1
00:00:04,170 --> 00:00:08,450
In this episode, The Changing Character of War program welcomes Dr. Kenneth Payne,
2
00:00:08,830 --> 00:00:12,570
a senior lecturer in the School for Security Studies at King's College London.
3
00:00:13,230 --> 00:00:18,810
His talk explores the evolutionary basis of strategic behaviour from chimpanzees to computers,
4
00:00:19,140 --> 00:00:22,620
teasing out the connections between liberal warfare and human evolution.
5
00:00:24,770 --> 00:00:28,730
So the colour coding here is three parts of the book comes in three parts.
6
00:00:28,970 --> 00:00:34,120
The first part is all about evolution or fixes on evolution, and that's my thesis there.
7
00:00:34,130 --> 00:00:39,140
The second part is a bit about culture, which I'm going to gloss over in just three or four slides.
8
00:00:39,470 --> 00:00:43,340
And then the last part is probably the bit that some that might be of most interest to you is the bit
9
00:00:43,340 --> 00:00:47,120
about artificial intelligence and the ways in which artificial intelligence will shape strategy.
10
00:00:47,930 --> 00:00:57,440
And the overall thesis of the book in a nutshell is that there have been and everybody likes to be kind of reductive about revolutions in warfare.
11
00:00:58,370 --> 00:01:04,849
I think if the Toffler's, for example, that posited three revolutions in human history, the agricultural revolution,
12
00:01:04,850 --> 00:01:10,579
the information age, the story, the Industrial Revolution, and the Information Age Revolution I wouldn't bet it.
13
00:01:10,580 --> 00:01:17,450
The Toffler's beginning for two revolutions, a cognitive revolution early in human history that some, I argue,
14
00:01:17,450 --> 00:01:24,530
set up the basis for some underlying and enduring principles of strategy in the sense that most of us would understand strategy.
15
00:01:25,370 --> 00:01:30,200
And then an artificial intelligence revolution that challenges the evolved
16
00:01:30,830 --> 00:01:36,020
cognitive psychology that had been a longstanding part of our human history.
17
00:01:36,440 --> 00:01:45,920
So two revolutions, the evolution of a cognitive package to do with language and and theory of mind,
18
00:01:46,220 --> 00:01:51,470
and then latterly a revolution in artificial intelligence, which is just getting underway.
19
00:01:52,040 --> 00:02:01,040
So part one and the bits about evolved psychology, my thesis is that there is a dynamic relationship between strategy and evolution.
20
00:02:01,040 --> 00:02:06,079
That is to say that the need to think strategically was a strong factor in shaping our human
21
00:02:06,080 --> 00:02:11,870
evolution and also the way in which we evolved in pace with constraints on how we did strategy.
22
00:02:12,680 --> 00:02:20,180
Hopefully I wouldn't say too much about what strategy is, but in a nutshell, for those of you that might not be familiar with the term,
23
00:02:20,390 --> 00:02:27,080
I'm talking here primarily about military strategy, the use of organised social violence in pursuit of some political goal.
24
00:02:28,510 --> 00:02:36,250
My thesis is that some another Tom Jones who is not my thesis is not even Tom's, but he's done some tremendous work on it.
25
00:02:36,430 --> 00:02:44,650
The general thesis is that Greek surmise has an important effect on winning or losing in combat,
26
00:02:44,950 --> 00:02:56,230
and the need to cooperate together in progressively larger human groups has been one of the fundamental driving forces shaping human evolution.
27
00:03:00,160 --> 00:03:09,819
I got started on this project reading a book that I commend you all in the late seventies and early 80 seconds devours chimpanzee politics.
28
00:03:09,820 --> 00:03:18,550
If you haven't seen that. Grab a copy of that highly entertaining account of life and death in a chimpanzee colony colony in Arnhem,
29
00:03:19,240 --> 00:03:28,390
and the game's inverted commas that the chimpanzees played in order to get ahead and assume a dominant position within the hierarchy,
30
00:03:28,720 --> 00:03:32,080
colony and devil at the end of the book asks the question.
31
00:03:32,080 --> 00:03:36,550
Interesting question Do chimpanzees think strategically?
32
00:03:37,120 --> 00:03:42,790
And he answers that, yes, they do think strategically, but I think it is a slight oxymoron.
33
00:03:43,210 --> 00:03:48,340
They do so instinctively. There's not he doesn't necessarily suppose,
34
00:03:48,340 --> 00:03:54,129
a particularly sophisticated degree of self-reflection and reflection about the minds of
35
00:03:54,130 --> 00:03:59,560
others on the part of chimpanzees is much of their strategic behaviour is innate to them,
36
00:03:59,560 --> 00:04:00,970
is instinctively acquired.
37
00:04:01,720 --> 00:04:09,550
And I think not that I want to disagree on this eminent primatologists in the world, but I think it is something of a disservice to chimpanzees.
38
00:04:09,910 --> 00:04:16,720
And I think the view on chimpanzee cognition and theory of mind has moved on since the late seventies.
39
00:04:16,990 --> 00:04:24,220
So on the one hand, he's doing a disservice to chimpanzees because they are capable of thinking about the intentions of others,
40
00:04:24,280 --> 00:04:28,120
not as not to the same degree that we are, but they are to a certain degree.
41
00:04:28,660 --> 00:04:30,549
And on the other hand, he does limit service,
42
00:04:30,550 --> 00:04:38,940
because I think there's is the supposition that we as humans think in rational and calculating terms about strategy is unfair.
43
00:04:38,950 --> 00:04:47,860
We do a lot of unconscious, instinctive, horrific type thinking about strategy, just as the chimpanzees do.
44
00:04:48,610 --> 00:04:53,170
So Deval set me going and the question was, do animals fight wars?
45
00:04:53,380 --> 00:05:00,310
And if they do, do they fight wars in the strategic sense? Do they make use of violence in an instrumental and passive way?
46
00:05:00,700 --> 00:05:11,290
And my answer was this was answer fight pretty totalitarian wars of the entire destruction of other and colonies in other species,
47
00:05:11,650 --> 00:05:15,850
that they do so on a fairly automatic situational basis.
48
00:05:15,850 --> 00:05:21,670
They're responding in an evolved fashion to instinctive clues from the environment about how they ought to behave.
49
00:05:22,270 --> 00:05:26,290
Chimpanzees, on the other hand, certainly fight wars come on tonight in the second.
50
00:05:26,620 --> 00:05:34,270
And then he did so with a little bit more of what we might call strategic thought, a little bit more conscious reflection on what they're doing.
51
00:05:36,070 --> 00:05:41,170
And so that was my starting point. Way too much about the middle part of the book.
52
00:05:41,170 --> 00:05:45,570
Middle part of the book is a dilemma coming to anybody looking for general patterns in strategic behaviour.
53
00:05:45,580 --> 00:05:47,740
It's you've got the whole gamut of human history.
54
00:05:47,920 --> 00:05:54,820
What can you say about strategic behaviour that unites strategy across all these different varieties and cultural circumstances?
55
00:05:55,360 --> 00:06:02,200
That's book classes wrestle with, of course, is the tension between historicism something that's particular to the moment in theory,
56
00:06:02,200 --> 00:06:07,500
the ability to say something general across all particular areas or multiple areas.
57
00:06:08,080 --> 00:06:14,229
The second tension that the tension between historicism, the local, the particular and history with a capital H,
58
00:06:14,230 --> 00:06:19,120
the idea that history is driving forward towards some particular purpose.
59
00:06:19,840 --> 00:06:23,079
That that's the that's the tension I wrestle with in the middle part of the book,
60
00:06:23,080 --> 00:06:27,700
the bit about culture that I'm not going to spend too much time thinking thinking about here today.
61
00:06:28,090 --> 00:06:32,650
And then lastly, the last part of the book, can artificial intelligence strategy save that for the end?
62
00:06:34,270 --> 00:06:38,659
Here to kick it off, we come to chimpanzees and their ability to think strategically.
63
00:06:38,660 --> 00:06:45,250
This is a paper in 2014 in which the chimpanzees demonstrated an ability that might have
64
00:06:45,250 --> 00:06:51,400
appealed to social scientists from the middle of the 20th century to win a game of strategy.
65
00:06:51,640 --> 00:06:54,760
This is the game they're playing here is a little bit like pieces of rock.
66
00:06:55,000 --> 00:06:58,410
You have to guess what your adversary actually can't see.
67
00:06:58,420 --> 00:07:02,170
The chimpanzee is play. It's the chimpanzee here. Is that back to back?
68
00:07:02,170 --> 00:07:06,190
So they can't see what the adversary is choosing. One of them is tasked with matching the shapes.
69
00:07:06,190 --> 00:07:11,320
And and if he guesses, if he matches the shapes that the other chimp throws, he wins.
70
00:07:11,680 --> 00:07:14,299
The other is tasked with mismatching the shapes that the chimp race.
71
00:07:14,300 --> 00:07:19,540
So he knows he's got to think about what the other chimp is going to throw and then produce a mismatch.
72
00:07:19,870 --> 00:07:21,879
So you've got a strategic game to play,
73
00:07:21,880 --> 00:07:26,920
a strategic game in which they've got to try and guess what the other person going to do and behave accordingly.
74
00:07:27,610 --> 00:07:37,360
Chimpanzees outperform humans at this game, which if you were a scholar of the standard social science model of the rational activity,
75
00:07:37,360 --> 00:07:43,870
expected utility payoffs, that would be a puzzle. Why are the chimpanzees so good at it and why the humans are less good at it?
76
00:07:43,870 --> 00:07:49,899
And if you change the payoffs or if you change whether one's a match or a mismatch and they tell them very quickly,
77
00:07:49,900 --> 00:07:58,800
the chimpanzees change tack and start behaving according to how the change suggests that they should on a mathematical basis, they come.
78
00:07:58,860 --> 00:08:02,760
Very close to the optimal mathematics solution that humans are hopeless at it comparatively.
79
00:08:03,510 --> 00:08:11,610
The question is why and what does that mean? And the answer for me is not that chimpanzees are better strategists.
80
00:08:12,030 --> 00:08:20,280
It's that they're pursuing a different strategic logic. The chimpanzees are responding more instinctively to the payoffs.
81
00:08:20,550 --> 00:08:23,910
The humans are trying to second guess what the adversary is doing.
82
00:08:24,720 --> 00:08:27,960
Humans are making that strategy using theory of mind.
83
00:08:27,990 --> 00:08:34,450
They're trying to play the idea that, you know, the innate game and the chimpanzees aren't playing that game.
84
00:08:34,470 --> 00:08:40,830
They just want to sweep that they get for winning the game. And it's the over complication that we put into it,
85
00:08:40,830 --> 00:08:47,040
trying to read what the intention of others are the producers in this narrowly to play game a suboptimal outcome.
86
00:08:49,650 --> 00:08:54,060
So chimpanzees more rational actors than humans are in that game.
87
00:08:55,500 --> 00:09:00,329
Chimpanzees, we know because they're evolutionarily close to us.
88
00:09:00,330 --> 00:09:08,819
We parted company for them some 6 million years ago. Very useful comparator in terms of thinking about strategy.
89
00:09:08,820 --> 00:09:13,500
And in answer to Duval's question, you know, do chimpanzees fight wars?
90
00:09:13,740 --> 00:09:18,540
When he wrote that in 79, I think it was the goomba extreme war had just ended.
91
00:09:18,630 --> 00:09:21,720
I'm not sure whether where they could all have actually published about it,
92
00:09:21,720 --> 00:09:27,300
but the dominant view at the time in anthropology and actually in climatology as well was still,
93
00:09:27,420 --> 00:09:30,990
you know, chimpanzees were fairly peaceful, fun loving kind of characters.
94
00:09:31,740 --> 00:09:40,080
Since when that view has changed almost 180 degrees and the survey that shouldn't say we're going to say random, of course, Richard Wrangham,
95
00:09:40,950 --> 00:09:42,560
famous primatologist at Harvard,
96
00:09:43,410 --> 00:09:52,270
it turns out that some chimpanzee life is extremely violent at best interest group when you're struggling for dominance as it develops.
97
00:09:52,320 --> 00:09:56,790
Chimpanzees were on them and also between groups in terms of these chimpanzee wars.
98
00:09:57,330 --> 00:10:05,010
And the the paper by Wrangham that compiles all the available evidence around in 2006
99
00:10:05,730 --> 00:10:10,020
about how many wars were fought and what the levels of casualties were in these wars.
100
00:10:10,020 --> 00:10:16,860
And it turns out the chimpanzee life inside the group is much, much more violent than Detroit in the 1970s,
101
00:10:17,580 --> 00:10:24,120
and that some the levels of war were between chimpanzee groups allowing for small
102
00:10:24,120 --> 00:10:28,259
sample size was quite likely to be on a par with hunter gatherer communities.
103
00:10:28,260 --> 00:10:30,060
And we'll see what that was in the moment.
104
00:10:30,300 --> 00:10:38,400
Here's some chimpanzees behaving you might think strategically, perhaps in response to innate subconscious inclinations.
105
00:10:39,120 --> 00:10:45,779
They're conducting a chimpanzee patrol here and they patrol into the territory of neighbouring chimpanzees and they're doing the tactics
106
00:10:45,780 --> 00:10:54,420
when it comes to waging a war like the going to extreme war is the raid and ambush that typically attack the 10 to 1 force ratios.
107
00:10:55,410 --> 00:11:02,040
And so if they can't not heavily outnumber their adversary, they'll stage a silent retreat back away until such time as they do.
108
00:11:02,040 --> 00:11:07,860
WRANGHAM Theory is that this behaviour, which contrasts markedly with that of bonobos,
109
00:11:08,760 --> 00:11:11,430
they tend to fight wars, spend a lot of time having a good time.
110
00:11:13,320 --> 00:11:18,990
Range theory is that the distribution of food which forced chimpanzees to forage in an isolated fashion,
111
00:11:18,990 --> 00:11:22,980
may be particularly vulnerable to predation in this way by their fellow chimpanzees,
112
00:11:23,310 --> 00:11:31,170
because they had to fissure into small groups to go foraging for food, they could very quickly be outnumbered and defeated by other chimpanzees.
113
00:11:31,980 --> 00:11:36,660
The question is, is that strategic behaviour or not?
114
00:11:37,530 --> 00:11:44,759
On the one hand, the chimpanzees are capable of waging a sustained war against another group of chimpanzees.
115
00:11:44,760 --> 00:11:48,930
In that extreme war they annihilated the other group of chimpanzees.
116
00:11:48,930 --> 00:11:51,360
That survey every last chimpanzee.
117
00:11:51,400 --> 00:11:59,130
I think there were a couple of females left over that were absorbed back into the original population and they waited over a period of years,
118
00:12:01,800 --> 00:12:04,470
you might say, well, that's not necessarily strategic behaviour and you know,
119
00:12:04,500 --> 00:12:10,889
they're following an evolved inclination to conduct these patrols and to stage ambushes and I think you'd probably be right about that.
120
00:12:10,890 --> 00:12:19,110
For me, the big difference that makes our behaviour more strategic in the sense that we're familiar with is our capacity for this theory of mind,
121
00:12:19,110 --> 00:12:27,000
for intuiting what other people want and communicating on the basis of that in order to derive an effective strategy.
122
00:12:27,840 --> 00:12:31,049
So I start with chimpanzee wars and I say yes.
123
00:12:31,050 --> 00:12:35,010
On the one hand they can do instinctive strategy, but there's something different about humans.
124
00:12:35,400 --> 00:12:42,780
And what's different about humans is our ability to communicate and to plan for an imagined future.
125
00:12:43,950 --> 00:12:51,269
And the basic theory that elaborate in part one is that there is enough violence in
126
00:12:51,270 --> 00:12:54,870
the hunter gatherer world in which we evolved for almost all of our human history.
127
00:12:55,290 --> 00:12:58,820
There is a sufficiently high level of violence like in that chimpanzee world.
128
00:12:59,610 --> 00:13:04,170
To force us to become effective co-operators, we have to cooperate.
129
00:13:04,440 --> 00:13:08,160
Otherwise, we go out of business. That's essentially the theory.
130
00:13:08,880 --> 00:13:14,790
So, as you might know, Robin Dunbar's work here may have heard Dunbar's number.
131
00:13:15,390 --> 00:13:18,690
Dunbar's number? Yeah. And under the back? I hope not.
132
00:13:18,690 --> 00:13:25,559
Robin Dunbar. I've never met Robin Dunbar. So. So Dunbar's number is the size.
133
00:13:25,560 --> 00:13:33,660
Your neocortex correlates very effectively with the size of social group in which you live in the larger neocortex, the theory goes.
134
00:13:34,230 --> 00:13:38,490
The greater the ability. You have to track the key essentials of life in a social group.
135
00:13:38,790 --> 00:13:42,690
And these key centres are essentially who's up, who's down and who's shagging who.
136
00:13:42,900 --> 00:13:47,430
These are the things that you won't be able t t to understand.
137
00:13:48,150 --> 00:13:55,670
And if you if you correlate a chimpanzee neocortex sized chimpanzee cortex size relationship,
138
00:13:55,980 --> 00:13:59,370
you can extend that line outwards to humans and come up with what's called Dunbar's number.
139
00:13:59,400 --> 00:14:09,300
Dunbar's number for humans is 150. That's the size of hunter gatherer community in which we spent a lot of our early human evolution.
140
00:14:09,540 --> 00:14:15,180
It's a very rough rule of thumb because actually the site you hang out with on a day to day basis would be near 50 people.
141
00:14:15,600 --> 00:14:19,710
That would be your kinship band, people you primarily be related to.
142
00:14:20,310 --> 00:14:25,140
And you come together in this gathering of 150 for an overnight camp, perhaps for reasons of security.
143
00:14:26,250 --> 00:14:35,459
Dunbar's number is one aspect of the theory of human evolution called the social theory of cognitive evolution,
144
00:14:35,460 --> 00:14:43,170
which says a lot of our functions are derived to help us navigate this complex, increasingly co-operative,
145
00:14:43,380 --> 00:14:49,170
interdependent world 50 to 150 people with whom we would spend most of our time.
146
00:14:49,530 --> 00:14:52,860
That's the group that we hung out here now.
147
00:14:55,930 --> 00:15:03,520
The second big theory to think about is Frederick Manchester's mathematical laws of armed conflict dating to 1916.
148
00:15:03,520 --> 00:15:09,330
Anybody familiar with those yet to be familiar with? They say without without stretching the point.
149
00:15:09,410 --> 00:15:12,879
Point one, this is an abstraction. It's a mathematical abstraction.
150
00:15:12,880 --> 00:15:16,390
It's not meant to correlate precisely to real world behaviour.
151
00:15:17,080 --> 00:15:26,150
But Manchester's argument was that the benefits that would accrue from being in a larger group would be disproportionate to the size of that group.
152
00:15:26,770 --> 00:15:39,190
That's what he call square. If fighting is structured so that it's a melee, basically a wall against all, then the fighting power of the group,
153
00:15:39,190 --> 00:15:44,019
all of the things being equal will increase geometric t relative to the size of the group.
154
00:15:44,020 --> 00:15:50,469
The advantages from being in a larger group will increase geometrically if fighting is structured.
155
00:15:50,470 --> 00:15:53,650
So it's one on one. So imagine a series of duels at dawn.
156
00:15:54,490 --> 00:16:04,200
Or imagine, as I do later in the book, Hoplite Warfare, where the formations force you to fight 1 to 1 rather than in an overall melee.
157
00:16:04,930 --> 00:16:09,790
Then it still matters to be big, but it matters in a linear fashion.
158
00:16:10,720 --> 00:16:17,740
Say provided fighting is disorganised or there is a capacity to concentrate fire, not renown,
159
00:16:17,740 --> 00:16:23,830
strategic principle, concentrating force where you can then the fighting power increases geometrically.
160
00:16:24,550 --> 00:16:32,170
So the question is, is, is there enough fighting in hunter gatherer communities early in human evolution to
161
00:16:32,170 --> 00:16:38,680
force the agglomeration of humans together into larger groups and in larger groups?
162
00:16:38,680 --> 00:16:40,750
Perforce, they have to fight.
163
00:16:40,960 --> 00:16:48,640
You have to be in a group with people to whom you're not directly related, and you have to find a way of cooperating with strangers.
164
00:16:49,660 --> 00:16:54,580
And the answer I make, the claim I make in the book and I'm not alone in making plenty of other people make it as well,
165
00:16:54,580 --> 00:17:00,790
is that, yes, there was enough fighting in hunter gatherer times to force people to cooperate together.
166
00:17:02,140 --> 00:17:07,780
So the result I suggest is that there was sufficient violence early in human history to put
167
00:17:07,780 --> 00:17:13,660
pressure on humans to start cooperating with people with whom they might not know each other.
168
00:17:13,760 --> 00:17:17,230
Manchester's law provides a mathematical formulation of that.
169
00:17:17,590 --> 00:17:23,200
That is to say the prevalence of war solves what's known as the free riding problem.
170
00:17:23,650 --> 00:17:30,940
Free riding problem is essentially a problem of loafing. If you get enough people together, there'll always be some people who slack off.
171
00:17:31,270 --> 00:17:36,220
And if it's a matter of fighting somebody to the death, I personally would be on the side of the slacker office.
172
00:17:36,970 --> 00:17:40,600
So what is it that's going to make me cooperate?
173
00:17:40,600 --> 00:17:44,050
That's going to stop me from slacking off?
174
00:17:44,320 --> 00:17:48,100
One answer is to this the prevalence of war forces,
175
00:17:48,730 --> 00:17:55,420
cooperation and scrutiny of how people are behaving within a particular book say the problem to be solved,
176
00:17:56,050 --> 00:18:03,879
you'll be aware, is the classic problem of the prisoner's dilemma. And the payoffs is just in the prisoner's dilemma,
177
00:18:03,880 --> 00:18:09,730
such that non-cooperation tends to be the favoured option, even if cooperating would be better for you.
178
00:18:10,150 --> 00:18:14,020
The solution to that problem, as Samuel Bowles argues, as Peter Turchin argues,
179
00:18:14,290 --> 00:18:20,200
is that there is sufficient violence to force the read out that on co-operators and to force cooperation.
180
00:18:20,620 --> 00:18:23,799
And actually you can see echoes of that in a couple of days.
181
00:18:23,800 --> 00:18:30,730
Research is there. So the basic theory of cooperation in an evolutionary sense is from Robert TRIVERS
182
00:18:30,730 --> 00:18:35,680
Trivers argues that we cooperate with non kin on the basis of reciprocal altruism.
183
00:18:37,030 --> 00:18:43,750
That is, if I scratch your back, I do so with the supposition that in years to come you might scratch my back.
184
00:18:43,960 --> 00:18:48,580
So I will behave altruistically to you on the assumption that I'll get some payback later on.
185
00:18:48,580 --> 00:18:51,070
That's the basic theory of reciprocal altruism,
186
00:18:51,310 --> 00:18:57,730
and that requires you to have a make a pretty good bet about it that the probability of being paid back in that future date,
187
00:18:58,060 --> 00:19:05,140
of course, the incentive then is to cheat, to get away with having your back scratched without returning the favour, so to speak.
188
00:19:05,470 --> 00:19:13,750
And Dan Ariely and Travis himself have demonstrated that very ably trivers in the in the world of animals.
189
00:19:13,990 --> 00:19:24,310
Ariely, in some fascinating experiments on his students, proved that our inclination is to cheat and then lie about it.
190
00:19:25,030 --> 00:19:33,099
But what actually proved this is quite interesting is that we lie about it not just to the people we're cheating, we lie about it to ourselves.
191
00:19:33,100 --> 00:19:38,170
So we have the capacity to persuade ourselves that we are good, honest co-operators, even while we're cheating.
192
00:19:39,100 --> 00:19:43,360
So we can see that the natural inclination is not to be a free,
193
00:19:43,360 --> 00:19:47,470
cooperative member of society, even though the pressure of war suggests that we should be.
194
00:19:48,460 --> 00:19:53,950
But there is, because of that pressure of war, I argue, still some inclination to cooperate.
195
00:19:54,280 --> 00:19:55,300
And you can see that.
196
00:19:55,330 --> 00:20:05,170
When you do real world experiments with the prisoner's dilemma, in theory, in game theoretical terms, the incentive is to cheat and not to cooperate.
197
00:20:05,770 --> 00:20:12,370
In practice, if you give the prisoner's dilemma to normal people, they tend to cooperate, even where it's a one shot game.
198
00:20:12,370 --> 00:20:18,550
So there's no possibility of playing the same person again. So you can't build a reputation for trust that might shape your behaviour.
199
00:20:18,880 --> 00:20:21,100
Was a one shot game with a stranger.
200
00:20:21,310 --> 00:20:28,690
We have a tendency to wait to cooperate and the argument is that tendency to cooperate has evolved within us as a result,
201
00:20:28,690 --> 00:20:35,890
as I say, these pressure of warfare. And in fact, even if you give the prisoner's dilemma to genuine real life prisoners who ought to know better,
202
00:20:36,340 --> 00:20:39,880
they tend to cooperate more than is suggested by the prisoner's dilemma.
203
00:20:42,810 --> 00:20:51,180
So this evolutionary picture then that I take out of the first part combines this notion of reciprocal altruism.
204
00:20:52,200 --> 00:20:53,510
Lanchester His idea,
205
00:20:53,540 --> 00:21:05,040
the victory goes to the big battalions in a disproportionate sense and comes up with what I think is the the dominant approach to strategy,
206
00:21:05,040 --> 00:21:08,430
which is to be strong everywhere, if you're strong everywhere. Of course, strategy is very easy.
207
00:21:09,270 --> 00:21:15,750
And the strategy that the strong evolved with involves scale, mass and concentration of force.
208
00:21:16,230 --> 00:21:19,860
And and that couldn't I gave you Voltaire a moment ago.
209
00:21:20,130 --> 00:21:22,680
Victory doesn't belong to the big chains. It belongs to the best shooters.
210
00:21:22,800 --> 00:21:28,740
Lanchester would disagree with that, and I would suggest the evidence from evolutionary history is against that as well,
211
00:21:28,740 --> 00:21:34,350
which we generally get used to the big battalions, as Stone said of the pope.
212
00:21:34,590 --> 00:21:41,160
How many divisions? SIEGEL And say there is still that in-built supposition that mass pays off in the end.
213
00:21:42,300 --> 00:21:45,450
But you will say, you know, even if you're strong, you don't want to be a muffett.
214
00:21:45,450 --> 00:21:50,009
You also want to combine your strategy, the strong with being strong everywhere with some sensible tactics.
215
00:21:50,010 --> 00:21:51,180
And that's what the chimpanzees do.
216
00:21:51,180 --> 00:22:00,000
They make they in conducting a raid and an ambush rather than pitched battle against the same number of adversary chimpanzees.
217
00:22:00,360 --> 00:22:10,100
What they're doing effectively is building a force ratio that Lanchester would say is dominant to create a localised force ratios,
218
00:22:10,110 --> 00:22:15,960
even a strong use rate and an ambush. And if you look at this book wall before civilisation,
219
00:22:16,500 --> 00:22:22,470
that pattern is very strongly evident across ethnographic studies and archaeological studies
220
00:22:22,740 --> 00:22:27,240
that the dominant strategy in warfare is rate an ambush rather than pitched battle.
221
00:22:28,560 --> 00:22:32,100
And if you're on the on the weak side, that's when strategy gets really interesting,
222
00:22:32,100 --> 00:22:35,970
because you have to find a way of offsetting these advances to decrease the others.
223
00:22:36,300 --> 00:22:42,240
So you adopt all these sort of approaches and surprise, deception and fortification, manoeuvre.
224
00:22:42,900 --> 00:22:48,330
And if you are the borrowers, as the Americans call the British army, you come up with the British way of warfare,
225
00:22:48,330 --> 00:22:52,809
which is to try and be more flexible and you get throwing about it and reverse.
226
00:22:52,810 --> 00:23:02,890
Three. Max Fraser if you don't know, Max raises work and you should.
227
00:23:03,670 --> 00:23:09,010
He's a statistician at Oxford and he's produced extensive database on all sorts of fascinating stuff.
228
00:23:09,010 --> 00:23:15,159
The world, the world and figures and the general picture is that if he has a general motif is that the world is getting a better place,
229
00:23:15,160 --> 00:23:18,910
which is not always how it feels. A day's out for an American election,
230
00:23:19,870 --> 00:23:30,309
but he has compiled a whole range of statistical information about it's the most comprehensive list that I've seen of archaeological
231
00:23:30,310 --> 00:23:37,990
evidence for prevalence of warfare and hunter gatherer communities and ethnographic studies of warfare in hunter gatherer communities.
232
00:23:38,740 --> 00:23:47,080
And on the right here, you can't make out any of the detail here, but on the right this is violent deaths per 100,000 in state societies.
233
00:23:47,440 --> 00:23:52,000
Down at the bottom here is the world in 2007.
234
00:23:52,540 --> 00:23:56,110
So that's that's that's the the [INAUDIBLE] pit that we live in at the moment.
235
00:23:56,680 --> 00:24:03,910
Battle deaths 0.33 deaths per 100,000 people for the world in 2007.
236
00:24:04,750 --> 00:24:11,110
Whereas if you look at this hunter gatherer community study in the 1840s in California.
237
00:24:14,430 --> 00:24:21,569
1450 deaths per hundred thousand, saying hunter gatherer communities tend to have the most violent men,
238
00:24:21,570 --> 00:24:27,910
but they tend to have more violence, more violent deaths than do more modern state communities.
239
00:24:27,930 --> 00:24:34,320
So was there enough warfare in the course of human history to drive consumption behaviour and get rid of free riders?
240
00:24:34,980 --> 00:24:40,110
The answer for me is yes and that's the claim I take out of the first part of the book.
241
00:24:42,270 --> 00:24:48,120
I didn't spend a bit of time thinking through the implications of what that means for human cognition and
242
00:24:48,120 --> 00:24:53,640
for the heretics that we've developed to help us think strategically when we go into any of that here.
243
00:24:54,480 --> 00:24:59,010
That's part one of the book and an evolutionary history of warfare. It pays to be big.
244
00:24:59,010 --> 00:25:06,780
There's enough fights around to force us to agglomerate in large groups and to force it to learn patterns of cooperative
245
00:25:06,780 --> 00:25:14,580
behaviour and shortcuts for thinking about who we trust and who we shouldn't trust to allow us to think strategically.
246
00:25:15,180 --> 00:25:22,120
In part to the book. I take that forward and see if it holds in the cultural era.
247
00:25:22,560 --> 00:25:26,640
And the answer is that it kind of does.
248
00:25:27,160 --> 00:25:30,330
But stuff complicates it for reasons of time. I'm not going to get into.
249
00:25:30,330 --> 00:25:33,900
I use three fairly well-trodden examples.
250
00:25:34,200 --> 00:25:36,509
It's all about the Greeks and hoplites.
251
00:25:36,510 --> 00:25:44,910
And I talk about the invention of writing and particularly the invention of strategic studies as a separate way of thinking reflectively about war.
252
00:25:45,210 --> 00:25:51,630
And the take out from that study. The Greeks is that writing provided a first example of artificial intelligence,
253
00:25:51,630 --> 00:25:56,250
if you like, a means of offshoring, part of our cognition outside of our brain.
254
00:25:56,610 --> 00:26:01,870
I won't say any more about that example to Clausewitz and Napoleonic warfare.
255
00:26:01,890 --> 00:26:06,360
Again, I'm not going to say too much more about that. I will say a little bit about nuclear weapons.
256
00:26:07,980 --> 00:26:17,010
And the take out from the middle part of the book is that Manchester's law breaks down in the cultural era because
257
00:26:17,310 --> 00:26:23,280
fighting power is not just about the number of men that you have at your disposal and their ability to concentrate force.
258
00:26:23,880 --> 00:26:31,410
It's also about technology in the cultural areas technologies of warfare, weapons technologies, but not just weapons technologies.
259
00:26:31,800 --> 00:26:41,430
Technology screws Manchester's abstract thinking about the importance of scale when it comes to strategy.
260
00:26:41,670 --> 00:26:50,729
And the apotheosis of that, of course, is nuclear weapons, where the analogy can inflict as much damage as a thousand plane fighter bombing raid.
261
00:26:50,730 --> 00:26:55,320
On take, you say scale, which you have nuclear weapons becomes less of an issue.
262
00:26:55,830 --> 00:26:59,190
Of course, you need scale in order to develop nuclear weapons in the first place.
263
00:27:01,340 --> 00:27:07,100
I said at the beginning, I posited two revolutions in military affairs, the one the cognitive revolution,
264
00:27:07,140 --> 00:27:12,110
the flexible package to construct grips flexibly in order to deal with violence in the world.
265
00:27:12,590 --> 00:27:25,130
And second, artificial intelligence. And one of the other contenders for a revolution in military affairs that changed everything is nuclear weapons.
266
00:27:25,190 --> 00:27:30,910
Robert Jervis, many of you will know, said that nuclear weapons were a revolutionary development in warfare.
267
00:27:30,930 --> 00:27:33,290
They were the revolutionary development in warfare.
268
00:27:33,650 --> 00:27:40,220
Bernard Brody, the American strategic thinker on hearing about the destination of the first nuclear bomb,
269
00:27:40,640 --> 00:27:45,920
said that he thought it made everything else he had ever written redundant and obsolete.
270
00:27:46,220 --> 00:27:51,140
He thought that the entirety of strategic history was redundant and obsolete because nuclear weapons are transforming so much.
271
00:27:52,400 --> 00:27:59,750
I think he was wrong. He was wrong. But I also think that this is wrong in the sense that nuclear weapons are revolutionary.
272
00:28:00,860 --> 00:28:03,170
They're an important development in the history of warfare, for sure.
273
00:28:03,410 --> 00:28:07,800
But I didn't think the revolution and the reason is at the bottom of their strategy.
274
00:28:07,820 --> 00:28:13,280
Even with nuclear weapons remains a matter of our evolved human psychology.
275
00:28:13,610 --> 00:28:17,030
We still making decisions about the employment of nuclear weapons,
276
00:28:17,390 --> 00:28:25,430
using precisely the same cognitive architecture that we evolved with 4050, 200,000 years of our human evolution.
277
00:28:25,460 --> 00:28:34,160
So all these factors in nuclear strategic theory, escalation, the need to demonstrate credibility, miscalculation,
278
00:28:34,170 --> 00:28:41,480
the scope for misperception, they are all subject to the same evolved cognitive, psychological foibles that we brought with us before.
279
00:28:42,470 --> 00:28:48,860
Jervis said that nuclear weapons were revolutionary on the basis that they would transform the balance between denial and punishment.
280
00:28:49,160 --> 00:28:57,280
And of course, you have a nuclear standoff through mutually assured destruction, and the threat of retaliation is how you use the weapon.
281
00:28:57,290 --> 00:29:03,350
So is the threat of using the weapon rather than the actual use of the weapon that delivers you more strategic result.
282
00:29:03,740 --> 00:29:08,330
And what he said is there's no point trying to deny nuclear weapons.
283
00:29:08,330 --> 00:29:15,590
You can't defend against them. And instead, your strategic use of nuclear weapons relies entirely on the threat of punishment.
284
00:29:17,760 --> 00:29:25,770
I'm not entirely sure that is revolutionary. Firstly, because denial and punishment both featured before the onset of nuclear weapons.
285
00:29:26,070 --> 00:29:30,810
You could certainly deny through the construction of fortification or the use of terrain, use of things.
286
00:29:31,140 --> 00:29:35,130
And you could also threaten punishments through retaliatory attack.
287
00:29:35,160 --> 00:29:40,800
You didn't need nuclear weapons. Individual nuclear weapons gave you an assured ability to punish.
288
00:29:41,520 --> 00:29:45,480
But punishment was certainly a feature of strategy even before the development of nuclear weapons.
289
00:29:46,320 --> 00:29:51,420
So I'm not sure in Genesis account that nuclear weapons are particularly revolutionary.
290
00:29:51,890 --> 00:29:55,980
That's what I'll say about them here. Reasons of time.
291
00:29:56,790 --> 00:29:59,669
I wanted to get ahead to the artificial intelligence because I'm aware that that's
292
00:29:59,670 --> 00:30:02,910
probably why most of you here not to hear me rambling on about chimpanzees warfare.
293
00:30:03,580 --> 00:30:12,810
So in very quick order, my headline for artificial intelligence is that it is a revolution in military affairs,
294
00:30:12,990 --> 00:30:19,560
is a revolution in strategic affairs precisely because it challenges the underlying evolved psychological
295
00:30:20,820 --> 00:30:26,670
abilities that we've brought to bear to tackle strategic problems since we emerged on the planet and even before,
296
00:30:26,820 --> 00:30:34,290
you know, antecedent species. Here are some examples of artificial intelligence from recent months.
297
00:30:34,860 --> 00:30:39,570
Let's go with the easy ones. Do you believe what's going on there? Yeah.
298
00:30:39,990 --> 00:30:43,470
Google's AlphaGo. Yeah. The part of the world.
299
00:30:43,620 --> 00:30:47,399
The world leading go. Yeah. Did you release it all for one? The World Championship.
300
00:30:47,400 --> 00:30:52,889
And before that? Very quickly. We very quickly adopt hindsight bias with achievements like that.
301
00:30:52,890 --> 00:30:59,580
And assuming, of course, it beat him before that pretty well. Every A.I. expert at DeepMind was saying it couldn't be done,
302
00:30:59,580 --> 00:31:06,930
couldn't be done for 20 years because they'd all been trying to do it goes right for those you don't know who wins or version where the ball game.
303
00:31:07,200 --> 00:31:11,160
It's a very simple constrained what now developer would call a toy universe.
304
00:31:11,370 --> 00:31:14,160
It's not like the complex, messy reality of winning in Vietnam.
305
00:31:14,490 --> 00:31:19,350
There are only a certain number of moves that can take place at two different varieties moves that can take place on the keyboard,
306
00:31:19,650 --> 00:31:27,690
but they can do so with such dazzling combinatorial complexity that there are more possible moves mathematically than there are atoms in the universe.
307
00:31:28,560 --> 00:31:32,010
So you can't solve a game of going through brute force calculation,
308
00:31:32,340 --> 00:31:36,270
which is why A.I. developers have said it would be 20 odd years before anybody came
309
00:31:36,270 --> 00:31:40,830
close to performing at the level of the world champion still is beating him anyway.
310
00:31:41,430 --> 00:31:44,520
Within DeepMind, they were pretty confident they trash him and they truly did.
311
00:31:45,600 --> 00:31:51,450
So that's not one this one you. But you know, that one is a wreckage or helicopter flying upside down.
312
00:31:51,450 --> 00:31:56,730
But more significantly, the Knights and rings, remote control helicopter flying upside down from 2006.
313
00:31:57,120 --> 00:32:02,969
And this is one of the first deep learning nets that you don't know how deep learning net and rough terms.
314
00:32:02,970 --> 00:32:07,230
It's analogous to the connections of neurones in your brain.
315
00:32:07,470 --> 00:32:12,780
It's a very loose analogy, but that was the inspiration that started AI researchers off down this sort of
316
00:32:14,250 --> 00:32:17,970
what's called constructivist constructionist approach to artificial intelligence.
317
00:32:18,330 --> 00:32:22,830
So Hayes was the first one to figure out how to fly a helicopter from first principles.
318
00:32:23,460 --> 00:32:28,020
It encountered the universe of the helicopter and it flew and crashed the flew and crashed.
319
00:32:28,020 --> 00:32:32,280
It flew and crashed on a simulator. But he could afford the models to actually and even over again.
320
00:32:32,820 --> 00:32:39,330
And then eventually it reached that takeoff point and it mastered helicopter flight to the point at which it could roll a helicopter,
321
00:32:39,330 --> 00:32:43,860
which is no mean feat for any helicopter pilots, because normally they crash at that point.
322
00:32:44,790 --> 00:32:53,609
So not only could you grow the form to pretty complex manoeuvres as well, this one down here is in the last couple of weeks.
323
00:32:53,610 --> 00:33:01,169
Anybody seen this one? You know, this guy is he's an American Air Force colonel.
324
00:33:01,170 --> 00:33:06,450
He retired. He is a a superiority combat instructor.
325
00:33:06,450 --> 00:33:10,500
Who is his job? Dogfighting. Top gun stalwarts we've spent his career doing.
326
00:33:10,890 --> 00:33:13,620
Now he works as a consultant for air companies.
327
00:33:13,950 --> 00:33:22,770
And this is him, as they say, having his ass handed to him by an artificial intelligence system that's running off blackberry pie,
328
00:33:23,070 --> 00:33:31,980
running on the BlackBerry Pi mobile phone. And he was unable to beat it in repeated air to air combat activities.
329
00:33:32,340 --> 00:33:39,059
And that's only now in the last couple of weeks. And it's really combat against human vessels is something that obviously the Pentagon
330
00:33:39,060 --> 00:33:43,049
has engineers been working on since the 1980s with no particular degree of success.
331
00:33:43,050 --> 00:33:47,910
It's just a complicated control problem to be able to process all the relevant
332
00:33:47,910 --> 00:33:53,819
information in the required amount of time to to defeat a sophisticated human adversary.
333
00:33:53,820 --> 00:34:01,440
But it's now been done. Pentagon, by the way, still planning to put a man in the cockpit of the next generation long range bomber,
334
00:34:01,440 --> 00:34:04,989
which is something that they might think about again.
335
00:34:04,990 --> 00:34:09,149
And you know what this is? That's that's an intelligent design.
336
00:34:09,150 --> 00:34:13,030
He's a visionary design are where they brand the computers.
337
00:34:13,140 --> 00:34:21,150
How did you know that? And it was ever known that if you actually geek out yet that's exactly what it is
338
00:34:21,150 --> 00:34:25,889
it's very old now is a nicer antenna to go on a satellite communication antenna.
339
00:34:25,890 --> 00:34:29,400
And it was designed using A.I., which is why it looks like that wacky shape.
340
00:34:29,850 --> 00:34:33,540
It's not the sort of thing that I would sit down with it, design me an antenna I'd come up with,
341
00:34:33,540 --> 00:34:37,739
used a straight forward thing like that one bend and it came out looking like that.
342
00:34:37,740 --> 00:34:41,970
And that is the optimal solution that is designed better than it than a human could do it.
343
00:34:42,930 --> 00:34:51,690
And I think together these things show a range of the sort of military relevant activities that I is involved with already.
344
00:34:52,050 --> 00:34:56,360
This is this is weapons design right now.
345
00:34:56,380 --> 00:35:07,770
Already you can read this in stations for them on on the US Army website solicitations to use air to design bunker penetrating munitions.
346
00:35:08,100 --> 00:35:11,829
So air is already involved in designing weapon systems.
347
00:35:11,830 --> 00:35:20,670
As I'm walking over here with Rob and we were talking about the carriers, the Navy carriers was the favourite project of the Royal Navy.
348
00:35:20,940 --> 00:35:24,720
We might also have been talking about the US Air Force man and long range bomber.
349
00:35:25,320 --> 00:35:31,440
These are weapon systems that are designed partly with the efficiency imperative in mind.
350
00:35:31,440 --> 00:35:35,100
You know, war has an efficiency imperative. If you don't do it well, you go out of business.
351
00:35:35,490 --> 00:35:40,090
But partly also of course to reflect the institutional proclivities of the organisations that design them.
352
00:35:40,110 --> 00:35:43,620
Bureaucratic politics in action that's only sustainable.
353
00:35:45,390 --> 00:35:48,810
Well, have been designed by human and I probably wouldn't design you an aircraft carrier.
354
00:35:48,840 --> 00:35:55,200
You might not like what it designed you, but it would design you something a bit different from from what you would come up with.
355
00:35:55,860 --> 00:36:02,420
So alien weapons design, this is not just to be in strategic situations or operational situations,
356
00:36:02,550 --> 00:36:10,410
situations with less structure to them when it comes to decision making and this is off to a tactical level,
357
00:36:11,730 --> 00:36:16,740
activities across a range of military savings activities.
358
00:36:17,400 --> 00:36:27,000
And my argument is to my good friends who work for DeepMind, who show me on a regular basis that they are not involved in defence related research,
359
00:36:27,300 --> 00:36:30,930
and indeed it is one of their criteria for being bought by Google that they
360
00:36:30,930 --> 00:36:36,870
would not engage in defence related activities a little analogous to to nuclear
361
00:36:36,870 --> 00:36:45,570
weapons development and the idea of dual use technologies where you have civil nuclear capability capable of being switched over to military ones,
362
00:36:45,840 --> 00:36:52,020
but even more so in this case, because air is entirely about autonomous decision making in imperfectly understood environments,
363
00:36:52,020 --> 00:36:58,080
and that is an inherently military strategic capability, regardless of their desire to keep them separated.
364
00:36:58,500 --> 00:37:02,370
So let me let me close this section for reasons of time. Some questions.
365
00:37:02,670 --> 00:37:08,980
The way I've tackled air here in the book is through some dualism isms that I think a
366
00:37:09,090 --> 00:37:13,680
particularly relevant considering how I will affect strategic level decision making.
367
00:37:14,760 --> 00:37:17,130
First, the distinction between narrow and generally I.
368
00:37:17,340 --> 00:37:24,720
So when Stephen Hawking tells you that we've unleashed the demon we must unleash, the demon air is coming to take us over.
369
00:37:25,260 --> 00:37:33,510
He's talking about generally that is not just unbounded flexibility, capable of reaching decisions across a range of activities.
370
00:37:33,510 --> 00:37:40,499
It's not just a chess computer. It might also decide what steel garden, but it's its air that is capable,
371
00:37:40,500 --> 00:37:45,030
that is motivated to some extent intrinsically by itself, to do these things.
372
00:37:45,030 --> 00:37:49,200
That's what he's frightened about. It's a flexibly I is motivated to do it.
373
00:37:49,200 --> 00:37:54,840
So in fact, I'm not so much concerned about that for all sorts of reasons.
374
00:37:55,440 --> 00:37:56,610
But I think there's plenty of you because.
375
00:37:56,630 --> 00:38:02,460
Concerned about women still talking about narrow domains, specifically in the tactical impact of narrow domain specific.
376
00:38:02,700 --> 00:38:06,630
So that's one big distinction when we're talking about narrowly.
377
00:38:06,690 --> 00:38:12,929
I was talking about, Lisa, of being beaten there. I mean, to my generally, I have a look at the top, a robotics challenge from this year.
378
00:38:12,930 --> 00:38:19,320
The robots are still incapable of doing fairly basic stuff like opening doors, climbing over fences without toppling over.
379
00:38:20,070 --> 00:38:21,930
And I think, you know, the complexity,
380
00:38:21,930 --> 00:38:27,330
the combinatorial explosion in doing some of the things that we take for granted is still beyond what our current level of I can do.
381
00:38:27,840 --> 00:38:31,410
Second, you ask, is between tactical and strategic decision making.
382
00:38:31,830 --> 00:38:36,740
So flying that helicopter up upside down or defeating us, colonel,
383
00:38:36,760 --> 00:38:44,610
in that's where combat is a tactical level activity tactical level military activity still require some generalist knowledge,
384
00:38:44,610 --> 00:38:48,060
flexible skills across a range of activities. And they're still hard.
385
00:38:48,420 --> 00:38:53,340
It's not a chess computer. It's altogether more complex to learn to fly a helicopter.
386
00:38:57,990 --> 00:39:02,290
The last couple of of Judaism's before I. I stopped talking with you.
387
00:39:03,040 --> 00:39:08,950
One is this question of autonomy and self-directed A.I., and this opens a real can of worms.
388
00:39:09,190 --> 00:39:13,450
To what extent is artificial intelligence capable of making decisions for itself?
389
00:39:14,260 --> 00:39:20,740
And to what extent is it responding to inputs that you put in ahead of time ex-ante?
390
00:39:21,340 --> 00:39:28,240
And in an ideal world, you would like to have the idea of your strategic goal within your tactical goal,
391
00:39:28,480 --> 00:39:31,570
and you would tell the machine to achieve that, and it would go out and achieve it.
392
00:39:31,570 --> 00:39:36,250
And that's what the Army calls mission command. It does it today. You tell them what you want to achieve.
393
00:39:36,250 --> 00:39:40,389
You tell them your commander's intent, and you leave them to figure out how it is to achieve them.
394
00:39:40,390 --> 00:39:45,660
So in an ideal world, you would quite like that to be your relationship with autonomous performance.
395
00:39:45,670 --> 00:39:52,030
In terms of chief, they are autonomous to the extent that they are carrying out activities that you have told me to do upfront,
396
00:39:52,270 --> 00:39:53,470
and that's fine, well and good,
397
00:39:53,740 --> 00:40:00,460
except that you don't always know what it is that you want them to do upfront because you can't anticipate situations that may arise down the street.
398
00:40:02,350 --> 00:40:06,700
It's fine, well and good because you may be Satisficing is the term is a dishonest term
399
00:40:06,700 --> 00:40:11,409
is between different goals that you want to achieve that might be intention.
400
00:40:11,410 --> 00:40:14,830
Can you specify which goal you want machine to achieve?
401
00:40:14,980 --> 00:40:18,460
Which goal is the most important or otherwise?
402
00:40:20,750 --> 00:40:29,270
Boston's Beijing paid ticket counter is going to be read Boston superintelligence if you if you want an intro to the to the to the
403
00:40:29,270 --> 00:40:36,410
scary Elon Musk Stephen Hawking site of artificial intelligence Bostrom book on superintelligence is a pretty good place to start.
404
00:40:36,680 --> 00:40:41,270
And his nightmare scenario is the paperclip counter the paper accounts it has the
405
00:40:41,270 --> 00:40:45,410
best intentions in the world in fact has no intention is not is not out to get you.
406
00:40:45,630 --> 00:40:53,060
It's just been tasked with counting to 100. You want it to count 100 paperclips and tell you when it's done, that there are 100 paperclips there.
407
00:40:54,320 --> 00:41:00,680
And then they end. In Bostrom scenario, it conquers the known universe in order to count these paper because it can't count
408
00:41:00,680 --> 00:41:05,240
them to the degree of certainty that it is happy 100% that they're 100 paperclips.
409
00:41:05,390 --> 00:41:08,750
So in 98, 99, 100. Job done.
410
00:41:09,560 --> 00:41:13,310
Did I get it right? Just have one more check. 93, four, five, six.
411
00:41:13,520 --> 00:41:16,729
And it gets 90 to 100 and it's reasonably short. It has done it twice.
412
00:41:16,730 --> 00:41:23,090
Now you can have one more check because it's a vanishingly small probability that it counted wrong twice and it will carry on doing this.
413
00:41:23,450 --> 00:41:29,030
And as a genius artificial intelligence system that has been tasked with this thing and nothing else,
414
00:41:29,510 --> 00:41:34,730
it will secure all the resources, Bostrom argues. We don't know that it's obsidian.
415
00:41:34,790 --> 00:41:41,600
It will secure all the resources in the universe in order to keep counting the paperclips to make sure it's got 100 paperclips.
416
00:41:42,080 --> 00:41:45,239
That is an example of the agent principal problem gone disastrously wrong.
417
00:41:45,240 --> 00:41:49,910
All you wanted to know is is the roughly 100 paperclips that unless you specify it up front,
418
00:41:49,910 --> 00:41:56,660
your very clever artificial intelligence will keep counting and securing the resources to enable it to keep counting.
419
00:41:56,960 --> 00:41:59,300
So the principle problem is a key thing.
420
00:41:59,300 --> 00:42:06,350
How do you put in enough safeguards to ensure that the machine is going to do exactly what you want and not what it thinks you want?
421
00:42:07,340 --> 00:42:11,240
And if you can see the military dimension flashing autonomous weapons systems,
422
00:42:11,570 --> 00:42:16,940
I just asked you to kill this one guy, and now you're doing all sorts of other stuff that I didn't ask me to do.
423
00:42:19,550 --> 00:42:26,390
I talk in the book about consciousness and the possibility of machines becoming conscious, and I won't go into that here.
424
00:42:27,290 --> 00:42:30,860
The last the last few before I finish,
425
00:42:33,140 --> 00:42:41,959
what is the distinction between that I make between evolved human cognition and artificially intelligent machines?
426
00:42:41,960 --> 00:42:44,990
Doing strategy is probably a false dichotomy.
427
00:42:45,140 --> 00:42:48,980
It's not a dichotomy. It is not probably it is a false dichotomy.
428
00:42:50,030 --> 00:42:57,110
Artificial intelligence and human intelligence are probably not going to be seen as as separate entities for all that long.
429
00:42:57,350 --> 00:43:00,650
And HCI is human computer interface.
430
00:43:01,010 --> 00:43:05,600
And the extent to which you are using artificial intelligence systems as part of
431
00:43:05,600 --> 00:43:10,070
your normal cognition and pokémon's augmented reality is an example of that.
432
00:43:10,850 --> 00:43:16,850
Or the use of a chip in your brain that's reading your prosthetic arm.
433
00:43:17,390 --> 00:43:25,460
To tell you what your prosthetic arm is feeling as an example is where mind computer merge if you like another dollop of project.
434
00:43:25,760 --> 00:43:32,089
The quadriplegic woman at some at MIT has flown in F-35 through the power of her thoughts.
435
00:43:32,090 --> 00:43:41,180
She's flying distant, not real. I 35 about ten of these in existence but she's flown in a simulator using the power of her mind.
436
00:43:41,180 --> 00:43:49,700
So I think the breakdown between what is evolved human cognition and what is artificially intelligent cognition will probably blend a little bit.
437
00:43:50,000 --> 00:43:55,040
So that's another another duality. They talk about defence and civilian research.
438
00:43:55,160 --> 00:44:02,280
I'm pretty hard over on in the book. It's all defence research. It's all about decision making in complicated environments.
439
00:44:02,680 --> 00:44:06,250
And to me, this is real universes. A lot of A.I. recession.
440
00:44:06,510 --> 00:44:14,629
Longstanding A.I. researchers are pessimistic about A.I., and I think they are say, by inclination because they've been at it a long time.
441
00:44:14,630 --> 00:44:15,680
And it's a frustrating game.
442
00:44:16,340 --> 00:44:24,470
And they are by inclination because they want to counter unjustified hyperbole, the sort of, you know, massacres with is talking about.
443
00:44:24,830 --> 00:44:28,770
But but I think that that pessimism is to a certain extent, unwarranted.
444
00:44:28,790 --> 00:44:34,310
You know, toy universes can be pretty complicated and they are doing pretty well in these twin universes.
445
00:44:37,210 --> 00:44:45,280
Let me finish for for reasons of time, if nothing else with I then offer by way of conclusion the book Ways in which artificial
446
00:44:45,280 --> 00:44:49,180
intelligence is going to change strategy and perhaps take a lot more in questions,
447
00:44:49,900 --> 00:44:52,890
the sort of the sorts of things in regards to, oh,
448
00:44:52,900 --> 00:44:58,360
that A.I. changes the balance between offence and defence because it allows like a concentration of force.
449
00:44:58,720 --> 00:45:02,920
Hey, I shift your tolerance for risk. Firstly, because you're not losing any of your own people.
450
00:45:03,190 --> 00:45:11,169
Secondly, because your actions are not being driven by your own emotional cognition, A.I. makes the man in the loop unsustainable.
451
00:45:11,170 --> 00:45:14,650
Apartment issued defence position on artificial intelligence weapons.
452
00:45:14,710 --> 00:45:19,990
There will always be a man in the loop, a lawyer making a decision on what machines can and can't kill.
453
00:45:20,740 --> 00:45:29,950
That is inherently unsustainable. And it's fine when you're attacking the Taliban in Helmand province and you can lawyer everything back up to London.
454
00:45:30,280 --> 00:45:35,500
It's not fine if you're tackling an autonomous system operated by a high tech adversary in, say, China,
455
00:45:36,400 --> 00:45:42,280
because they might not care to have their lawyer in the loop deliberating or reporting their machine in court.
456
00:45:42,280 --> 00:45:47,350
To which case you eventually say the man in the loop will be forced out of the loop.
457
00:45:47,980 --> 00:45:52,900
By dint of the speed of artificial intelligence, A.I. will change radically.
458
00:45:52,900 --> 00:45:58,720
The structure of armed forces is another point that I make in the Strip.
459
00:45:58,740 --> 00:46:03,220
A lot of people out of the armed forces to kick off with no change, the sort of people that are employed.
460
00:46:04,120 --> 00:46:11,979
So that's my last slide. I started from the point of considering the way in which our evolved human
461
00:46:11,980 --> 00:46:17,080
psychology has shaped our strategic behaviour and for our entire human history.
462
00:46:17,590 --> 00:46:22,719
Up until fairly recently and really fairly recently,
463
00:46:22,720 --> 00:46:27,640
the Pentagon is now using artificial intelligence systems in rapid change to identify and prosecute targets.
464
00:46:28,240 --> 00:46:37,900
So fairly, fairly recently, our entire strategic decision making about the use of violence in the collective good
465
00:46:38,080 --> 00:46:42,880
has been taken on of these evolved cognitions that I say started with Manchester
466
00:46:42,900 --> 00:46:50,140
Law one and continues to cooperate that is now gradually being replaced with what and
467
00:46:51,010 --> 00:46:56,770
that I'll leave it and take any questions that you may have about the statements.
00:00:04,170 --> 00:00:08,450
In this episode, The Changing Character of War program welcomes Dr. Kenneth Payne,
2
00:00:08,830 --> 00:00:12,570
a senior lecturer in the School for Security Studies at King's College London.
3
00:00:13,230 --> 00:00:18,810
His talk explores the evolutionary basis of strategic behaviour from chimpanzees to computers,
4
00:00:19,140 --> 00:00:22,620
teasing out the connections between liberal warfare and human evolution.
5
00:00:24,770 --> 00:00:28,730
So the colour coding here is three parts of the book comes in three parts.
6
00:00:28,970 --> 00:00:34,120
The first part is all about evolution or fixes on evolution, and that's my thesis there.
7
00:00:34,130 --> 00:00:39,140
The second part is a bit about culture, which I'm going to gloss over in just three or four slides.
8
00:00:39,470 --> 00:00:43,340
And then the last part is probably the bit that some that might be of most interest to you is the bit
9
00:00:43,340 --> 00:00:47,120
about artificial intelligence and the ways in which artificial intelligence will shape strategy.
10
00:00:47,930 --> 00:00:57,440
And the overall thesis of the book in a nutshell is that there have been and everybody likes to be kind of reductive about revolutions in warfare.
11
00:00:58,370 --> 00:01:04,849
I think if the Toffler's, for example, that posited three revolutions in human history, the agricultural revolution,
12
00:01:04,850 --> 00:01:10,579
the information age, the story, the Industrial Revolution, and the Information Age Revolution I wouldn't bet it.
13
00:01:10,580 --> 00:01:17,450
The Toffler's beginning for two revolutions, a cognitive revolution early in human history that some, I argue,
14
00:01:17,450 --> 00:01:24,530
set up the basis for some underlying and enduring principles of strategy in the sense that most of us would understand strategy.
15
00:01:25,370 --> 00:01:30,200
And then an artificial intelligence revolution that challenges the evolved
16
00:01:30,830 --> 00:01:36,020
cognitive psychology that had been a longstanding part of our human history.
17
00:01:36,440 --> 00:01:45,920
So two revolutions, the evolution of a cognitive package to do with language and and theory of mind,
18
00:01:46,220 --> 00:01:51,470
and then latterly a revolution in artificial intelligence, which is just getting underway.
19
00:01:52,040 --> 00:02:01,040
So part one and the bits about evolved psychology, my thesis is that there is a dynamic relationship between strategy and evolution.
20
00:02:01,040 --> 00:02:06,079
That is to say that the need to think strategically was a strong factor in shaping our human
21
00:02:06,080 --> 00:02:11,870
evolution and also the way in which we evolved in pace with constraints on how we did strategy.
22
00:02:12,680 --> 00:02:20,180
Hopefully I wouldn't say too much about what strategy is, but in a nutshell, for those of you that might not be familiar with the term,
23
00:02:20,390 --> 00:02:27,080
I'm talking here primarily about military strategy, the use of organised social violence in pursuit of some political goal.
24
00:02:28,510 --> 00:02:36,250
My thesis is that some another Tom Jones who is not my thesis is not even Tom's, but he's done some tremendous work on it.
25
00:02:36,430 --> 00:02:44,650
The general thesis is that Greek surmise has an important effect on winning or losing in combat,
26
00:02:44,950 --> 00:02:56,230
and the need to cooperate together in progressively larger human groups has been one of the fundamental driving forces shaping human evolution.
27
00:03:00,160 --> 00:03:09,819
I got started on this project reading a book that I commend you all in the late seventies and early 80 seconds devours chimpanzee politics.
28
00:03:09,820 --> 00:03:18,550
If you haven't seen that. Grab a copy of that highly entertaining account of life and death in a chimpanzee colony colony in Arnhem,
29
00:03:19,240 --> 00:03:28,390
and the game's inverted commas that the chimpanzees played in order to get ahead and assume a dominant position within the hierarchy,
30
00:03:28,720 --> 00:03:32,080
colony and devil at the end of the book asks the question.
31
00:03:32,080 --> 00:03:36,550
Interesting question Do chimpanzees think strategically?
32
00:03:37,120 --> 00:03:42,790
And he answers that, yes, they do think strategically, but I think it is a slight oxymoron.
33
00:03:43,210 --> 00:03:48,340
They do so instinctively. There's not he doesn't necessarily suppose,
34
00:03:48,340 --> 00:03:54,129
a particularly sophisticated degree of self-reflection and reflection about the minds of
35
00:03:54,130 --> 00:03:59,560
others on the part of chimpanzees is much of their strategic behaviour is innate to them,
36
00:03:59,560 --> 00:04:00,970
is instinctively acquired.
37
00:04:01,720 --> 00:04:09,550
And I think not that I want to disagree on this eminent primatologists in the world, but I think it is something of a disservice to chimpanzees.
38
00:04:09,910 --> 00:04:16,720
And I think the view on chimpanzee cognition and theory of mind has moved on since the late seventies.
39
00:04:16,990 --> 00:04:24,220
So on the one hand, he's doing a disservice to chimpanzees because they are capable of thinking about the intentions of others,
40
00:04:24,280 --> 00:04:28,120
not as not to the same degree that we are, but they are to a certain degree.
41
00:04:28,660 --> 00:04:30,549
And on the other hand, he does limit service,
42
00:04:30,550 --> 00:04:38,940
because I think there's is the supposition that we as humans think in rational and calculating terms about strategy is unfair.
43
00:04:38,950 --> 00:04:47,860
We do a lot of unconscious, instinctive, horrific type thinking about strategy, just as the chimpanzees do.
44
00:04:48,610 --> 00:04:53,170
So Deval set me going and the question was, do animals fight wars?
45
00:04:53,380 --> 00:05:00,310
And if they do, do they fight wars in the strategic sense? Do they make use of violence in an instrumental and passive way?
46
00:05:00,700 --> 00:05:11,290
And my answer was this was answer fight pretty totalitarian wars of the entire destruction of other and colonies in other species,
47
00:05:11,650 --> 00:05:15,850
that they do so on a fairly automatic situational basis.
48
00:05:15,850 --> 00:05:21,670
They're responding in an evolved fashion to instinctive clues from the environment about how they ought to behave.
49
00:05:22,270 --> 00:05:26,290
Chimpanzees, on the other hand, certainly fight wars come on tonight in the second.
50
00:05:26,620 --> 00:05:34,270
And then he did so with a little bit more of what we might call strategic thought, a little bit more conscious reflection on what they're doing.
51
00:05:36,070 --> 00:05:41,170
And so that was my starting point. Way too much about the middle part of the book.
52
00:05:41,170 --> 00:05:45,570
Middle part of the book is a dilemma coming to anybody looking for general patterns in strategic behaviour.
53
00:05:45,580 --> 00:05:47,740
It's you've got the whole gamut of human history.
54
00:05:47,920 --> 00:05:54,820
What can you say about strategic behaviour that unites strategy across all these different varieties and cultural circumstances?
55
00:05:55,360 --> 00:06:02,200
That's book classes wrestle with, of course, is the tension between historicism something that's particular to the moment in theory,
56
00:06:02,200 --> 00:06:07,500
the ability to say something general across all particular areas or multiple areas.
57
00:06:08,080 --> 00:06:14,229
The second tension that the tension between historicism, the local, the particular and history with a capital H,
58
00:06:14,230 --> 00:06:19,120
the idea that history is driving forward towards some particular purpose.
59
00:06:19,840 --> 00:06:23,079
That that's the that's the tension I wrestle with in the middle part of the book,
60
00:06:23,080 --> 00:06:27,700
the bit about culture that I'm not going to spend too much time thinking thinking about here today.
61
00:06:28,090 --> 00:06:32,650
And then lastly, the last part of the book, can artificial intelligence strategy save that for the end?
62
00:06:34,270 --> 00:06:38,659
Here to kick it off, we come to chimpanzees and their ability to think strategically.
63
00:06:38,660 --> 00:06:45,250
This is a paper in 2014 in which the chimpanzees demonstrated an ability that might have
64
00:06:45,250 --> 00:06:51,400
appealed to social scientists from the middle of the 20th century to win a game of strategy.
65
00:06:51,640 --> 00:06:54,760
This is the game they're playing here is a little bit like pieces of rock.
66
00:06:55,000 --> 00:06:58,410
You have to guess what your adversary actually can't see.
67
00:06:58,420 --> 00:07:02,170
The chimpanzee is play. It's the chimpanzee here. Is that back to back?
68
00:07:02,170 --> 00:07:06,190
So they can't see what the adversary is choosing. One of them is tasked with matching the shapes.
69
00:07:06,190 --> 00:07:11,320
And and if he guesses, if he matches the shapes that the other chimp throws, he wins.
70
00:07:11,680 --> 00:07:14,299
The other is tasked with mismatching the shapes that the chimp race.
71
00:07:14,300 --> 00:07:19,540
So he knows he's got to think about what the other chimp is going to throw and then produce a mismatch.
72
00:07:19,870 --> 00:07:21,879
So you've got a strategic game to play,
73
00:07:21,880 --> 00:07:26,920
a strategic game in which they've got to try and guess what the other person going to do and behave accordingly.
74
00:07:27,610 --> 00:07:37,360
Chimpanzees outperform humans at this game, which if you were a scholar of the standard social science model of the rational activity,
75
00:07:37,360 --> 00:07:43,870
expected utility payoffs, that would be a puzzle. Why are the chimpanzees so good at it and why the humans are less good at it?
76
00:07:43,870 --> 00:07:49,899
And if you change the payoffs or if you change whether one's a match or a mismatch and they tell them very quickly,
77
00:07:49,900 --> 00:07:58,800
the chimpanzees change tack and start behaving according to how the change suggests that they should on a mathematical basis, they come.
78
00:07:58,860 --> 00:08:02,760
Very close to the optimal mathematics solution that humans are hopeless at it comparatively.
79
00:08:03,510 --> 00:08:11,610
The question is why and what does that mean? And the answer for me is not that chimpanzees are better strategists.
80
00:08:12,030 --> 00:08:20,280
It's that they're pursuing a different strategic logic. The chimpanzees are responding more instinctively to the payoffs.
81
00:08:20,550 --> 00:08:23,910
The humans are trying to second guess what the adversary is doing.
82
00:08:24,720 --> 00:08:27,960
Humans are making that strategy using theory of mind.
83
00:08:27,990 --> 00:08:34,450
They're trying to play the idea that, you know, the innate game and the chimpanzees aren't playing that game.
84
00:08:34,470 --> 00:08:40,830
They just want to sweep that they get for winning the game. And it's the over complication that we put into it,
85
00:08:40,830 --> 00:08:47,040
trying to read what the intention of others are the producers in this narrowly to play game a suboptimal outcome.
86
00:08:49,650 --> 00:08:54,060
So chimpanzees more rational actors than humans are in that game.
87
00:08:55,500 --> 00:09:00,329
Chimpanzees, we know because they're evolutionarily close to us.
88
00:09:00,330 --> 00:09:08,819
We parted company for them some 6 million years ago. Very useful comparator in terms of thinking about strategy.
89
00:09:08,820 --> 00:09:13,500
And in answer to Duval's question, you know, do chimpanzees fight wars?
90
00:09:13,740 --> 00:09:18,540
When he wrote that in 79, I think it was the goomba extreme war had just ended.
91
00:09:18,630 --> 00:09:21,720
I'm not sure whether where they could all have actually published about it,
92
00:09:21,720 --> 00:09:27,300
but the dominant view at the time in anthropology and actually in climatology as well was still,
93
00:09:27,420 --> 00:09:30,990
you know, chimpanzees were fairly peaceful, fun loving kind of characters.
94
00:09:31,740 --> 00:09:40,080
Since when that view has changed almost 180 degrees and the survey that shouldn't say we're going to say random, of course, Richard Wrangham,
95
00:09:40,950 --> 00:09:42,560
famous primatologist at Harvard,
96
00:09:43,410 --> 00:09:52,270
it turns out that some chimpanzee life is extremely violent at best interest group when you're struggling for dominance as it develops.
97
00:09:52,320 --> 00:09:56,790
Chimpanzees were on them and also between groups in terms of these chimpanzee wars.
98
00:09:57,330 --> 00:10:05,010
And the the paper by Wrangham that compiles all the available evidence around in 2006
99
00:10:05,730 --> 00:10:10,020
about how many wars were fought and what the levels of casualties were in these wars.
100
00:10:10,020 --> 00:10:16,860
And it turns out the chimpanzee life inside the group is much, much more violent than Detroit in the 1970s,
101
00:10:17,580 --> 00:10:24,120
and that some the levels of war were between chimpanzee groups allowing for small
102
00:10:24,120 --> 00:10:28,259
sample size was quite likely to be on a par with hunter gatherer communities.
103
00:10:28,260 --> 00:10:30,060
And we'll see what that was in the moment.
104
00:10:30,300 --> 00:10:38,400
Here's some chimpanzees behaving you might think strategically, perhaps in response to innate subconscious inclinations.
105
00:10:39,120 --> 00:10:45,779
They're conducting a chimpanzee patrol here and they patrol into the territory of neighbouring chimpanzees and they're doing the tactics
106
00:10:45,780 --> 00:10:54,420
when it comes to waging a war like the going to extreme war is the raid and ambush that typically attack the 10 to 1 force ratios.
107
00:10:55,410 --> 00:11:02,040
And so if they can't not heavily outnumber their adversary, they'll stage a silent retreat back away until such time as they do.
108
00:11:02,040 --> 00:11:07,860
WRANGHAM Theory is that this behaviour, which contrasts markedly with that of bonobos,
109
00:11:08,760 --> 00:11:11,430
they tend to fight wars, spend a lot of time having a good time.
110
00:11:13,320 --> 00:11:18,990
Range theory is that the distribution of food which forced chimpanzees to forage in an isolated fashion,
111
00:11:18,990 --> 00:11:22,980
may be particularly vulnerable to predation in this way by their fellow chimpanzees,
112
00:11:23,310 --> 00:11:31,170
because they had to fissure into small groups to go foraging for food, they could very quickly be outnumbered and defeated by other chimpanzees.
113
00:11:31,980 --> 00:11:36,660
The question is, is that strategic behaviour or not?
114
00:11:37,530 --> 00:11:44,759
On the one hand, the chimpanzees are capable of waging a sustained war against another group of chimpanzees.
115
00:11:44,760 --> 00:11:48,930
In that extreme war they annihilated the other group of chimpanzees.
