1
99:59:59,999 --> 99:59:59,999
(Male announcer) Thank you for downloading from the BBC.
2
99:59:59,999 --> 99:59:59,999
For details of our complete range of podcasts and our Terms of Use,
3
99:59:59,999 --> 99:59:59,999
go to bbcworldservice.com/podcasts .
4
99:59:59,999 --> 99:59:59,999
(Female announcer) Governments worldwide battle to control and contain terrorism.
5
99:59:59,999 --> 99:59:59,999
Police and the courts struggle to separate harmless loners from dangerous lone wolves.
6
99:59:59,999 --> 99:59:59,999
Opinions differ on the most effective way to combat terrorist attacks,
7
99:59:59,999 --> 99:59:59,999
from military interventions on the ground, to curbing political and religious radicalisation.
8
99:59:59,999 --> 99:59:59,999
But in this edition of Discovery, we'll be hearing about a more unusual new weapon
9
99:59:59,999 --> 99:59:59,999
that might be used in the future to fight terrorism:
10
99:59:59,999 --> 99:59:59,999
maths (1:00)
11
99:59:59,999 --> 99:59:59,999
(A) We were looking at the data in a new way, we were
using tools that were somewhat foreign,
12
99:59:59,999 --> 99:59:59,999
(A) these were tools that came out of physics
and complex systems, not tools that necessarily came out of the political science community.
13
99:59:59,999 --> 99:59:59,999
(A) And we were saying things that were kind of weird.
14
99:59:59,999 --> 99:59:59,999
(B) One thinks of terrorism as something very random,
something so strange that it must be done in a very
chaotic way.
15
99:59:59,999 --> 99:59:59,999
(B) But of course, in the end, it is an activity, it's a
human activity, so it's quite interesting then that the
patterns that you seen in the events are not random.
16
99:59:59,999 --> 99:59:59,999
(A) For terrorism, that had somewhat shocking implications.
17
99:59:59,999 --> 99:59:59,999
(A)If you understand the frequencies of the small
events, you can extrapolate,
18
99:59:59,999 --> 99:59:59,999
(A) and then make a forecast out into the future,
about what the probability should be
19
99:59:59,999 --> 99:59:59,999
(A) for a very large event.
20
99:59:59,999 --> 99:59:59,999
(HF) So, could maths predict the next 9/11?
21
99:59:59,999 --> 99:59:59,999
(HF) You're listening to the BBC World Service, and today
on Discovery, I'll be looking at the hidden mathematical
pattern that is being discovered in global terrorism;
22
99:59:59,999 --> 99:59:59,999
(HF) a pattern that lies behind a host of diverse phenomena,
from economics to earthquakes.
23
99:59:59,999 --> 99:59:59,999
(HF) I'm Dr Hannah Fry, and I'm a mathematician from
University College London, working on complex systems.
24
99:59:59,999 --> 99:59:59,999
(HF) These are systems, like terrorism, which at first seem
complex, and random
25
99:59:59,999 --> 99:59:59,999
(HF) but if you stand back and study the bigger picture, then
a surprising number of patterns can appear;
26
99:59:59,999 --> 99:59:59,999
(HF) patterns which you can describe using mathematics.
27
99:59:59,999 --> 99:59:59,999
(HF) Dr (?), a computer scientist at the University of
Colorado, was one of the first to find a tangible,
mathematical connection underlying terrorism.
28
99:59:59,999 --> 99:59:59,999
(HF) He looked at 30,000 terror attacks worldwide, over
40 years
29
99:59:59,999 --> 99:59:59,999
(HF) and for all of the events, he counted how many times a
certain number of people were killed, and plotted it on a graph
30
99:59:59,999 --> 99:59:59,999
(HF) and the results were remarkable.
31
99:59:59,999 --> 99:59:59,999
(Dr) This initial analysis we did, it was quite shocking,
we found this thing that looked like
32
99:59:59,999 --> 99:59:59,999
(Dr) what's called a power law distribution, which is a
very special kind of mathematical pattern that usually
33
99:59:59,999 --> 99:59:59,999
(Dr) crops up in physics, in fact, but increasingly is
observed in social and biological systems,
34
99:59:59,999 --> 99:59:59,999
(Dr) and this is somewhat surprising, because when we
think about terrorism,
35
99:59:59,999 --> 99:59:59,999
(Dr) we think mainly about the capricious, highly
contextualised nature of the individual actors
36
99:59:59,999 --> 99:59:59,999
(Dr) that carry out these events,
37
99:59:59,999 --> 99:59:59,999
(Dr) and yet, at the global level, we see this remarkable
pattern, this power law pattern, emerge.
38
99:59:59,999 --> 99:59:59,999
(HF) So it's obviously a bit tricky to describe a graph
on the radio. (Dr) [Laughs] Yes.
39
99:59:59,999 --> 99:59:59,999
(HF) but could you give us an idea of what a power law
looks like,
40
99:59:59,999 --> 99:59:59,999
(HF) perhaps, compared to some other distributions that
people might be familiar with?
41
99:59:59,999 --> 99:59:59,999
(Dr) So a power law distribution is very different from
42
99:59:59,999 --> 99:59:59,999
(Dr) what most of us experience, and our intuition
is built around, as human beings.
43
99:59:59,999 --> 99:59:59,999
(Dr) Most of our world is wrapped up in what are called
Gaussian, or normal, distributions.
44
99:59:59,999 --> 99:59:59,999
(Dr) So, the range of heights that we experience
among other humans,
45
99:59:59,999 --> 99:59:59,999
(Dr) has what's called a normal distribution.
(HF) Like a bell curve.
46
99:59:59,999 --> 99:59:59,999
(Dr) [Confirming] Like a bell curve. Which means that
there's an average,
47
99:59:59,999 --> 99:59:59,999
(Dr) that is representative of essentially the entire population.
48
99:59:59,999 --> 99:59:59,999
(NJ) One of the first graphs that we ever draw in
school is one of heights of people in the classroom.
49
99:59:59,999 --> 99:59:59,999
(HF) Neil Johnson, professor of Physics at the
University of Miami.
50
99:59:59,999 --> 99:59:59,999
(NJ) There's usually some great big peak, and there's a
little bit of spreading either side of the peak,
51
99:59:59,999 --> 99:59:59,999
(NJ) