1
00:00:01,020 --> 00:00:07,230
Thank you. Thank you very much. Also, to come and stay on on keeping your hair for me rather than the free sandwiches as impoverished kids.

2
00:00:08,430 --> 00:00:16,140
So I'm Al Brown and I work at an organisation called DC DC, which is described as the Moody's Independent think tank.

3
00:00:16,830 --> 00:00:20,190
Because we're an organ of government. Semi-independent is probably here.

4
00:00:20,190 --> 00:00:27,450
It's a trade and our job is to think about stuff and study trends and try and work out what it means for the future

5
00:00:27,810 --> 00:00:34,560
and then provide advice to sort of governments can form strategies and try to work out what to do about these things.

6
00:00:35,130 --> 00:00:42,209
And we do that by plagiarising. We feel that we thing a lot relevant to themselves, which is one of the reasons I'm in the region.

7
00:00:42,210 --> 00:00:46,710
So I will get at least as much from you, from the questions as you think you're going to get from me.

8
00:00:48,070 --> 00:00:50,219
There's a very long list there and you don't get to read all of it.

9
00:00:50,220 --> 00:00:54,540
It's meant to just look impressive as you scan across the people that we steal from.

10
00:00:55,290 --> 00:01:02,040
And I wrote a blurb for this, but Robert and I went back and looked at it and realised it's not my promise to talk about

11
00:01:02,040 --> 00:01:08,160
some strongly held certainties that we have about the future and why they might not be safe,

12
00:01:09,090 --> 00:01:13,230
in which they might not be accurate, let's put it that way, and I'll put them on here.

13
00:01:15,060 --> 00:01:20,040
And at the end of that, if you recognise anyone has any of these before then that's good.

14
00:01:20,040 --> 00:01:24,299
I hear these all the time. And then on I ask, What do you mean by that?

15
00:01:24,300 --> 00:01:29,340
Frequently the conversation sort of gets a bit awkward and embarrassing as we start the times,

16
00:01:29,340 --> 00:01:33,330
and hopefully this will sort of skim over some of these bits because there's a lot to cover.

17
00:01:33,540 --> 00:01:39,749
So we have to get quite a little bit, not terribly deep on everything. I just thought it was just natural.

18
00:01:39,750 --> 00:01:43,080
Do you want to say something about caveats? Not you misconceived speak for.

19
00:01:43,410 --> 00:01:48,420
Oh, we have a good point. So I don't mind lecturing here. I am not an organ of state delivering government policy.

20
00:01:49,150 --> 00:01:52,770
You are picking the brains of the black out rounds. That's just what I'm looking expecting.

21
00:01:53,010 --> 00:01:56,370
But they're not here so happy with it as you're talking.

22
00:01:56,370 --> 00:01:59,400
So it's always good to start with a definition in economic thinking.

23
00:01:59,700 --> 00:02:08,310
So in this case, I would say that artificial intelligence is something that sits at the top of the Gartner Gartner point curve.

24
00:02:08,610 --> 00:02:15,690
So the peak of inflated expectations, but simultaneously manages to be something that is frequently underestimated and surprises us.

25
00:02:16,020 --> 00:02:21,450
It's quite an achievement. So something that's always better and simultaneously worse than we think is kind of worth understanding.

26
00:02:21,840 --> 00:02:28,940
And I use that here as a sort of bucket because you can find a million different definitions and that equals terribly well put together.

27
00:02:29,340 --> 00:02:33,200
But the useful way I found in thinking about this it is consists of three things really.

28
00:02:33,630 --> 00:02:39,540
One is algorithms, which is essentially just maths, much of which has been around since the fifties or sixties,

29
00:02:39,900 --> 00:02:43,740
but we didn't actually have other components that allow us to do anything with it.

30
00:02:45,150 --> 00:02:52,290
And then in 2012, we had a revolution in sort of the next leg, which is the processing power, particularly graphics processing chips.

31
00:02:52,710 --> 00:03:00,300
And they allowed us to use some of those algorithms, but they also only became useful once we have the data to exploit.

32
00:03:00,990 --> 00:03:04,980
And this is where the internet comes in and of itself to be enormously valuable.

33
00:03:05,610 --> 00:03:12,120
Not only is it a tool for finding data, but it's also become a system which generates data which makes a difference.

34
00:03:12,900 --> 00:03:17,580
So one of the things we do discover is pretty trends, which means all of our presentations in the polygraph system.

35
00:03:18,250 --> 00:03:26,880
So I going to tell you some things that you don't know already, but that's fine because Olympic athletes use priming by listening to like

36
00:03:27,330 --> 00:03:31,200
particularly inspiring music or watching films before they go and do an event.

37
00:03:31,210 --> 00:03:37,110
So I'm going to do the same to your Olympic level audience in this room and show you some of the things we know already to get you thinking in a way.

38
00:03:37,110 --> 00:03:40,589
So this is Moore's Law, so people talk about it running out.

39
00:03:40,590 --> 00:03:46,650
It kind of depends on how you choose to measure it. So there is a point where the size of a chip versus the size of the atom becomes a problem.

40
00:03:47,010 --> 00:03:51,840
But if you measure it in terms of calculations per thousand dollars, it continues to go strong strongly.

41
00:03:52,380 --> 00:03:57,240
And if you look at data generation, that is increasing exponentially.

42
00:03:58,530 --> 00:04:03,780
And if you look at why that is, again, going back to the Internet connectivity, the Internet matters.

43
00:04:04,650 --> 00:04:11,219
So I think we had 1 million devices in 1992 connected to the Internet whereabouts billion now.

44
00:04:11,220 --> 00:04:16,560
And the 200 billion is the estimate by 2050, probably slightly off on these figures, but kind of doesn't matter.

45
00:04:16,680 --> 00:04:24,419
It's a huge growth. I think it 50 billion. But if I tell you those things that make up the box, they weren't sticking your brain.

46
00:04:24,420 --> 00:04:26,580
So if I show you another image to do that sort of framing thing,

47
00:04:26,820 --> 00:04:31,290
you will instinctively recognise these on almost an emotional as well as an intellectual level.

48
00:04:31,620 --> 00:04:35,310
So we talk about data being generated and occurring automatically.

49
00:04:35,640 --> 00:04:40,320
That's kind of why I'm and that presents us with the sort of the first problem,

50
00:04:40,650 --> 00:04:45,930
which is there's an idea somehow that in all warfare you will just be able to alter and that has its

51
00:04:45,930 --> 00:04:50,280
place in things like you stood the line from home and if you were to be a successful terrorist,

52
00:04:50,610 --> 00:04:56,700
behaving like a 12th century man makes you quite difficult to find on the Internet, but that doesn't necessarily work for warfare.

53
00:04:56,700 --> 00:04:59,940
In an example I give you, is that a typical 99?

54
00:05:00,120 --> 00:05:06,780
The sortie generates between 20 to 40 laptops worth of data, one that comes back and it's presented to a human analyst.

55
00:05:06,810 --> 00:05:13,190
He then sits there and tries to make sense of all that data that necessarily means in the tempo that warfare occurs.

56
00:05:13,710 --> 00:05:17,030
The vast majority of that information then gathered on the cutting room floor.

57
00:05:17,820 --> 00:05:22,770
And if you're making information or making decisions on fractions of the data available to you,

58
00:05:23,160 --> 00:05:27,660
you are necessarily taking risk that you are not making the best decisions that you could do.

59
00:05:28,230 --> 00:05:33,900
So automation and how we ultimately how we process information is going to be fundamental warfare,

60
00:05:34,350 --> 00:05:36,960
but also in all other aspects of life, I would say too.

61
00:05:36,990 --> 00:05:43,830
So those graphs you saw about the increase in rates of data availability are true for all elements of the economy, economics and society as well.

62
00:05:46,380 --> 00:05:51,630
In 2015 we had an interesting development which kind of helps response of the response to that.

63
00:05:52,380 --> 00:05:58,740
And then it goes back to where Office Intelligence plays into this pretty much the first time on a standard benchmark test,

64
00:05:59,400 --> 00:06:04,630
the machines became better than humans at recognising faces and attributes.

65
00:06:06,120 --> 00:06:12,630
And you can see the error rates going down. And unsurprisingly, the error rate for humans has not changed because we're not evolving at that pace.

66
00:06:14,880 --> 00:06:19,230
So our ability to automate processing is going to necessarily depend on some of those technologies.

67
00:06:19,530 --> 00:06:25,110
And I described before how those three constituent elements I'd say there's two underpinning elements that make a difference as well.

68
00:06:25,800 --> 00:06:30,330
The first one, the image of the guy on the left of a subject matter, experts who are able to do this.

69
00:06:30,750 --> 00:06:37,110
And at moment they are a relatively low population figure and that makes them highly sought after.

70
00:06:37,530 --> 00:06:39,659
Now, in the future, there are going to be economic drivers,

71
00:06:39,660 --> 00:06:44,100
which means that everybody is going to be interested in doing more education, more research and creating more specialists.

72
00:06:44,430 --> 00:06:49,470
But right now they are a rare breed. And that ties into the second picture, which is to do with investment.

73
00:06:49,890 --> 00:06:59,549
And this is probably best described by a statistic in time, Wendy Hall's report for the government's view of economics and technology.

74
00:06:59,550 --> 00:07:04,950
When they were looking at artificial standards for machine learning and they highlighted the fact that there's a process called factory hiring,

75
00:07:04,950 --> 00:07:10,050
which is where you go out as a large company and you acquire in order to hire

76
00:07:10,470 --> 00:07:16,230
a small start ups or specialists on the average person is 5 to £10 million,

77
00:07:16,740 --> 00:07:21,959
which is a very significant figure. And you all have heard of sort of the amazing amounts the tiny little company DeepMind

78
00:07:21,960 --> 00:07:26,200
were built for from things that didn't necessarily means this ferocious competition.

79
00:07:26,260 --> 00:07:28,320
It's interesting as well, because if you are Google,

80
00:07:28,500 --> 00:07:33,930
you won't necessarily just go out and buy the Start-Up that has a gap that you need to fill in your stable.

81
00:07:34,440 --> 00:07:38,790
You will buy things that you may already have because you can deny it to your competition.

82
00:07:39,180 --> 00:07:42,300
So, so quite often there are lots of routes to achieve the same end.

83
00:07:42,600 --> 00:07:45,720
And if you own all of them, you start out towards being a natural monopoly.

84
00:07:48,270 --> 00:07:54,990
And the reason I put up a picture here of the Manhattan Project is I think this tends to lead towards an interesting phenomenon,

85
00:07:55,020 --> 00:07:56,700
which I call the Manhattan problem.

86
00:07:57,660 --> 00:08:06,000
So in the Manhattan Project, Oppenheimer is running a program where the state has a monopoly on all the constituent ingredients,

87
00:08:06,340 --> 00:08:11,450
can't go to the shops and buy uranium. You don't have big stables of scientists wandering around in the fray today.

88
00:08:11,460 --> 00:08:16,920
And Fred Oppenheimer and a number of people are all working in a very closely confined environment with

89
00:08:16,920 --> 00:08:21,180
walls of secrecy built around because they don't want to be counted on to be good at building atomic bomb,

90
00:08:21,180 --> 00:08:27,510
you need to build one. You need to bounce the ideas around those small groups of specialists in order to then accelerate the program to get further.

91
00:08:28,560 --> 00:08:33,000
And the state had a monopoly on that, which is one of the reasons, of course, that the U.S. have one.

92
00:08:33,000 --> 00:08:36,570
And then there's a gap of years before other major world powers start to catch up.

93
00:08:37,290 --> 00:08:43,139
In this instance, the reason I call a problem is that the constituent ingredients are no longer really owned by states,

94
00:08:43,140 --> 00:08:50,970
but they are owned by technology giants. So the data say 90% of all searches on the Internet were done through Google.

95
00:08:51,870 --> 00:08:58,200
The nearest competitor, Bing, I think, has the same number of searches as all conducted just through Google Maps.

96
00:08:58,680 --> 00:09:04,890
So you really have start hoovering up that sort of data and controlling it becomes important and then it starts to belong in the commercial sector.

97
00:09:05,010 --> 00:09:11,120
And indeed, there are regulations that prevent states and certain countries from holding or acquiring data.

98
00:09:11,130 --> 00:09:17,490
So you'll have had a GDPR and how that is radically altering what people are allowed to own, particularly state institutions.

99
00:09:17,790 --> 00:09:21,550
And there was this piece of regulation, which is this treaty now in the UK called RIP,

100
00:09:21,610 --> 00:09:29,010
which was the Regulation of Investigatory Powers Act, which stops the states acquiring large amounts of data just because its citizens.

101
00:09:30,830 --> 00:09:34,799
And I think that sort of purchasing power doesn't really tend to belong to the states as well.

102
00:09:34,800 --> 00:09:38,760
It largely belongs to commercial organisations because they can exploit it better as well.

103
00:09:41,190 --> 00:09:45,480
And you would have seen it in Zuckerberg in front of the Senate recently.

104
00:09:45,510 --> 00:09:54,060
You would have seen press statements on Cambridge Analytica and part of how society feels is kind of emerging in that field.

105
00:09:54,810 --> 00:09:59,820
But one thing, and this goes back to one of the least certainties I would be slightly cautious of is if.

106
00:09:59,930 --> 00:10:03,560
You ever hear someone say other people will be prepared to do things that we really wouldn't?

107
00:10:04,130 --> 00:10:09,440
Well, that's always true. That's kind of us having a different moral framework to those around us just in this room.

108
00:10:09,800 --> 00:10:15,380
It's always a fact of life, and different social groups having different ethical frameworks is a natural space as well.

109
00:10:16,010 --> 00:10:17,870
I'll be very careful about any exceptionalism,

110
00:10:17,870 --> 00:10:24,859
assuming that you will always have some sort of moral standing which you sits on which precludes you from doing something the other guys.

111
00:10:24,860 --> 00:10:28,550
Would you tend to alienate the enemy at it, but they will have different views.

112
00:10:28,640 --> 00:10:36,230
I would suggest that if you were in North Korea, for example, your comfort with the concept of autonomy is very different to in a Western state.

113
00:10:37,040 --> 00:10:43,429
So if you live in an organisation where autonomy isn't socially acceptable and yet you can control where people's families are,

114
00:10:43,430 --> 00:10:47,540
every piece of education they receive and what kind of behaviours you expect.

115
00:10:48,110 --> 00:10:51,860
The idea that you would then expect in machines vastly disproportionate level

116
00:10:51,860 --> 00:10:55,160
also to meet with none of those same levers or institutions of competence,

117
00:10:55,490 --> 00:11:00,800
I think is perhaps to paint the opponents in a negative light rather than actually looking to them for power.

118
00:11:01,700 --> 00:11:06,460
And we also you have to be careful as well about the idea of some sort of fixed ethical framework.

119
00:11:06,470 --> 00:11:11,930
We change our ethics all the time, and you can go back and look at things like in-vitro fertilisation,

120
00:11:12,650 --> 00:11:15,200
genetic screening for things like Down's syndrome.

121
00:11:15,410 --> 00:11:20,990
And you go back and you will find that a lot of a furore tends to disappear in retrospect as we change our ethical framework,

122
00:11:21,020 --> 00:11:25,790
as we become comfortable with changes in society and technology, and indeed we're not even that stable now.

123
00:11:26,240 --> 00:11:30,980
So one third of Britons surveyed in this newspaper report said they feared the rise of the robot,

124
00:11:31,100 --> 00:11:37,730
but nearly a fifth were prepared to have sex with an android. That's not a very coherent position, emotionally speaking, because not intellectually.

125
00:11:38,660 --> 00:11:44,630
And one of the reasons we tend to have these images is because we're trying to do we do it here mistake substitution.

126
00:11:45,110 --> 00:11:48,890
So we're not familiar. We can't see you do either of them at work.

127
00:11:48,950 --> 00:11:54,740
It's just a call and response effect on computers mean anything to us on an intellectual or emotional basis.

128
00:11:55,010 --> 00:11:58,100
So we substitute in what we do now, which is sci fi.

129
00:11:58,430 --> 00:12:02,390
And this particular picture and I mean this particular picture,

130
00:12:02,540 --> 00:12:08,240
it's the most common image that people have for the rise of robots, because this is the one that is most licensed by newspapers.

131
00:12:09,140 --> 00:12:15,140
So The Daily Telegraph printed full stories in four months using the same picture and the same story.

132
00:12:15,620 --> 00:12:20,460
So to find that your population then starts to develop a mental image of what the future looks like, and it looks a bit like this.

133
00:12:20,780 --> 00:12:28,309
So it's for us, there are more new exposure to, say, the Zuckerberg and sort of fear of big data is one form.

134
00:12:28,310 --> 00:12:33,740
This is from vehicle slaughter bots. Let's get to my institution of life, which is from the context of killer robots.

135
00:12:34,120 --> 00:12:41,090
And this is if you want to have an influence, campaigns that exist in Britain, this is definitely how to change public opinion.

136
00:12:41,360 --> 00:12:45,860
And it's really good. And the idea is at the beginning, you have a sort of Steve Jobs like military to get he hands up.

137
00:12:46,130 --> 00:12:51,350
He's created quadcopters with little shape charges in. They go out to some people who are nearby.

138
00:12:51,350 --> 00:12:53,480
Well it's like it's never really clear who they are.

139
00:12:54,110 --> 00:12:57,739
And then they end up going out in assassinating thousands of students based on things like Facebook

140
00:12:57,740 --> 00:13:01,640
profiles and leading all the students along because they seem to be right or left wing enough.

141
00:13:02,840 --> 00:13:06,290
And it's a terrifying image and it's quite emotionally powerful.

142
00:13:06,650 --> 00:13:10,790
But one of the reasons it hasn't led to sort of it's frustrated the company,

143
00:13:10,880 --> 00:13:16,100
the robots and others in the UN who are looking at it as an institution called The Gate,

144
00:13:16,340 --> 00:13:19,610
which is looking at whether to ban lethal some of US weapon systems.

145
00:13:20,420 --> 00:13:24,890
Are they really struggling? And they're not struggling because that evil people trying to dissuade it,

146
00:13:25,340 --> 00:13:29,210
they're struggling because of the challenge, which is trying to work out what it is you write about.

147
00:13:29,690 --> 00:13:33,319
So nobody particularly wants to get rid of a heat seeking missile because we know

148
00:13:33,320 --> 00:13:37,340
those have been more firmly front to back and nobody wants the silver version of Army.

149
00:13:37,340 --> 00:13:43,100
It's got some bad storms around them, scores of thousands of people. But you can't write a law about you need to find a boundary.

150
00:13:43,250 --> 00:13:47,720
And all of the things that are in that video kind of exist in the real world already.

151
00:13:48,050 --> 00:13:51,720
So there's very little in that. And you couldn't pick any one of those and say, this is the thing that we hate.

152
00:13:51,950 --> 00:13:53,690
You can say facial recognition technology.

153
00:13:53,700 --> 00:14:00,320
She'll never be in service again because otherwise it will keep getting longer and your friends would recommend you on Facebook and things like this.

154
00:14:01,190 --> 00:14:07,670
So there are all also it's there's no constituency single policy that technology and autonomy is kind of an emergent term.

155
00:14:07,760 --> 00:14:14,960
You tend to think of it as being a an ability to leave a system alone to get along with its job without being supervised.

156
00:14:15,080 --> 00:14:18,620
The problem is that that's contextual and it's based on your assessment of risk,

157
00:14:18,860 --> 00:14:25,460
and it's based on familiarity and the predictability that we've seen in our system in the past, which makes generating a law particularly difficult.

158
00:14:25,910 --> 00:14:32,720
There are issues that are very pertinent, so things like assurance and trust, how that is going to be collective training,

159
00:14:33,080 --> 00:14:39,170
which is more they're more like how sort of car assurance or airline assurance works in some ways.

160
00:14:39,170 --> 00:14:43,729
And those are probably going to get tackled first and will probably be the way that

161
00:14:43,730 --> 00:14:47,520
this starts to resolve itself in the same way as you have legislation around airlines,

162
00:14:47,810 --> 00:14:52,190
which we now would agree. So I would be very wary of fixed conceits or views.

163
00:14:52,730 --> 00:14:58,700
And and there's another problem, I'd say. So this is a system that sits on a North South Korean border northwards.

164
00:14:59,330 --> 00:15:02,850
It's made by. Subsidiary of Samsung. It's a remote weapons station.

165
00:15:02,850 --> 00:15:07,350
And what it does is it has the capability to do automatic detection and targeting.

166
00:15:08,010 --> 00:15:12,450
There are various as well, which can do voice recognition and facial recognition.

167
00:15:12,720 --> 00:15:18,900
And right now, it's it's laid back to a guy in a control room. So there's a bloke with a joystick deciding whether to shoot or not.

168
00:15:19,920 --> 00:15:23,340
But fundamentally, it's actually a software patch away from not needing that.

169
00:15:25,020 --> 00:15:30,720
And it kind of does everything that a century does. So it's great because he's not just sitting in a little box getting rained on,

170
00:15:31,140 --> 00:15:35,600
but he's also more vigilant so he doesn't get distracted by pulling on his iPhone all the time.

171
00:15:36,090 --> 00:15:39,180
Of course, because he's tired and had a big lunch. So none of those things happen.

172
00:15:40,320 --> 00:15:44,340
And yet this is something created by a man in America who doesn't like to kick out his lawn.

173
00:15:48,490 --> 00:15:52,630
This is one of the fundamental problems with the idea of the banner that this guy has put this together,

174
00:15:52,740 --> 00:15:59,309
and it's essentially the same system more or less, and yet he's running it off a laptop.

175
00:15:59,310 --> 00:16:03,450
No difference in the thing over there running it. And he's used some actual systems of service and some software.

176
00:16:03,450 --> 00:16:10,120
It's downloaded from the Internet. So one of the problems that we've got is that this is not, you know,

177
00:16:10,140 --> 00:16:14,120
in the idea that there's a Steve Jobs like figure out that leading all this technology is not the case.

178
00:16:14,130 --> 00:16:20,040
So going back to that kind of Manhattan problem idea, this is very much being developed in the civil commercial.

179
00:16:20,580 --> 00:16:27,390
These technologies that are accelerating in that world first and it's not like any here is somebody who's got a really old breakwater.

180
00:16:27,750 --> 00:16:34,110
And yet if you were to say they're going to put a sighting system on, it has the world's most advanced military accelerometers.

181
00:16:34,320 --> 00:16:36,350
It's links to G.P.S. mapping system.

182
00:16:36,360 --> 00:16:45,190
It's also a communication system that also does azimuth and has data for compass and can get light to mental updates on weather environment.

183
00:16:45,420 --> 00:16:49,710
It sounds really cool. And then you say, I'm talking about knowing that bunch of stuff on the side.

184
00:16:50,190 --> 00:16:58,620
You're able to, you know, generate some quite complex effects or quite complex technological things by using quite simple off the shelf technologies.

185
00:16:58,890 --> 00:17:02,400
The U.S. Marine Corps, I've got a great phrase for this. They call it the democratisation of precision.

186
00:17:02,880 --> 00:17:06,810
And I think it's going to lead to a lower cost of entry to, well, you know,

187
00:17:06,810 --> 00:17:11,190
precision warfare used to be a first world nations force, and it's not going to be for very long.

188
00:17:13,050 --> 00:17:16,290
But don't just think about it in terms of technology as well.

189
00:17:16,550 --> 00:17:19,710
Think instead of old actors being able to do new things.

190
00:17:20,520 --> 00:17:23,690
I would say that you have to think about it in social structures as well.

191
00:17:23,700 --> 00:17:28,800
So that is a self-driving car would have millions that are out there. If I was a member of the Animal Liberation Front,

192
00:17:29,550 --> 00:17:33,720
one of the things I really like doing is forming Huntington like scientists and trying to kill people out.

193
00:17:34,680 --> 00:17:43,320
And I have in my family Animal Liberation Front, I know no tolerance for being a martyr or even being harmed or even being arrested.

194
00:17:43,620 --> 00:17:50,189
And right now, that limits me to sending crappy pass the bombs in the mail to try and get someone to continue the life sciences which funny.

195
00:17:50,190 --> 00:17:58,830
I think they're really stupid, you know, and it does limit what you can say if in the future I can move my bomb to target the target,

196
00:17:58,890 --> 00:18:06,150
I no longer needs to have somebody pressed into service behind will have access to what has been described as the poor man's cruise missile.

197
00:18:06,630 --> 00:18:12,330
So the nature of violence is going to change and it's almost impossible to regulate out all of these things, obviously.

198
00:18:13,830 --> 00:18:19,020
Now, I wouldn't say also some people think that this just leads to a completely flat playing field and everybody's going to have other things.

199
00:18:19,680 --> 00:18:25,940
You will have an economic divergence. So again, it's a bit of footage from now rather than future, say the world's cheapest retailer,

200
00:18:26,670 --> 00:18:33,120
which is basically a small grenade which has been attached to this silly blue like tail structure.

201
00:18:33,120 --> 00:18:36,840
And some of these flying, of course, comes out bombing some of their opponents.

202
00:18:38,820 --> 00:18:44,730
And the ability to do that is not the same as the ability that is shown here.

203
00:18:45,240 --> 00:18:50,940
And there are other variants. So you can see locust or vertical on the internet, which is generated by the U.S. It's pretty similar.

204
00:18:50,940 --> 00:18:55,080
So this is developed by Chinese university in consultation with the Defence Forces.

205
00:18:55,350 --> 00:18:59,160
And the idea here is to build a swarm of systems which are able to interact.

206
00:18:59,460 --> 00:19:03,000
So, you know, there isn't going to be a one size fits all solution either.

207
00:19:04,440 --> 00:19:09,330
And I would also say that some of them are things of the simplicity of the idea and also natural compromise.

208
00:19:09,810 --> 00:19:16,170
So this is a little remote controlled aircraft that can land itself on a wall and then take off again and fly somewhere else.

209
00:19:16,260 --> 00:19:25,990
Maybe Chevron or Chevron domestic, forget which. And their idea here is to actually if I'm trying to look for people in an aircraft, say,

210
00:19:26,040 --> 00:19:31,559
I'm sorry, that my time, my ability to survey the area is limited by battery life.

211
00:19:31,560 --> 00:19:35,820
So if I can park, then I can extend the right the duration of my sensor.

212
00:19:36,690 --> 00:19:40,229
But actually parking on the floor is difficult because everything is fallen over and it says earthquake here.

213
00:19:40,230 --> 00:19:44,160
But, you know, you could easily imagine that would be severe somewhat today.

214
00:19:45,060 --> 00:19:48,780
But walls are self-cleaning if it doesn't stay there. So you're ready to talk?

215
00:19:49,230 --> 00:19:54,450
Actually, it's quite useful. But the battery life issue could be extended by another university,

216
00:19:54,450 --> 00:19:59,790
which you're looking at sending the propeller that sits on the top into a wind farm that regenerates the battery.

217
00:19:59,830 --> 00:20:02,920
As it sits and hangs on more. So you start to change some of the signs.

218
00:20:03,130 --> 00:20:11,980
The parishes. Some of these ideas are also farms from other variations say, you know, you will see stuff that's biomimetic.

219
00:20:12,250 --> 00:20:16,690
So again, one of the ideas I present to you is it's going to be a plethora of different things.

220
00:20:17,020 --> 00:20:23,860
We tend to think about robots back again in that sort of only picture, but they'll be all sorts of shapes and sizes and they won't all be fixed.

221
00:20:24,520 --> 00:20:26,200
So if not all solid.

222
00:20:26,620 --> 00:20:32,390
I also like to put this one up because then as I look across the audience, I can find work one in five, all from that newspaper report.

223
00:20:36,850 --> 00:20:43,390
Some people think that the information of these systems across the future tobacco space means that there isn't going to be any more surprise.

224
00:20:43,630 --> 00:20:48,700
Everything will be seen and anticipated in future. But I'm sorry I've missed a bit.

225
00:20:49,420 --> 00:20:54,670
One of the other things that I think artificial intelligence enables is going back to what's right now.

226
00:20:54,730 --> 00:21:00,010
The other tool for power that's quite significant is you find permanently broadcasting a signal like drain my batteries quite quickly.

227
00:21:00,280 --> 00:21:07,150
And in a military sense, being noisy has always been that and it's tuning the electromagnetic spectrum is what is walking around in the dark.

228
00:21:08,380 --> 00:21:16,120
So the ability to compress that data imagery, so I'm not sending you a full motion video, but I'm now actually sending you the words crank games.

229
00:21:16,120 --> 00:21:20,260
I can send a lot less data and I can be a lot more quiet and I can consider my battery life.

230
00:21:20,740 --> 00:21:26,380
So the ability to develop these algorithms, stick them on those systems, is going to change how the communication works,

231
00:21:28,060 --> 00:21:34,810
which indeed will help with that sort of volume of data, not just getting to the individual, but being processed across the airwaves masses.

232
00:21:35,440 --> 00:21:40,690
The electromagnetic spectrum is already too full and we have to give it to the person as well.

233
00:21:40,690 --> 00:21:45,310
So this isn't just going to be how it gets into the robots, it's going to be how it gets into the human events decisions.

234
00:21:45,590 --> 00:21:47,460
And this is an example from Arrow Glass.

235
00:21:48,160 --> 00:21:53,320
So the idea here was that rather than pilots having to sort of read or memorise their folder before they take off,

236
00:21:54,130 --> 00:22:01,720
they can actually have all of that data put in that the very cool set of glasses that you'll see the operator using the face on it.

237
00:22:02,530 --> 00:22:10,540
And it presents the data in a way that's fast and intuitive because speed of decision matters in all forms of life, but definitely in warfare.

238
00:22:11,560 --> 00:22:16,480
And you can see how the immediate read across to how people conduct warfare and could

239
00:22:16,480 --> 00:22:20,709
make better decisions with better information is going to be not just a robotics thing,

240
00:22:20,710 --> 00:22:28,660
but a how you integrate with the people. I'm going back to the idea of those little senses being everywhere, meaning there's no more surprising war.

241
00:22:28,660 --> 00:22:33,100
I don't think that's true. And the first example I would give you is just so chess.

242
00:22:33,170 --> 00:22:36,700
You have a completely glass bottle, you have perfect situational awareness.

243
00:22:36,700 --> 00:22:41,880
You know where everything is. You know all of the rules and capabilities, bases, and you also know the tempo as well.

244
00:22:42,010 --> 00:22:43,580
And you can. So your opponent takes a turn.

245
00:22:43,610 --> 00:22:49,180
You take it to an example, but if you play against a better player, you will still find yourself being surprised.

246
00:22:50,680 --> 00:22:55,270
The other thing I would say to you and without a quiz is just to be thinking, why does anyone know who that is?

247
00:22:57,310 --> 00:23:02,040
Well, Google's face in that algorithm is absolutely certain that that is military jobs.

248
00:23:03,670 --> 00:23:07,540
And the reason that is certain is because it is slightly semantic like glasses.

249
00:23:07,540 --> 00:23:15,129
That, again, star doesn't seem to be a feature in this presentation. And what it's doing is it's something called up a serial example.

250
00:23:15,130 --> 00:23:18,100
So the best description of it is on the top left here.

251
00:23:18,520 --> 00:23:24,070
So the picture on the top left of the panda is an original source picture, a crazy static looking box.

252
00:23:24,400 --> 00:23:27,640
Awesome changes to the colours of that picture that are really subtle.

253
00:23:28,190 --> 00:23:30,880
Now let's do that first, which it's create a picture on the right.

254
00:23:31,460 --> 00:23:38,020
What it's doing is it's increasing the signals that a computer sees that give it a different outcome.

255
00:23:38,440 --> 00:23:42,310
So you can see the scores below that again from Google in this recognition system.

256
00:23:42,580 --> 00:23:45,730
So it's 57% confident that it's a panda.

257
00:23:46,090 --> 00:23:50,740
And then you and I see what looks like the same picture, and yet it is 93% confident that it is a given.

258
00:23:51,790 --> 00:23:56,019
And that's sort of the subtle pictures there of the subtle pixel changes of water, actually,

259
00:23:56,020 --> 00:24:03,009
and that so the glasses that make the computer think about this military challenge, it's not just linked to that site, that star as well.

260
00:24:03,010 --> 00:24:09,669
There are more simple versions that are out there. So the crazy looking picture on the bottom left a slightly is a fractal pattern developed by an

261
00:24:09,670 --> 00:24:15,190
artist called Army and who has sort of a big brother theme to a lot of his designs for fashion.

262
00:24:15,490 --> 00:24:23,140
And it's a fabric pattern. And what it does is that is the the set of shadows, gaps, distances, the facial recognition software is looking for.

263
00:24:23,560 --> 00:24:27,879
And as you change the folds of your clothes, as you move around, it changes those distances.

264
00:24:27,880 --> 00:24:31,150
And when you walk past the facial recognition camera, it doesn't just get too far.

265
00:24:31,420 --> 00:24:39,730
That's wrong. It says that Robyn Rodney is buried in a list of thousands of other names and if there are enough on the computer isn't good enough,

266
00:24:40,000 --> 00:24:45,520
it just has a little over 500 shots that otherwise it has to sift through to find robots of all of those other names,

267
00:24:45,520 --> 00:24:50,440
which probably correlate to real people as well. So deception is like, well, I like this one.

268
00:24:50,440 --> 00:24:55,690
So this sees it in like real time for something that's 3D. So you can see this live picture there.

269
00:24:55,690 --> 00:24:59,710
It is being so largely recognised as a total 3D printed and.

270
00:25:00,080 --> 00:25:04,580
Some of those adversarial patterns in it and you can see it largely consistently being recognised as a.

271
00:25:06,140 --> 00:25:12,230
So the prospect of surprise disappearing or indeed machines being in front of them, that judgement seems to be extreme.

272
00:25:13,770 --> 00:25:17,690
Of these four people are all interesting. Does anyone know how they what they have in common?

273
00:25:18,860 --> 00:25:21,470
Well, the answer is that none of them exist. One would have.

274
00:25:23,180 --> 00:25:30,530
And so these are all generated by computers that we're trying to create from their own imagination.

275
00:25:30,530 --> 00:25:35,660
If you would like a better sense of what a human looks like and you can see it increases quite radically.

276
00:25:35,960 --> 00:25:39,080
So the prospect of deception isn't just that we can form machines,

277
00:25:39,080 --> 00:25:42,830
but it's increasingly going to be machines will allow us to fool each other as well.

278
00:25:44,330 --> 00:25:48,550
And another example I'll give you is live, real time video stitching.

279
00:25:49,490 --> 00:25:53,320
Now, somebody will know this from slightly more insane resources on leave up to your own morals.

280
00:25:54,590 --> 00:25:58,930
But you can see here somebody is able to in real time change the expressions.

281
00:25:58,940 --> 00:26:06,170
And indeed, there are other videos of people changing horses. And if you stare at George Bush's mouth on this, it looks a little bit grainy.

282
00:26:06,590 --> 00:26:11,150
But it's worth remembering that it will never be this bad again. This is only going to get better.

283
00:26:11,840 --> 00:26:15,500
So our ability to say seeing is believing on the Internet is going to diminish.

284
00:26:15,860 --> 00:26:23,610
And this is an area that is rapidly increasing and capable and an object that I'm thinking of, things that are ongoing.

285
00:26:24,130 --> 00:26:27,770
So DeepMind is is a fascinating organisation been speaking to and a lot of this.

286
00:26:29,270 --> 00:26:34,099
One of the things that is interesting I think is in the world of cybersecurity,

287
00:26:34,100 --> 00:26:40,730
one of the things that professional cybersecurity analysts are worried about is how some of these things can transpose.

288
00:26:41,120 --> 00:26:46,610
So Deepmind's next trick is to try and win a game called Starcraft.

289
00:26:47,180 --> 00:26:51,860
The interesting thing about Starcraft from a machine learning perspective is that you don't get to see over the battlefield.

290
00:26:51,860 --> 00:26:55,939
The other side is shrouded in fog until you get there and look, they aren't the same as you.

291
00:26:55,940 --> 00:26:59,569
They don't necessarily have the same features. You play real time.

292
00:26:59,570 --> 00:27:05,780
So there is a deterrence thing going on and there's a long term as a short term strategy component.

293
00:27:05,780 --> 00:27:09,350
So if I take a my guys have moved over and punched all your guys in the face of the very beginning,

294
00:27:09,800 --> 00:27:14,450
do I do that or do I go a little stunt by going off to build an arms factory so that we're coming back,

295
00:27:14,450 --> 00:27:17,540
makes a big takes versus maybe you're doing something in an intermediate term.

296
00:27:17,840 --> 00:27:26,059
So there's a series of judgements that need to be made. And the thing that is interesting to the cybersecurity world is that this could

297
00:27:26,060 --> 00:27:30,740
be readily transposed to what's called advanced persistent threat techniques.

298
00:27:31,730 --> 00:27:37,700
And how that works in cybersecurity is if you have a shield ball in front of you protecting yourself.

299
00:27:38,180 --> 00:27:43,759
And I am an incredible hacker. What I would do is I came up and I just tried your share war until I find a gap.

300
00:27:43,760 --> 00:27:50,540
So I'm looking for a weakness in your system. But more than that, I might do something which causes you to move one of your shields or responders.

301
00:27:50,690 --> 00:27:51,830
So I'm not trying to solve both.

302
00:27:52,040 --> 00:27:57,350
But if I keep pressing here, I notice you move this to here and I think that expresses weakness and then I can get here.

303
00:27:57,890 --> 00:28:05,360
And right now that requires very skilled hackers and a lot of time and expertise because then manually reading code we can get into the game.

304
00:28:05,720 --> 00:28:11,630
But if you're doing that, the cyclical rate of minute machine learning, your tempo becomes vastly faster.

305
00:28:12,140 --> 00:28:17,270
And so this means that not only in defence in terms of trying to hack in, but in terms of defence,

306
00:28:17,780 --> 00:28:24,650
you are going to find that systems like this are probably going to dominate the online version of cyber attack and cyber defence.

307
00:28:24,950 --> 00:28:28,669
And the other thing as well as you don't have to understand it from the ground up in the same

308
00:28:28,670 --> 00:28:33,079
way as the pilots does not have to be able to go to an aeroplane to be able to operate it.

309
00:28:33,080 --> 00:28:37,520
You just need star to fly. You don't need to be able to code this for first principles.

310
00:28:37,790 --> 00:28:43,520
If you can take it real, write it to a target and send it off on its way so it lowers the threshold for some of those sorts of attacks.

311
00:28:44,630 --> 00:28:47,960
Another factor that's going to change, I think the economics of war is,

312
00:28:49,460 --> 00:28:56,090
is the idea of huge headquarters and massive logistics supply base is not background or capacity.

313
00:28:57,110 --> 00:29:04,040
And the reason I say that is the idea of what's called parasitic warheads. So if you can really accurately target something more accurately,

314
00:29:04,040 --> 00:29:07,879
you can pick your target the less explosives, and that's what you need to take to be targets.

315
00:29:07,880 --> 00:29:14,630
And so if I can fly, if I can send a small unmanned drone to fly to Camp Bastion and it's looking

316
00:29:14,630 --> 00:29:18,510
for the wings of aeroplanes and it just puts bomb the whole street diamonds.

317
00:29:18,830 --> 00:29:24,140
I'm starting to make some really expensive aircraft, unreliable, relatively low risk.

318
00:29:25,040 --> 00:29:30,830
And there's a great quote which says, lining up all these aircraft on the runway might be the next version of lining up the ships at the moment.

319
00:29:31,940 --> 00:29:38,750
So there are some old people will have to reload. So in the Cold War, a whole bunch of an army wasn't just finding some big carrier somewhere.

320
00:29:38,990 --> 00:29:45,370
It was to move and hide in forests. And one of the lessons I think is coming out of the Ukraine, where both sides are using hard to detect these,

321
00:29:45,610 --> 00:29:53,440
is that both sides have to behave as if the other side has kind of superiority because whoever is fighter is dominant,

322
00:29:53,460 --> 00:29:59,470
doesn't determine whether somebody is looking down from the sky. And if you are seeing seen, obviously you can be targeted with alternative.

323
00:29:59,920 --> 00:30:05,920
So the concept of a benevolent sky, I think is going to disappear and it opens up other tactics as well.

324
00:30:06,190 --> 00:30:11,919
So that's the idea of pilot autonomy. So a great example here is David Perkins.

325
00:30:11,920 --> 00:30:20,770
The U.S. general pointed out that a U.S. ally in the Middle East and fighting $3 million paycheques on to destroy a drone have been fired,

326
00:30:21,010 --> 00:30:24,370
which costs $200 for hours. And it was successful.

327
00:30:24,680 --> 00:30:30,729
But Frank Hoffman describes this as a strategy set, by my defence, to measure cost vastly more than your productive measure.

328
00:30:30,730 --> 00:30:33,610
And Perkins went on to say. And if I was the guy who applied to join,

329
00:30:33,610 --> 00:30:39,100
I'd be delighted because I'd now be going out and spending is decreasing because many of these children go to training as I can,

330
00:30:39,310 --> 00:30:42,880
because I know it costs you 300 or $300 every time I do that.

331
00:30:44,260 --> 00:30:51,640
So the prospect of overwhelming things is going to change. And you know, this the one of the other organs from his people tend to be swarms.

332
00:30:52,120 --> 00:30:56,680
I don't think it's as simple as that again. So there are environments where it makes loads of sense to be a swarm.

333
00:30:57,070 --> 00:31:01,540
Say, for example, one missile equals one subset of this radar field does disappeared.

334
00:31:02,110 --> 00:31:08,340
But there are real world signs waiting parachutes to come in. And the reason that we have things like AWACS is for early warnings.

335
00:31:08,980 --> 00:31:14,140
And if you are looking over a border, you need a really big radar dish to look at one on the way.

336
00:31:14,500 --> 00:31:18,160
So the concepts, if you will, of a swarm suddenly starts to become not early warning.

337
00:31:18,310 --> 00:31:19,870
It just tells you what's crossing the border.

338
00:31:20,260 --> 00:31:26,740
So it's going to be a mix of the big and the things we recognise now, and especially along with new systems.

339
00:31:28,690 --> 00:31:31,240
And there are other practical prospects that people cross over.

340
00:31:31,250 --> 00:31:36,130
And so some people have described the advantage of having an autonomous armoured vehicles like tanks.

341
00:31:36,490 --> 00:31:41,250
I don't think necessarily swapping the humans out for robots is going to be the way that this thing plays out.

342
00:31:41,290 --> 00:31:49,209
Since and the example I'll give you here is this a robot tank has been designed and created and it could be part of a platoon of,

343
00:31:49,210 --> 00:31:53,320
let's say, four tanks where three of them, a robot, takes number six to the back of the mountain.

344
00:31:53,740 --> 00:31:59,620
Well, that sounds fantastic. The robots can go forward and do the dangerous stuff and the humans in the back making sort of tactical judgement

345
00:32:00,040 --> 00:32:06,700
until one of those tanks has a moment where the equivalent of an arm and a flat tire is throwing a truck.

346
00:32:07,270 --> 00:32:12,520
And now the only people who can get out and put the truck back on the tank of the crew and you only have one crew.

347
00:32:12,970 --> 00:32:17,320
So I've now gone from having four times to having no tanks, whereas when they were all gone,

348
00:32:17,650 --> 00:32:23,350
I went from having four tanks to having three tanks, and sometimes those small real world practices moments.

349
00:32:23,350 --> 00:32:30,639
Again, there was a prominent guy who was showing this tank driver, but he hated the idea.

350
00:32:30,640 --> 00:32:35,500
And it wasn't the tank truck thing, it was the weapon being because his whole prospect was at the end of the battle.

351
00:32:35,500 --> 00:32:39,190
They're going to spend the next two days exhausting themselves, just feeding the weapons.

352
00:32:40,540 --> 00:32:44,590
So the small, pragmatic realities are going to make a difference and understanding how these things play out.

353
00:32:44,980 --> 00:32:48,760
And I'll run a video here. First of all, see, that's the Tom Cruise dialogue, which it was fun.

354
00:32:49,930 --> 00:32:55,150
So this is from a film called The Edge of Tomorrow. And this goes back to this idea that they're going to be super brilliant.

355
00:32:56,740 --> 00:33:00,160
It's worth understanding that Tom Cruise has the same effect in his films,

356
00:33:00,160 --> 00:33:06,550
that he starts out rubbish fighting this war in the same way as Alphazero was rubbish in fighting for war playing chess the first time.

357
00:33:07,120 --> 00:33:14,050
And yet in 4 hours you've got the grandmaster level. But like Tom Cruise, it's fighting the same battle again and again and again and again.

358
00:33:14,410 --> 00:33:17,350
And there are changes within it, but it's kind of always the same battle.

359
00:33:17,770 --> 00:33:22,180
One of the reasons we always say that we're ill prepared for the next war is because we've never fought the same war twice.