116
00:11:48,930 --> 00:11:51,360
That survey every last chimpanzee.
117
00:11:51,400 --> 00:11:59,130
I think there were a couple of females left over that were absorbed back into the original population and they waited over a period of years,
118
00:12:01,800 --> 00:12:04,470
you might say, well, that's not necessarily strategic behaviour and you know,
119
00:12:04,500 --> 00:12:10,889
they're following an evolved inclination to conduct these patrols and to stage ambushes and I think you'd probably be right about that.
120
00:12:10,890 --> 00:12:19,110
For me, the big difference that makes our behaviour more strategic in the sense that we're familiar with is our capacity for this theory of mind,
121
00:12:19,110 --> 00:12:27,000
for intuiting what other people want and communicating on the basis of that in order to derive an effective strategy.
122
00:12:27,840 --> 00:12:31,049
So I start with chimpanzee wars and I say yes.
123
00:12:31,050 --> 00:12:35,010
On the one hand they can do instinctive strategy, but there's something different about humans.
124
00:12:35,400 --> 00:12:42,780
And what's different about humans is our ability to communicate and to plan for an imagined future.
125
00:12:43,950 --> 00:12:51,269
And the basic theory that elaborate in part one is that there is enough violence in
126
00:12:51,270 --> 00:12:54,870
the hunter gatherer world in which we evolved for almost all of our human history.
127
00:12:55,290 --> 00:12:58,820
There is a sufficiently high level of violence like in that chimpanzee world.
128
00:12:59,610 --> 00:13:04,170
To force us to become effective co-operators, we have to cooperate.
129
00:13:04,440 --> 00:13:08,160
Otherwise, we go out of business. That's essentially the theory.
130
00:13:08,880 --> 00:13:14,790
So, as you might know, Robin Dunbar's work here may have heard Dunbar's number.
131
00:13:15,390 --> 00:13:18,690
Dunbar's number? Yeah. And under the back? I hope not.
132
00:13:18,690 --> 00:13:25,559
Robin Dunbar. I've never met Robin Dunbar. So. So Dunbar's number is the size.
133
00:13:25,560 --> 00:13:33,660
Your neocortex correlates very effectively with the size of social group in which you live in the larger neocortex, the theory goes.
134
00:13:34,230 --> 00:13:38,490
The greater the ability. You have to track the key essentials of life in a social group.
135
00:13:38,790 --> 00:13:42,690
And these key centres are essentially who's up, who's down and who's shagging who.
136
00:13:42,900 --> 00:13:47,430
These are the things that you won't be able t t to understand.
137
00:13:48,150 --> 00:13:55,670
And if you if you correlate a chimpanzee neocortex sized chimpanzee cortex size relationship,
138
00:13:55,980 --> 00:13:59,370
you can extend that line outwards to humans and come up with what's called Dunbar's number.
139
00:13:59,400 --> 00:14:09,300
Dunbar's number for humans is 150. That's the size of hunter gatherer community in which we spent a lot of our early human evolution.
140
00:14:09,540 --> 00:14:15,180
It's a very rough rule of thumb because actually the site you hang out with on a day to day basis would be near 50 people.
141
00:14:15,600 --> 00:14:19,710
That would be your kinship band, people you primarily be related to.
142
00:14:20,310 --> 00:14:25,140
And you come together in this gathering of 150 for an overnight camp, perhaps for reasons of security.
143
00:14:26,250 --> 00:14:35,459
Dunbar's number is one aspect of the theory of human evolution called the social theory of cognitive evolution,
144
00:14:35,460 --> 00:14:43,170
which says a lot of our functions are derived to help us navigate this complex, increasingly co-operative,
145
00:14:43,380 --> 00:14:49,170
interdependent world 50 to 150 people with whom we would spend most of our time.
146
00:14:49,530 --> 00:14:52,860
That's the group that we hung out here now.
147
00:14:55,930 --> 00:15:03,520
The second big theory to think about is Frederick Manchester's mathematical laws of armed conflict dating to 1916.
148
00:15:03,520 --> 00:15:09,330
Anybody familiar with those yet to be familiar with? They say without without stretching the point.
149
00:15:09,410 --> 00:15:12,879
Point one, this is an abstraction. It's a mathematical abstraction.
150
00:15:12,880 --> 00:15:16,390
It's not meant to correlate precisely to real world behaviour.
151
00:15:17,080 --> 00:15:26,150
But Manchester's argument was that the benefits that would accrue from being in a larger group would be disproportionate to the size of that group.
152
00:15:26,770 --> 00:15:39,190
That's what he call square. If fighting is structured so that it's a melee, basically a wall against all, then the fighting power of the group,
153
00:15:39,190 --> 00:15:44,019
all of the things being equal will increase geometric t relative to the size of the group.
154
00:15:44,020 --> 00:15:50,469
The advantages from being in a larger group will increase geometrically if fighting is structured.
155
00:15:50,470 --> 00:15:53,650
So it's one on one. So imagine a series of duels at dawn.
156
00:15:54,490 --> 00:16:04,200
Or imagine, as I do later in the book, Hoplite Warfare, where the formations force you to fight 1 to 1 rather than in an overall melee.
157
00:16:04,930 --> 00:16:09,790
Then it still matters to be big, but it matters in a linear fashion.
158
00:16:10,720 --> 00:16:17,740
Say provided fighting is disorganised or there is a capacity to concentrate fire, not renown,
159
00:16:17,740 --> 00:16:23,830
strategic principle, concentrating force where you can then the fighting power increases geometrically.
160
00:16:24,550 --> 00:16:32,170
So the question is, is, is there enough fighting in hunter gatherer communities early in human evolution to
161
00:16:32,170 --> 00:16:38,680
force the agglomeration of humans together into larger groups and in larger groups?
162
00:16:38,680 --> 00:16:40,750
Perforce, they have to fight.
163
00:16:40,960 --> 00:16:48,640
You have to be in a group with people to whom you're not directly related, and you have to find a way of cooperating with strangers.
164
00:16:49,660 --> 00:16:54,580
And the answer I make, the claim I make in the book and I'm not alone in making plenty of other people make it as well,
165
00:16:54,580 --> 00:17:00,790
is that, yes, there was enough fighting in hunter gatherer times to force people to cooperate together.
166
00:17:02,140 --> 00:17:07,780
So the result I suggest is that there was sufficient violence early in human history to put
167
00:17:07,780 --> 00:17:13,660
pressure on humans to start cooperating with people with whom they might not know each other.
168
00:17:13,760 --> 00:17:17,230
Manchester's law provides a mathematical formulation of that.
169
00:17:17,590 --> 00:17:23,200
That is to say the prevalence of war solves what's known as the free riding problem.
170
00:17:23,650 --> 00:17:30,940
Free riding problem is essentially a problem of loafing. If you get enough people together, there'll always be some people who slack off.
171
00:17:31,270 --> 00:17:36,220
And if it's a matter of fighting somebody to the death, I personally would be on the side of the slacker office.
172
00:17:36,970 --> 00:17:40,600
So what is it that's going to make me cooperate?
173
00:17:40,600 --> 00:17:44,050
That's going to stop me from slacking off?
174
00:17:44,320 --> 00:17:48,100
One answer is to this the prevalence of war forces,
175
00:17:48,730 --> 00:17:55,420
cooperation and scrutiny of how people are behaving within a particular book say the problem to be solved,
176
00:17:56,050 --> 00:18:03,879
you'll be aware, is the classic problem of the prisoner's dilemma. And the payoffs is just in the prisoner's dilemma,
177
00:18:03,880 --> 00:18:09,730
such that non-cooperation tends to be the favoured option, even if cooperating would be better for you.
178
00:18:10,150 --> 00:18:14,020
The solution to that problem, as Samuel Bowles argues, as Peter Turchin argues,
179
00:18:14,290 --> 00:18:20,200
is that there is sufficient violence to force the read out that on co-operators and to force cooperation.
180
00:18:20,620 --> 00:18:23,799
And actually you can see echoes of that in a couple of days.
181
00:18:23,800 --> 00:18:30,730
Research is there. So the basic theory of cooperation in an evolutionary sense is from Robert TRIVERS
182
00:18:30,730 --> 00:18:35,680
Trivers argues that we cooperate with non kin on the basis of reciprocal altruism.
183
00:18:37,030 --> 00:18:43,750
That is, if I scratch your back, I do so with the supposition that in years to come you might scratch my back.
184
00:18:43,960 --> 00:18:48,580
So I will behave altruistically to you on the assumption that I'll get some payback later on.
185
00:18:48,580 --> 00:18:51,070
That's the basic theory of reciprocal altruism,
186
00:18:51,310 --> 00:18:57,730
and that requires you to have a make a pretty good bet about it that the probability of being paid back in that future date,
187
00:18:58,060 --> 00:19:05,140
of course, the incentive then is to cheat, to get away with having your back scratched without returning the favour, so to speak.
188
00:19:05,470 --> 00:19:13,750
And Dan Ariely and Travis himself have demonstrated that very ably trivers in the in the world of animals.
189
00:19:13,990 --> 00:19:24,310
Ariely, in some fascinating experiments on his students, proved that our inclination is to cheat and then lie about it.
190
00:19:25,030 --> 00:19:33,099
But what actually proved this is quite interesting is that we lie about it not just to the people we're cheating, we lie about it to ourselves.
191
00:19:33,100 --> 00:19:38,170
So we have the capacity to persuade ourselves that we are good, honest co-operators, even while we're cheating.
192
00:19:39,100 --> 00:19:43,360
So we can see that the natural inclination is not to be a free,
193
00:19:43,360 --> 00:19:47,470
cooperative member of society, even though the pressure of war suggests that we should be.
194
00:19:48,460 --> 00:19:53,950
But there is, because of that pressure of war, I argue, still some inclination to cooperate.
195
00:19:54,280 --> 00:19:55,300
And you can see that.
196
00:19:55,330 --> 00:20:05,170
When you do real world experiments with the prisoner's dilemma, in theory, in game theoretical terms, the incentive is to cheat and not to cooperate.
197
00:20:05,770 --> 00:20:12,370
In practice, if you give the prisoner's dilemma to normal people, they tend to cooperate, even where it's a one shot game.
198
00:20:12,370 --> 00:20:18,550
So there's no possibility of playing the same person again. So you can't build a reputation for trust that might shape your behaviour.
199
00:20:18,880 --> 00:20:21,100
Was a one shot game with a stranger.
200
00:20:21,310 --> 00:20:28,690
We have a tendency to wait to cooperate and the argument is that tendency to cooperate has evolved within us as a result,
201
00:20:28,690 --> 00:20:35,890
as I say, these pressure of warfare. And in fact, even if you give the prisoner's dilemma to genuine real life prisoners who ought to know better,
202
00:20:36,340 --> 00:20:39,880
they tend to cooperate more than is suggested by the prisoner's dilemma.
203
00:20:42,810 --> 00:20:51,180
So this evolutionary picture then that I take out of the first part combines this notion of reciprocal altruism.
204
00:20:52,200 --> 00:20:53,510
Lanchester His idea,
205
00:20:53,540 --> 00:21:05,040
the victory goes to the big battalions in a disproportionate sense and comes up with what I think is the the dominant approach to strategy,
206
00:21:05,040 --> 00:21:08,430
which is to be strong everywhere, if you're strong everywhere. Of course, strategy is very easy.
207
00:21:09,270 --> 00:21:15,750
And the strategy that the strong evolved with involves scale, mass and concentration of force.
208
00:21:16,230 --> 00:21:19,860
And and that couldn't I gave you Voltaire a moment ago.
209
00:21:20,130 --> 00:21:22,680
Victory doesn't belong to the big chains. It belongs to the best shooters.
210
00:21:22,800 --> 00:21:28,740
Lanchester would disagree with that, and I would suggest the evidence from evolutionary history is against that as well,
211
00:21:28,740 --> 00:21:34,350
which we generally get used to the big battalions, as Stone said of the pope.
212
00:21:34,590 --> 00:21:41,160
How many divisions? SIEGEL And say there is still that in-built supposition that mass pays off in the end.
213
00:21:42,300 --> 00:21:45,450
But you will say, you know, even if you're strong, you don't want to be a muffett.
214
00:21:45,450 --> 00:21:50,009
You also want to combine your strategy, the strong with being strong everywhere with some sensible tactics.
215
00:21:50,010 --> 00:21:51,180
And that's what the chimpanzees do.
216
00:21:51,180 --> 00:22:00,000
They make they in conducting a raid and an ambush rather than pitched battle against the same number of adversary chimpanzees.
217
00:22:00,360 --> 00:22:10,100
What they're doing effectively is building a force ratio that Lanchester would say is dominant to create a localised force ratios,
218
00:22:10,110 --> 00:22:15,960
even a strong use rate and an ambush. And if you look at this book wall before civilisation,
219
00:22:16,500 --> 00:22:22,470
that pattern is very strongly evident across ethnographic studies and archaeological studies
220
00:22:22,740 --> 00:22:27,240
that the dominant strategy in warfare is rate an ambush rather than pitched battle.
221
00:22:28,560 --> 00:22:32,100
And if you're on the on the weak side, that's when strategy gets really interesting,
222
00:22:32,100 --> 00:22:35,970
because you have to find a way of offsetting these advances to decrease the others.
223
00:22:36,300 --> 00:22:42,240
So you adopt all these sort of approaches and surprise, deception and fortification, manoeuvre.
224
00:22:42,900 --> 00:22:48,330
And if you are the borrowers, as the Americans call the British army, you come up with the British way of warfare,
225
00:22:48,330 --> 00:22:52,809
which is to try and be more flexible and you get throwing about it and reverse.
226
00:22:52,810 --> 00:23:02,890
Three. Max Fraser if you don't know, Max raises work and you should.
227
00:23:03,670 --> 00:23:09,010
He's a statistician at Oxford and he's produced extensive database on all sorts of fascinating stuff.
228
00:23:09,010 --> 00:23:15,159
The world, the world and figures and the general picture is that if he has a general motif is that the world is getting a better place,
229
00:23:15,160 --> 00:23:18,910
which is not always how it feels. A day's out for an American election,
230
00:23:19,870 --> 00:23:30,309
but he has compiled a whole range of statistical information about it's the most comprehensive list that I've seen of archaeological
231
00:23:30,310 --> 00:23:37,990
evidence for prevalence of warfare and hunter gatherer communities and ethnographic studies of warfare in hunter gatherer communities.
232
00:23:38,740 --> 00:23:47,080
And on the right here, you can't make out any of the detail here, but on the right this is violent deaths per 100,000 in state societies.
233
00:23:47,440 --> 00:23:52,000
Down at the bottom here is the world in 2007.
234
00:23:52,540 --> 00:23:56,110
So that's that's that's the the [INAUDIBLE] pit that we live in at the moment.
235
00:23:56,680 --> 00:24:03,910
Battle deaths 0.33 deaths per 100,000 people for the world in 2007.
236
00:24:04,750 --> 00:24:11,110
Whereas if you look at this hunter gatherer community study in the 1840s in California.
237
00:24:14,430 --> 00:24:21,569
1450 deaths per hundred thousand, saying hunter gatherer communities tend to have the most violent men,
238
00:24:21,570 --> 00:24:27,910
but they tend to have more violence, more violent deaths than do more modern state communities.
239
00:24:27,930 --> 00:24:34,320
So was there enough warfare in the course of human history to drive consumption behaviour and get rid of free riders?
240
00:24:34,980 --> 00:24:40,110
The answer for me is yes and that's the claim I take out of the first part of the book.
241
00:24:42,270 --> 00:24:48,120
I didn't spend a bit of time thinking through the implications of what that means for human cognition and
242
00:24:48,120 --> 00:24:53,640
for the heretics that we've developed to help us think strategically when we go into any of that here.
243
00:24:54,480 --> 00:24:59,010
That's part one of the book and an evolutionary history of warfare. It pays to be big.
244
00:24:59,010 --> 00:25:06,780
There's enough fights around to force us to agglomerate in large groups and to force it to learn patterns of cooperative
245
00:25:06,780 --> 00:25:14,580
behaviour and shortcuts for thinking about who we trust and who we shouldn't trust to allow us to think strategically.
246
00:25:15,180 --> 00:25:22,120
In part to the book. I take that forward and see if it holds in the cultural era.
247
00:25:22,560 --> 00:25:26,640
And the answer is that it kind of does.
248
00:25:27,160 --> 00:25:30,330
But stuff complicates it for reasons of time. I'm not going to get into.
249
00:25:30,330 --> 00:25:33,900
I use three fairly well-trodden examples.
250
00:25:34,200 --> 00:25:36,509
It's all about the Greeks and hoplites.
251
00:25:36,510 --> 00:25:44,910
And I talk about the invention of writing and particularly the invention of strategic studies as a separate way of thinking reflectively about war.
252
00:25:45,210 --> 00:25:51,630
And the take out from that study. The Greeks is that writing provided a first example of artificial intelligence,
253
00:25:51,630 --> 00:25:56,250
if you like, a means of offshoring, part of our cognition outside of our brain.
254
00:25:56,610 --> 00:26:01,870
I won't say any more about that example to Clausewitz and Napoleonic warfare.
255
00:26:01,890 --> 00:26:06,360
Again, I'm not going to say too much more about that. I will say a little bit about nuclear weapons.
256
00:26:07,980 --> 00:26:17,010
And the take out from the middle part of the book is that Manchester's law breaks down in the cultural era because
257
00:26:17,310 --> 00:26:23,280
fighting power is not just about the number of men that you have at your disposal and their ability to concentrate force.
258
00:26:23,880 --> 00:26:31,410
It's also about technology in the cultural areas technologies of warfare, weapons technologies, but not just weapons technologies.
259
00:26:31,800 --> 00:26:41,430
Technology screws Manchester's abstract thinking about the importance of scale when it comes to strategy.
260
00:26:41,670 --> 00:26:50,729
And the apotheosis of that, of course, is nuclear weapons, where the analogy can inflict as much damage as a thousand plane fighter bombing raid.
261
00:26:50,730 --> 00:26:55,320
On take, you say scale, which you have nuclear weapons becomes less of an issue.
262
00:26:55,830 --> 00:26:59,190
Of course, you need scale in order to develop nuclear weapons in the first place.
263
00:27:01,340 --> 00:27:07,100
I said at the beginning, I posited two revolutions in military affairs, the one the cognitive revolution,
264
00:27:07,140 --> 00:27:12,110
the flexible package to construct grips flexibly in order to deal with violence in the world.
265
00:27:12,590 --> 00:27:25,130
And second, artificial intelligence. And one of the other contenders for a revolution in military affairs that changed everything is nuclear weapons.
266
00:27:25,190 --> 00:27:30,910
Robert Jervis, many of you will know, said that nuclear weapons were a revolutionary development in warfare.
267
00:27:30,930 --> 00:27:33,290
They were the revolutionary development in warfare.
268
00:27:33,650 --> 00:27:40,220
Bernard Brody, the American strategic thinker on hearing about the destination of the first nuclear bomb,
269
00:27:40,640 --> 00:27:45,920
said that he thought it made everything else he had ever written redundant and obsolete.
270
00:27:46,220 --> 00:27:51,140
He thought that the entirety of strategic history was redundant and obsolete because nuclear weapons are transforming so much.
271
00:27:52,400 --> 00:27:59,750
I think he was wrong. He was wrong. But I also think that this is wrong in the sense that nuclear weapons are revolutionary.
272
00:28:00,860 --> 00:28:03,170
They're an important development in the history of warfare, for sure.
273
00:28:03,410 --> 00:28:07,800
But I didn't think the revolution and the reason is at the bottom of their strategy.
274
00:28:07,820 --> 00:28:13,280
Even with nuclear weapons remains a matter of our evolved human psychology.
275
00:28:13,610 --> 00:28:17,030
We still making decisions about the employment of nuclear weapons,
276
00:28:17,390 --> 00:28:25,430
using precisely the same cognitive architecture that we evolved with 4050, 200,000 years of our human evolution.
277
00:28:25,460 --> 00:28:34,160
So all these factors in nuclear strategic theory, escalation, the need to demonstrate credibility, miscalculation,
278
00:28:34,170 --> 00:28:41,480
the scope for misperception, they are all subject to the same evolved cognitive, psychological foibles that we brought with us before.
279
00:28:42,470 --> 00:28:48,860
Jervis said that nuclear weapons were revolutionary on the basis that they would transform the balance between denial and punishment.
280
00:28:49,160 --> 00:28:57,280
And of course, you have a nuclear standoff through mutually assured destruction, and the threat of retaliation is how you use the weapon.
281
00:28:57,290 --> 00:29:03,350
So is the threat of using the weapon rather than the actual use of the weapon that delivers you more strategic result.
282
00:29:03,740 --> 00:29:08,330
And what he said is there's no point trying to deny nuclear weapons.
283
00:29:08,330 --> 00:29:15,590
You can't defend against them. And instead, your strategic use of nuclear weapons relies entirely on the threat of punishment.
284
00:29:17,760 --> 00:29:25,770
I'm not entirely sure that is revolutionary. Firstly, because denial and punishment both featured before the onset of nuclear weapons.
285
00:29:26,070 --> 00:29:30,810
You could certainly deny through the construction of fortification or the use of terrain, use of things.
286
00:29:31,140 --> 00:29:35,130
And you could also threaten punishments through retaliatory attack.
287
00:29:35,160 --> 00:29:40,800
You didn't need nuclear weapons. Individual nuclear weapons gave you an assured ability to punish.
288
00:29:41,520 --> 00:29:45,480
But punishment was certainly a feature of strategy even before the development of nuclear weapons.
289
00:29:46,320 --> 00:29:51,420
So I'm not sure in Genesis account that nuclear weapons are particularly revolutionary.
290
00:29:51,890 --> 00:29:55,980
That's what I'll say about them here. Reasons of time.
291
00:29:56,790 --> 00:29:59,669
I wanted to get ahead to the artificial intelligence because I'm aware that that's
292
00:29:59,670 --> 00:30:02,910
probably why most of you here not to hear me rambling on about chimpanzees warfare.
293
00:30:03,580 --> 00:30:12,810
So in very quick order, my headline for artificial intelligence is that it is a revolution in military affairs,
294
00:30:12,990 --> 00:30:19,560
is a revolution in strategic affairs precisely because it challenges the underlying evolved psychological
295
00:30:20,820 --> 00:30:26,670
abilities that we've brought to bear to tackle strategic problems since we emerged on the planet and even before,
296
00:30:26,820 --> 00:30:34,290
you know, antecedent species. Here are some examples of artificial intelligence from recent months.
297
00:30:34,860 --> 00:30:39,570
Let's go with the easy ones. Do you believe what's going on there? Yeah.
298
00:30:39,990 --> 00:30:43,470
Google's AlphaGo. Yeah. The part of the world.
299
00:30:43,620 --> 00:30:47,399
The world leading go. Yeah. Did you release it all for one? The World Championship.
300
00:30:47,400 --> 00:30:52,889
And before that? Very quickly. We very quickly adopt hindsight bias with achievements like that.
301
00:30:52,890 --> 00:30:59,580
And assuming, of course, it beat him before that pretty well. Every A.I. expert at DeepMind was saying it couldn't be done,
302
00:30:59,580 --> 00:31:06,930
couldn't be done for 20 years because they'd all been trying to do it goes right for those you don't know who wins or version where the ball game.
303
00:31:07,200 --> 00:31:11,160
It's a very simple constrained what now developer would call a toy universe.
304
00:31:11,370 --> 00:31:14,160
It's not like the complex, messy reality of winning in Vietnam.
305
00:31:14,490 --> 00:31:19,350
There are only a certain number of moves that can take place at two different varieties moves that can take place on the keyboard,
306
00:31:19,650 --> 00:31:27,690
but they can do so with such dazzling combinatorial complexity that there are more possible moves mathematically than there are atoms in the universe.
307
00:31:28,560 --> 00:31:32,010
So you can't solve a game of going through brute force calculation,
308
00:31:32,340 --> 00:31:36,270
which is why A.I. developers have said it would be 20 odd years before anybody came
309
00:31:36,270 --> 00:31:40,830
close to performing at the level of the world champion still is beating him anyway.
310
00:31:41,430 --> 00:31:44,520
Within DeepMind, they were pretty confident they trash him and they truly did.
311
00:31:45,600 --> 00:31:51,450
So that's not one this one you. But you know, that one is a wreckage or helicopter flying upside down.
312
00:31:51,450 --> 00:31:56,730
But more significantly, the Knights and rings, remote control helicopter flying upside down from 2006.
313
00:31:57,120 --> 00:32:02,969
And this is one of the first deep learning nets that you don't know how deep learning net and rough terms.
314
00:32:02,970 --> 00:32:07,230
It's analogous to the connections of neurones in your brain.
315
00:32:07,470 --> 00:32:12,780
It's a very loose analogy, but that was the inspiration that started AI researchers off down this sort of
316
00:32:14,250 --> 00:32:17,970
what's called constructivist constructionist approach to artificial intelligence.
317
00:32:18,330 --> 00:32:22,830
So Hayes was the first one to figure out how to fly a helicopter from first principles.
318
00:32:23,460 --> 00:32:28,020
It encountered the universe of the helicopter and it flew and crashed the flew and crashed.
319
00:32:28,020 --> 00:32:32,280
It flew and crashed on a simulator. But he could afford the models to actually and even over again.
320
00:32:32,820 --> 00:32:39,330
And then eventually it reached that takeoff point and it mastered helicopter flight to the point at which it could roll a helicopter,
321
00:32:39,330 --> 00:32:43,860
which is no mean feat for any helicopter pilots, because normally they crash at that point.
322
00:32:44,790 --> 00:32:53,609
So not only could you grow the form to pretty complex manoeuvres as well, this one down here is in the last couple of weeks.
323
00:32:53,610 --> 00:33:01,169
Anybody seen this one? You know, this guy is he's an American Air Force colonel.
324
00:33:01,170 --> 00:33:06,450
He retired. He is a a superiority combat instructor.
325
00:33:06,450 --> 00:33:10,500
Who is his job? Dogfighting. Top gun stalwarts we've spent his career doing.
326
00:33:10,890 --> 00:33:13,620
Now he works as a consultant for air companies.
327
00:33:13,950 --> 00:33:22,770
And this is him, as they say, having his ass handed to him by an artificial intelligence system that's running off blackberry pie,
328
00:33:23,070 --> 00:33:31,980
running on the BlackBerry Pi mobile phone. And he was unable to beat it in repeated air to air combat activities.
329
00:33:32,340 --> 00:33:39,059
And that's only now in the last couple of weeks. And it's really combat against human vessels is something that obviously the Pentagon
330
00:33:39,060 --> 00:33:43,049
has engineers been working on since the 1980s with no particular degree of success.
331
00:33:43,050 --> 00:33:47,910
It's just a complicated control problem to be able to process all the relevant
332
00:33:47,910 --> 00:33:53,819
information in the required amount of time to to defeat a sophisticated human adversary.
333
00:33:53,820 --> 00:34:01,440
But it's now been done. Pentagon, by the way, still planning to put a man in the cockpit of the next generation long range bomber,
334
00:34:01,440 --> 00:34:04,989
which is something that they might think about again.
335
00:34:04,990 --> 00:34:09,149
And you know what this is? That's that's an intelligent design.
336
00:34:09,150 --> 00:34:13,030
He's a visionary design are where they brand the computers.
337
00:34:13,140 --> 00:34:21,150
How did you know that? And it was ever known that if you actually geek out yet that's exactly what it is
338
00:34:21,150 --> 00:34:25,889
it's very old now is a nicer antenna to go on a satellite communication antenna.
339
00:34:25,890 --> 00:34:29,400
And it was designed using A.I., which is why it looks like that wacky shape.
340
00:34:29,850 --> 00:34:33,540
It's not the sort of thing that I would sit down with it, design me an antenna I'd come up with,
341
00:34:33,540 --> 00:34:37,739
used a straight forward thing like that one bend and it came out looking like that.
342
00:34:37,740 --> 00:34:41,970
And that is the optimal solution that is designed better than it than a human could do it.
343
00:34:42,930 --> 00:34:51,690
And I think together these things show a range of the sort of military relevant activities that I is involved with already.
344
00:34:52,050 --> 00:34:56,360
This is this is weapons design right now.
345
00:34:56,380 --> 00:35:07,770
Already you can read this in stations for them on on the US Army website solicitations to use air to design bunker penetrating munitions.
346
00:35:08,100 --> 00:35:11,829
So air is already involved in designing weapon systems.
347
00:35:11,830 --> 00:35:20,670
As I'm walking over here with Rob and we were talking about the carriers, the Navy carriers was the favourite project of the Royal Navy.
348
00:35:20,940 --> 00:35:24,720
We might also have been talking about the US Air Force man and long range bomber.
