions.

612
01:09:41,920 --> 01:09:52,030
Listen. Show the empathy that you described. I think we are well overdue that renewal of education and a focus also on the heart.

613
01:09:52,030 --> 01:09:56,780
Imago Dei, you know, what really makes us curious, kind and brave.

614
01:09:56,780 --> 01:10:04,790
I'd love to see that much more central to education, even at elite academic institutions like this.

615
01:10:04,790 --> 01:10:12,560
The second challenge, I think more traditional diplomacy, is who is going to do the peace process here between humans and technology?

616
01:10:12,560 --> 01:10:15,560
Maybe that is the Nobel peace prise of the future.

617
01:10:15,560 --> 01:10:22,160
I was involved four years ago, five years ago with a leading a report for the UN Secretary General on how we coexist with technology.

618
01:10:22,160 --> 01:10:26,120
40 recommendations, of which only one was adopted.

619
01:10:26,120 --> 01:10:31,040
And that was to create a new committee to think about it. Oh, well.

620
01:10:31,040 --> 01:10:35,300
But we, but we were looking at the real challenges that, you know 1815 Congress Vienna.

621
01:10:35,300 --> 01:10:38,090
It was clear who should be in the room and who convened the room,

622
01:10:38,090 --> 01:10:45,140
and now who decides who gets in the room to decide how we coexist with artificial intelligence, who has the authority to do that?

623
01:10:45,140 --> 01:10:50,230
The UN doesn't have the resource, the bandwidth, the strategic depth to do that.

624
01:10:50,230 --> 01:10:57,250
We had an exchange of letters with the UN Secretary General where 200 of us wrote in, Elon Musk and others, to say we are creating risks.

625
01:10:57,250 --> 01:11:03,190
This is the field of lethal autonomous weapons. We are creating these risks faster than we can actually manage them.

626
01:11:03,190 --> 01:11:11,540
What will you do about it? How will you update The Universal Declaration of Human Rights, that you quoted, in order to respond to those new challenges.

627
01:11:11,540 --> 01:11:15,260
Two years later, the UN is still arguing over who should draught the reply.

628
01:11:15,260 --> 01:11:19,820
You know, it's very, very difficult to see who brings that conversation together, and that,

629
01:11:19,820 --> 01:11:25,040
for me, is a real challenge for, for our fields, but also challenge for a university.

630
01:11:25,040 --> 01:11:32,960
Because if governments can't do that, if international organisations can't do that, then maybe this is a place where we can have those conversations,

631
01:11:32,960 --> 01:11:45,380
where we can convene the debates and the diplomacy that is necessary to update the rules of liberty and security for a digital age.

632
01:11:45,380 --> 01:11:50,810
And the third thought also in that world of peace and diplomacy,

633
01:11:50,810 --> 01:11:58,640
but really bringing it further on to how we renew society and a society in which humans flourish.

634
01:11:58,640 --> 01:12:02,780
When I arrived in the Middle East 10, 11 years ago, I used to go around saying something very silly,

635
01:12:02,780 --> 01:12:06,540
which was the most powerful weapon in the Middle East is the smartphone.

636
01:12:06,540 --> 01:12:13,130
You know, the Arab Spring was breaking out around us and we thought, we thought that this would actually help us connect in a really exciting,

637
01:12:13,130 --> 01:12:17,090
extraordinary new ways. And we were overoptimistic. And of course,

638
01:12:17,090 --> 01:12:24,590
what we learnt subsequently and what we have learnt in Ukraine is that actually there are more powerful weapons out there than the smart thing.

639
01:12:24,590 --> 01:12:31,700
But actually, your tool gave me fresh optimism that maybe this can help us find new ways to reach out,

640
01:12:31,700 --> 01:12:39,800
new ways to connect as humans, maybe A.I. can assist us with that effort to coexist with each other.

641
01:12:39,800 --> 01:12:45,400
I think that's an extraordinary piece of work to be done that brings together that work.

642
01:12:45,400 --> 01:12:50,590
What we're learning really about the future of technology, what we're learning about positive psychology,

643
01:12:50,590 --> 01:12:54,910
and also what we're learning about social media and the ability to connect to the whole society.

644
01:12:54,910 --> 01:12:59,350
We've watched people use that, Trump weaponized that against us.

645
01:12:59,350 --> 01:13:04,990
Can we turn that around and use these tools to actually connect people in the way you describe?

646
01:13:04,990 --> 01:13:09,040
And ultimately, can we use it to heal the wounds of history?

647
01:13:09,040 --> 01:13:13,360
And then education does become upstream diplomacy.

648
01:13:13,360 --> 01:13:18,460
And so there's another great peace process there, alongside the peace process between humans and technology.

649
01:13:18,460 --> 01:13:25,480
The peace process between humans and their past and perhaps the tech of the future can help us with that.

650
01:13:25,480 --> 01:13:30,310
And maybe then we get back. You know, as a non scientist, I grew up watching The Ascent of Man, Jacob Bronowski,

651
01:13:30,310 --> 01:13:35,740
and at the end, when he stands in the mud of Auschwitz, where he lost so many of his relatives.

652
01:13:35,740 --> 01:13:39,700
And he reaches down and picks up the mud from the puddle and just says at the end:

653
01:13:39,700 --> 01:13:43,510
We have to reach out and touch people. We have to reach out and connect.

654
01:13:43,510 --> 01:13:48,910
I think that's really where Imago Dei takes us. How do we connect as as humans?

655
01:13:48,910 --> 01:13:52,870
And perhaps that's where our fields connect as well?

656
01:13:52,870 --> 01:13:58,150
A final story I worked for for years on Northern Ireland, for the prime minister in Downing Street,

657
01:13:58,150 --> 01:14:04,300
and I was at one event on peace and reconciliation, and I asked one of the speakers: You know, why are you here?

658
01:14:04,300 --> 01:14:08,820
And she said: Well, my, my father was killed in the terrorist attack in the Brighton bombing.

659
01:14:08,820 --> 01:14:13,750
Alongside many others, and I'm here to talk about connexion, what makes us human.

660
01:14:13,750 --> 01:14:18,340
How do we connect? And I asked the gentleman next to her: You sir, why are you here?

661
01:14:18,340 --> 01:14:23,970
And he said: I was the bomber. You know, ultimately there is something,

662
01:14:23,970 --> 01:14:31,830
that is so human, as Lionel says. And if we can hold on to that, then that can really animate this work, but that can all, that

663
01:14:31,830 --> 01:14:45,670
can also make us better academics, better Heads of House, better citizens.