360
00:33:22,540 --> 00:33:28,869
That always different. And so your ability to sort of plug a machine in and it be optimised kind of depends on your

361
00:33:28,870 --> 00:33:33,580
ability to generate the data you need or have a repetition rate that allows you to learn with you.

362
00:33:34,060 --> 00:33:36,230
And another example I give you is fruit flies intelligence.

363
00:33:36,910 --> 00:33:42,670
So people study genetics and fruit flies because they breed fast, mutate fast and die fast.

364
00:33:43,360 --> 00:33:50,650
If we had studied genetics in elephants, the experiments would die at a faster rate than the experiments.

365
00:33:51,370 --> 00:33:53,319
So there's a fairly good reason we pick those once.

366
00:33:53,320 --> 00:33:58,870
And I think the same sort of understanding of what artificial intelligence is trying to learn is getting smarter.

367
00:33:59,170 --> 00:34:04,920
So the types of problems are going to shape what we can actually study these things and it plays itself out in funny ways.

368
00:34:04,930 --> 00:34:12,880
We also tend to think we talk about economics and mass unemployment, but which jobs are automatable is interesting.

369
00:34:13,240 --> 00:34:16,660
So there is quite often an assumption that the lowest paid jobs will be the ones that come

370
00:34:16,660 --> 00:34:20,320
in droves because they can't be that hard because we don't pay people to do them that much.

371
00:34:20,740 --> 00:34:24,670
But it's wrong. An example I'm giving you is on the left of the auto pilot.

372
00:34:24,730 --> 00:34:31,480
He was able to find a747 because it's doing things we think are complicated but are just maths and you can automate that.

373
00:34:31,750 --> 00:34:36,630
Whereas the lady on the right that has to do all sorts of things which we find really difficult to codify,

374
00:34:36,680 --> 00:34:44,020
like dealing with troublesome customers, you know, trying to find who really get ordered vegetarian meals, all of that stuff.

375
00:34:44,200 --> 00:34:51,669
Yeah, very poor situations for computers, which means that, you know, there are counterintuitive examples.

376
00:34:51,670 --> 00:34:59,460
So in legal, legal case study reviews and in diagnostics, there's loads of automation happening which is affecting people.

377
00:35:00,110 --> 00:35:06,100
God is the brightest and best the society's adopted at a glance. And yet the fruit pickers and office cleaners are brilliant.

378
00:35:06,160 --> 00:35:11,200
Remarkably difficult to automate, at least economically. And at the top of the picture.

379
00:35:11,500 --> 00:35:14,890
I'd say going back to that things that you codify idea.

380
00:35:15,280 --> 00:35:19,540
So if the odyssey is describing stay or going to be on an interest to being central to strategy.

381
00:35:19,960 --> 00:35:23,200
We struggle to articulate what those things are to ourselves.

382
00:35:23,560 --> 00:35:27,910
The possibility we can reduce those to maths seems a very distant prospect.

383
00:35:28,270 --> 00:35:34,210
So the odds of machine learning doing what we would be happy with a strategy seems quite far away.

384
00:35:34,240 --> 00:35:39,130
Although they will affect strategy because there'll be tools that matter. So understanding relative strengths becomes important.

385
00:35:39,460 --> 00:35:44,290
So machines are faster than us. They are already proving better at detection and recognition.

386
00:35:45,100 --> 00:35:50,620
They are typically more efficient at things that we can focus on and they are better optimising in all sorts of ways,

387
00:35:50,620 --> 00:35:56,030
especially problems which we do we have a lot of data for. We can cycle quickly through how many would say, don't you know,

388
00:35:56,090 --> 00:36:01,580
spending a lot of power getting tired, sleepy or distracted, but they own negative context.

389
00:36:01,630 --> 00:36:06,940
So the ability of a robots to walk into a room and make a determination as to whether that's a child who

390
00:36:06,940 --> 00:36:13,120
has picked up a rifle or a child soldier is extremely difficult and could well just be an intuitive,

391
00:36:13,120 --> 00:36:17,920
almost unconscious act by a soldier walking into a room because the context is important.

392
00:36:19,180 --> 00:36:23,380
Right now, they're pretty bad at doing this big push to try and change that.

393
00:36:24,130 --> 00:36:29,050
But the ability to take lessons in one environment and reapply them in another one is a very powerful YouTube.

394
00:36:29,290 --> 00:36:34,330
And it is no mistake that all of the longest running religions are books of allegory,

395
00:36:34,630 --> 00:36:40,030
because the lessons that were written about worlds that no longer exist can still be applied by humans successfully.

396
00:36:40,430 --> 00:36:45,600
Or that's how that even why they are still found yet facilitated.

397
00:36:45,670 --> 00:36:50,230
This is not within the realm of the machine, but it's new for the obvious reasons that there's no data.

398
00:36:50,680 --> 00:36:55,140
So if it's the first time something has been seen in the machine control, it's versatility.

399
00:36:55,160 --> 00:37:02,799
So you can take a human and you can kind of do all sorts of things moderately well, if not perfectly, but quite often take the machine and put it on.

400
00:37:02,800 --> 00:37:07,270
A new project requires re-engineering and there's a thing called catastrophic forgetting at the moment,

401
00:37:07,270 --> 00:37:14,079
which is if I teach machines to be amazing objects and then I teach it back and it kind of forgets all of the things that it's learned about.

402
00:37:14,080 --> 00:37:20,240
Yes, there are some progress. There's some progress being made to try and put sort of elastic bands in the memory, but it's all pretty nice.

403
00:37:20,320 --> 00:37:24,640
And right now they tend to be as bad as if you started with a completely blank slate machine.

404
00:37:25,750 --> 00:37:32,200
And I'm one of the one with this will probably be one of the first to disappear is going to be complex manual handling and it's if you

405
00:37:32,200 --> 00:37:38,200
want an idea of how difficult something that we have evolved over thousands of years to do really well so we take it for granted,

406
00:37:38,410 --> 00:37:40,240
I would suggest go home, sit on your hand.

407
00:37:40,390 --> 00:37:45,400
The story isn't going where you think it's going to come and try and strike a match and you'll probably go through the

408
00:37:45,400 --> 00:37:51,250
whole box of matches before you get anywhere near successfully doing it because you're relying on unconscious feedback,

409
00:37:51,250 --> 00:37:54,460
pressure modulation. There's a whole raft of things to do.

410
00:37:54,730 --> 00:37:59,379
That right now is why machines are doing squishy fruit rather than actually packaging stuff.

411
00:37:59,380 --> 00:38:06,530
And, you know, firms like a car, if you've managed to automate everything else in that supply chain apart from irregular handling of soft objects.

412
00:38:07,990 --> 00:38:14,020
Oh, and again, the promise of the bottom now, because they still make mistakes that seem completely alien in that we would never make.

413
00:38:15,460 --> 00:38:20,620
So it's going to be a cocktail of human and machine, as one might say, that there isn't just going to be the machines coming out of it.

414
00:38:20,950 --> 00:38:28,600
And how you apply human consciousness as it becomes more economical to have lots of cheap presence means the human consciousness,

415
00:38:29,140 --> 00:38:35,970
naturally and logically must become proportionally scarcer on the battlefield compared to the different points of presence, you know, housing.

416
00:38:36,430 --> 00:38:40,659
And you then start to get to a place where you can't afford to waste it on doing,

417
00:38:40,660 --> 00:38:45,160
filling in Excel spreadsheets or other stuff that is kind of mechanistic.

418
00:38:45,160 --> 00:38:50,709
If done, you're going to have to start to preserve it for doing the context, judgement, things that you know,

419
00:38:50,710 --> 00:38:55,960
the machine is weaker and I could give you an example, it won't just be one black, one machine.

420
00:38:56,230 --> 00:39:01,990
You will have teams that do this probably with some sort of obviously my own symbol is because I had a kind of simple process intelligence

421
00:39:02,320 --> 00:39:07,120
and the brackets are where those people start with their consciousness and their intelligence and what they're thinking about.

422
00:39:07,420 --> 00:39:12,550
And there will be some places. So on the far right there, you have one driver, maybe he's near the French president.

423
00:39:12,970 --> 00:39:17,020
So you really don't want any emergent good ideas for the machine that you weren't expecting.

424
00:39:17,560 --> 00:39:21,160
So that's you sitting there very carefully with one person flying that system.

425
00:39:21,610 --> 00:39:26,920
And if it's complex to do, you'll find multitasking diminishes, which is one of the reasons, you know, have sex when you drive.

426
00:39:27,670 --> 00:39:33,129
So you focus your consciousness that you might find that you are actually controlling one thing, which is, in fact, a swarm.

427
00:39:33,130 --> 00:39:37,680
So you'll do it in concert with a machine to say, take this woman to that which you can give as an instruction,

428
00:39:37,690 --> 00:39:40,749
but the machine needs to coordinate because there's no way you can fly 50 aircraft.

429
00:39:40,750 --> 00:39:44,190
At the same time, it will also be how you put together a package of things.

430
00:39:44,200 --> 00:39:45,669
Save the little bracket on the left.

431
00:39:45,670 --> 00:39:52,060
There might be that we want to have a surveillance system that's not carrying any heavy metal because we wanted to be for a long time.

432
00:39:52,400 --> 00:39:57,890
We want something with a gun on the floor so that we can actually do the shooting and you want them concert.

433
00:39:57,910 --> 00:40:03,740
It's concerted team of. Things that are actually genius to maximise different kinds of strengths.

434
00:40:04,010 --> 00:40:08,780
But you kind of want have one person to join in flight or so that you're minimising your decision, actually.

435
00:40:09,590 --> 00:40:15,740
And then there'll be other things where, for example, the AWACS in the top left you you don't want to have to spend any time thinking about something.

436
00:40:16,190 --> 00:40:18,440
You just wanted to sit there and say. Right.

437
00:40:18,630 --> 00:40:23,270
Their only point is now I'm not going to pay attention to that because looking at a blank space waste of that consciousness,

438
00:40:23,570 --> 00:40:27,380
you will go there, you will fly around in this area and you will shout when something happens.

439
00:40:27,590 --> 00:40:29,510
But otherwise, just let me know you're going to run out of fuel.

440
00:40:30,020 --> 00:40:33,890
So it's going to be how you balance those things and it isn't going to all arrive overnight.

441
00:40:34,010 --> 00:40:37,880
And I think the way it's going to arrive is going to be the kind of series of overlapping phases.

442
00:40:38,060 --> 00:40:40,900
And I couldn't give you a timeframe. I'll tell you, in ten years it'll be this.

443
00:40:41,540 --> 00:40:49,940
And I think if anyone does a very, very, very clever war room, and it's going to probably play out in the same way as air power did.

444
00:40:50,120 --> 00:40:53,810
So the beginning used to have battleships, which were the kings of the sea.

445
00:40:54,200 --> 00:40:59,000
And in the First World War that was you know, that was in control of the oceans and seas.

446
00:40:59,570 --> 00:41:08,210
Between the sea wars, you added enormous amounts of money to get incremental improvements in what were essentially the same systems.

447
00:41:08,510 --> 00:41:12,520
So by the time the Second World War starts, you know, that got they've got better armour,

448
00:41:12,530 --> 00:41:16,690
they can shoot better and they're more accurate, but they've paid a fortune to do that.

449
00:41:17,060 --> 00:41:19,820
And you have the little box type things that are very likely to fall out of

450
00:41:19,820 --> 00:41:23,120
the sky in the First World War to kill the pilots as the other guy was like, to shoot down.

451
00:41:23,570 --> 00:41:27,920
I've had tens or hundreds of dollars for much bigger jumps in technology.

452
00:41:28,280 --> 00:41:34,950
And the three pictures you see, there were a string of battleships, which are tiny little planes flung up and catapulted into a suicide bomber.

453
00:41:36,170 --> 00:41:40,790
The middle picture is a software terminal taking off from HMS Pegasus, which is a World War One aircraft carrier.

454
00:41:41,480 --> 00:41:46,100
And then by 1942, the battleship is still the most amazing thing in close combat.

455
00:41:46,190 --> 00:41:51,310
But it's kind of irrelevant because you can never get that because the aircraft carrier was able to just a bit range.

456
00:41:51,740 --> 00:41:53,330
So I think we'll see the same sort of thing.

457
00:41:53,330 --> 00:41:59,719
So the first things will be stuff to augment what we already do because we're not in a position to understand those

458
00:41:59,720 --> 00:42:06,440
technologies well enough to know which are definitely going to change stuff and then still operate in parallel.

459
00:42:06,710 --> 00:42:11,480
And then I think eventually much of it will just be some things which become redundant because they're no longer.

460
00:42:12,290 --> 00:42:15,979
But I also don't think it's going to just be one box. And the reason is the picture of a radio.

461
00:42:15,980 --> 00:42:19,610
That is because it's kind of more of a component level technology in some ways.

462
00:42:20,060 --> 00:42:24,170
So in the same way as the French probably have the best time to start of World War two,

463
00:42:25,580 --> 00:42:33,049
it was actually the combination of all the artillery and transportation and people all melded together

464
00:42:33,050 --> 00:42:37,780
through the medium of the radio to allow Blitzkrieg to be as successful as it was at the start of my life.

465
00:42:38,840 --> 00:42:45,530
And I have given us a number of people who tend to get very depressed because they see all of the threats of this.

466
00:42:45,860 --> 00:42:50,410
I don't think there is one natural automatic winner. They do have to cross that my next target.

467
00:42:52,310 --> 00:42:57,980
But I think the future is neutral. I don't think any one person naturally owns this.

468
00:42:57,980 --> 00:43:02,810
There are places that have natural advantages because of research that is going on.

469
00:43:03,440 --> 00:43:05,509
If you do things to the cynical science and metrics,

470
00:43:05,510 --> 00:43:13,399
which is the study of patents and the three leading countries of the world finding patents of currently the US and China.

471
00:43:13,400 --> 00:43:16,850
And depending on how you measure it, the order changes there and then the UK.

472
00:43:17,300 --> 00:43:23,780
So being in that bracket is certainly very healthy. But understanding how these things will play out for the fact that it's going to be driven by

473
00:43:23,780 --> 00:43:27,740
the commercial sector largely means that quite a lot of it's going to be available for everyone.

474
00:43:28,160 --> 00:43:32,600
So it's going to be proactive experimenters who make the most of this.

475
00:43:32,610 --> 00:43:39,800
I think in the same way as was true for the emergence of Echo. And speaking of proactive, inventive people,

476
00:43:40,430 --> 00:43:46,940
this is so I I'm going to show you this that's a test from yesterday I was doing and I saw a preview of what's a robot.

477
00:43:49,340 --> 00:43:57,110
Now, if I'm labelling these pages, I think that is volume two on Google.

478
00:43:57,470 --> 00:44:05,450
They were trying to build some driving cars and I'm like really proving, no, I'm not a robot or am I a slave doing data labelling,

479
00:44:05,450 --> 00:44:13,430
all of it, especially since the old CAPTCHA process was beaten in, I think 2017 by a robot.

480
00:44:13,790 --> 00:44:17,060
So there are robots out there that can be other forms of this test.

481
00:44:17,540 --> 00:44:29,380
So it's all going to be about how your clever in getting everything you need. And on that one, I'll get back and see how much this will.