349
00:35:25,320 --> 00:35:31,440
These are weapon systems that are designed partly with the efficiency imperative in mind.
350
00:35:31,440 --> 00:35:35,100
You know, war has an efficiency imperative. If you don't do it well, you go out of business.
351
00:35:35,490 --> 00:35:40,090
But partly also of course to reflect the institutional proclivities of the organisations that design them.
352
00:35:40,110 --> 00:35:43,620
Bureaucratic politics in action that's only sustainable.
353
00:35:45,390 --> 00:35:48,810
Well, have been designed by human and I probably wouldn't design you an aircraft carrier.
354
00:35:48,840 --> 00:35:55,200
You might not like what it designed you, but it would design you something a bit different from from what you would come up with.
355
00:35:55,860 --> 00:36:02,420
So alien weapons design, this is not just to be in strategic situations or operational situations,
356
00:36:02,550 --> 00:36:10,410
situations with less structure to them when it comes to decision making and this is off to a tactical level,
357
00:36:11,730 --> 00:36:16,740
activities across a range of military savings activities.
358
00:36:17,400 --> 00:36:27,000
And my argument is to my good friends who work for DeepMind, who show me on a regular basis that they are not involved in defence related research,
359
00:36:27,300 --> 00:36:30,930
and indeed it is one of their criteria for being bought by Google that they
360
00:36:30,930 --> 00:36:36,870
would not engage in defence related activities a little analogous to to nuclear
361
00:36:36,870 --> 00:36:45,570
weapons development and the idea of dual use technologies where you have civil nuclear capability capable of being switched over to military ones,
362
00:36:45,840 --> 00:36:52,020
but even more so in this case, because air is entirely about autonomous decision making in imperfectly understood environments,
363
00:36:52,020 --> 00:36:58,080
and that is an inherently military strategic capability, regardless of their desire to keep them separated.
364
00:36:58,500 --> 00:37:02,370
So let me let me close this section for reasons of time. Some questions.
365
00:37:02,670 --> 00:37:08,980
The way I've tackled air here in the book is through some dualism isms that I think a
366
00:37:09,090 --> 00:37:13,680
particularly relevant considering how I will affect strategic level decision making.
367
00:37:14,760 --> 00:37:17,130
First, the distinction between narrow and generally I.
368
00:37:17,340 --> 00:37:24,720
So when Stephen Hawking tells you that we've unleashed the demon we must unleash, the demon air is coming to take us over.
369
00:37:25,260 --> 00:37:33,510
He's talking about generally that is not just unbounded flexibility, capable of reaching decisions across a range of activities.
370
00:37:33,510 --> 00:37:40,499
It's not just a chess computer. It might also decide what steel garden, but it's its air that is capable,
371
00:37:40,500 --> 00:37:45,030
that is motivated to some extent intrinsically by itself, to do these things.
372
00:37:45,030 --> 00:37:49,200
That's what he's frightened about. It's a flexibly I is motivated to do it.
373
00:37:49,200 --> 00:37:54,840
So in fact, I'm not so much concerned about that for all sorts of reasons.
374
00:37:55,440 --> 00:37:56,610
But I think there's plenty of you because.
375
00:37:56,630 --> 00:38:02,460
Concerned about women still talking about narrow domains, specifically in the tactical impact of narrow domain specific.
376
00:38:02,700 --> 00:38:06,630
So that's one big distinction when we're talking about narrowly.
377
00:38:06,690 --> 00:38:12,929
I was talking about, Lisa, of being beaten there. I mean, to my generally, I have a look at the top, a robotics challenge from this year.
378
00:38:12,930 --> 00:38:19,320
The robots are still incapable of doing fairly basic stuff like opening doors, climbing over fences without toppling over.
379
00:38:20,070 --> 00:38:21,930
And I think, you know, the complexity,
380
00:38:21,930 --> 00:38:27,330
the combinatorial explosion in doing some of the things that we take for granted is still beyond what our current level of I can do.
381
00:38:27,840 --> 00:38:31,410
Second, you ask, is between tactical and strategic decision making.
382
00:38:31,830 --> 00:38:36,740
So flying that helicopter up upside down or defeating us, colonel,
383
00:38:36,760 --> 00:38:44,610
in that's where combat is a tactical level activity tactical level military activity still require some generalist knowledge,
384
00:38:44,610 --> 00:38:48,060
flexible skills across a range of activities. And they're still hard.
385
00:38:48,420 --> 00:38:53,340
It's not a chess computer. It's altogether more complex to learn to fly a helicopter.
386
00:38:57,990 --> 00:39:02,290
The last couple of of Judaism's before I. I stopped talking with you.
387
00:39:03,040 --> 00:39:08,950
One is this question of autonomy and self-directed A.I., and this opens a real can of worms.
388
00:39:09,190 --> 00:39:13,450
To what extent is artificial intelligence capable of making decisions for itself?
389
00:39:14,260 --> 00:39:20,740
And to what extent is it responding to inputs that you put in ahead of time ex-ante?
390
00:39:21,340 --> 00:39:28,240
And in an ideal world, you would like to have the idea of your strategic goal within your tactical goal,
391
00:39:28,480 --> 00:39:31,570
and you would tell the machine to achieve that, and it would go out and achieve it.
392
00:39:31,570 --> 00:39:36,250
And that's what the Army calls mission command. It does it today. You tell them what you want to achieve.
393
00:39:36,250 --> 00:39:40,389
You tell them your commander's intent, and you leave them to figure out how it is to achieve them.
394
00:39:40,390 --> 00:39:45,660
So in an ideal world, you would quite like that to be your relationship with autonomous performance.
395
00:39:45,670 --> 00:39:52,030
In terms of chief, they are autonomous to the extent that they are carrying out activities that you have told me to do upfront,
396
00:39:52,270 --> 00:39:53,470
and that's fine, well and good,
397
00:39:53,740 --> 00:40:00,460
except that you don't always know what it is that you want them to do upfront because you can't anticipate situations that may arise down the street.
398
00:40:02,350 --> 00:40:06,700
It's fine, well and good because you may be Satisficing is the term is a dishonest term
399
00:40:06,700 --> 00:40:11,409
is between different goals that you want to achieve that might be intention.
400
00:40:11,410 --> 00:40:14,830
Can you specify which goal you want machine to achieve?
401
00:40:14,980 --> 00:40:18,460
Which goal is the most important or otherwise?
402
00:40:20,750 --> 00:40:29,270
Boston's Beijing paid ticket counter is going to be read Boston superintelligence if you if you want an intro to the to the to the
403
00:40:29,270 --> 00:40:36,410
scary Elon Musk Stephen Hawking site of artificial intelligence Bostrom book on superintelligence is a pretty good place to start.
404
00:40:36,680 --> 00:40:41,270
And his nightmare scenario is the paperclip counter the paper accounts it has the
405
00:40:41,270 --> 00:40:45,410
best intentions in the world in fact has no intention is not is not out to get you.
406
00:40:45,630 --> 00:40:53,060
It's just been tasked with counting to 100. You want it to count 100 paperclips and tell you when it's done, that there are 100 paperclips there.
407
00:40:54,320 --> 00:41:00,680
And then they end. In Bostrom scenario, it conquers the known universe in order to count these paper because it can't count
408
00:41:00,680 --> 00:41:05,240
them to the degree of certainty that it is happy 100% that they're 100 paperclips.
409
00:41:05,390 --> 00:41:08,750
So in 98, 99, 100. Job done.
410
00:41:09,560 --> 00:41:13,310
Did I get it right? Just have one more check. 93, four, five, six.
411
00:41:13,520 --> 00:41:16,729
And it gets 90 to 100 and it's reasonably short. It has done it twice.
412
00:41:16,730 --> 00:41:23,090
Now you can have one more check because it's a vanishingly small probability that it counted wrong twice and it will carry on doing this.
413
00:41:23,450 --> 00:41:29,030
And as a genius artificial intelligence system that has been tasked with this thing and nothing else,
414
00:41:29,510 --> 00:41:34,730
it will secure all the resources, Bostrom argues. We don't know that it's obsidian.
415
00:41:34,790 --> 00:41:41,600
It will secure all the resources in the universe in order to keep counting the paperclips to make sure it's got 100 paperclips.
416
00:41:42,080 --> 00:41:45,239
That is an example of the agent principal problem gone disastrously wrong.
417
00:41:45,240 --> 00:41:49,910
All you wanted to know is is the roughly 100 paperclips that unless you specify it up front,
418
00:41:49,910 --> 00:41:56,660
your very clever artificial intelligence will keep counting and securing the resources to enable it to keep counting.
419
00:41:56,960 --> 00:41:59,300
So the principle problem is a key thing.
420
00:41:59,300 --> 00:42:06,350
How do you put in enough safeguards to ensure that the machine is going to do exactly what you want and not what it thinks you want?
421
00:42:07,340 --> 00:42:11,240
And if you can see the military dimension flashing autonomous weapons systems,
422
00:42:11,570 --> 00:42:16,940
I just asked you to kill this one guy, and now you're doing all sorts of other stuff that I didn't ask me to do.
423
00:42:19,550 --> 00:42:26,390
I talk in the book about consciousness and the possibility of machines becoming conscious, and I won't go into that here.
424
00:42:27,290 --> 00:42:30,860
The last the last few before I finish,
425
00:42:33,140 --> 00:42:41,959
what is the distinction between that I make between evolved human cognition and artificially intelligent machines?
426
00:42:41,960 --> 00:42:44,990
Doing strategy is probably a false dichotomy.
427
00:42:45,140 --> 00:42:48,980
It's not a dichotomy. It is not probably it is a false dichotomy.
428
00:42:50,030 --> 00:42:57,110
Artificial intelligence and human intelligence are probably not going to be seen as as separate entities for all that long.
429
00:42:57,350 --> 00:43:00,650
And HCI is human computer interface.
430
00:43:01,010 --> 00:43:05,600
And the extent to which you are using artificial intelligence systems as part of
431
00:43:05,600 --> 00:43:10,070
your normal cognition and pokémon's augmented reality is an example of that.
432
00:43:10,850 --> 00:43:16,850
Or the use of a chip in your brain that's reading your prosthetic arm.
433
00:43:17,390 --> 00:43:25,460
To tell you what your prosthetic arm is feeling as an example is where mind computer merge if you like another dollop of project.
434
00:43:25,760 --> 00:43:32,089
The quadriplegic woman at some at MIT has flown in F-35 through the power of her thoughts.
435
00:43:32,090 --> 00:43:41,180
She's flying distant, not real. I 35 about ten of these in existence but she's flown in a simulator using the power of her mind.
436
00:43:41,180 --> 00:43:49,700
So I think the breakdown between what is evolved human cognition and what is artificially intelligent cognition will probably blend a little bit.
437
00:43:50,000 --> 00:43:55,040
So that's another another duality. They talk about defence and civilian research.
438
00:43:55,160 --> 00:44:02,280
I'm pretty hard over on in the book. It's all defence research. It's all about decision making in complicated environments.
439
00:44:02,680 --> 00:44:06,250
And to me, this is real universes. A lot of A.I. recession.
440
00:44:06,510 --> 00:44:14,629
Longstanding A.I. researchers are pessimistic about A.I., and I think they are say, by inclination because they've been at it a long time.
441
00:44:14,630 --> 00:44:15,680
And it's a frustrating game.
442
00:44:16,340 --> 00:44:24,470
And they are by inclination because they want to counter unjustified hyperbole, the sort of, you know, massacres with is talking about.
443
00:44:24,830 --> 00:44:28,770
But but I think that that pessimism is to a certain extent, unwarranted.
444
00:44:28,790 --> 00:44:34,310
You know, toy universes can be pretty complicated and they are doing pretty well in these twin universes.
445
00:44:37,210 --> 00:44:45,280
Let me finish for for reasons of time, if nothing else with I then offer by way of conclusion the book Ways in which artificial
446
00:44:45,280 --> 00:44:49,180
intelligence is going to change strategy and perhaps take a lot more in questions,
447
00:44:49,900 --> 00:44:52,890
the sort of the sorts of things in regards to, oh,
448
00:44:52,900 --> 00:44:58,360
that A.I. changes the balance between offence and defence because it allows like a concentration of force.
449
00:44:58,720 --> 00:45:02,920
Hey, I shift your tolerance for risk. Firstly, because you're not losing any of your own people.
450
00:45:03,190 --> 00:45:11,169
Secondly, because your actions are not being driven by your own emotional cognition, A.I. makes the man in the loop unsustainable.
451
00:45:11,170 --> 00:45:14,650
Apartment issued defence position on artificial intelligence weapons.
452
00:45:14,710 --> 00:45:19,990
There will always be a man in the loop, a lawyer making a decision on what machines can and can't kill.
453
00:45:20,740 --> 00:45:29,950
That is inherently unsustainable. And it's fine when you're attacking the Taliban in Helmand province and you can lawyer everything back up to London.
454
00:45:30,280 --> 00:45:35,500
It's not fine if you're tackling an autonomous system operated by a high tech adversary in, say, China,
455
00:45:36,400 --> 00:45:42,280
because they might not care to have their lawyer in the loop deliberating or reporting their machine in court.
456
00:45:42,280 --> 00:45:47,350
To which case you eventually say the man in the loop will be forced out of the loop.
457
00:45:47,980 --> 00:45:52,900
By dint of the speed of artificial intelligence, A.I. will change radically.
458
00:45:52,900 --> 00:45:58,720
The structure of armed forces is another point that I make in the Strip.
459
00:45:58,740 --> 00:46:03,220
A lot of people out of the armed forces to kick off with no change, the sort of people that are employed.
460
00:46:04,120 --> 00:46:11,979
So that's my last slide. I started from the point of considering the way in which our evolved human
461
00:46:11,980 --> 00:46:17,080
psychology has shaped our strategic behaviour and for our entire human history.
462
00:46:17,590 --> 00:46:22,719
Up until fairly recently and really fairly recently,
463
00:46:22,720 --> 00:46:27,640
the Pentagon is now using artificial intelligence systems in rapid change to identify and prosecute targets.
464
00:46:28,240 --> 00:46:37,900
So fairly, fairly recently, our entire strategic decision making about the use of violence in the collective good
465
00:46:38,080 --> 00:46:42,880
has been taken on of these evolved cognitions that I say started with Manchester
466
00:46:42,900 --> 00:46:50,140
Law one and continues to cooperate that is now gradually being replaced with what and
467
00:46:51,010 --> 00:46:56,770
that I'll leave it and take any questions that you may have about the statements.
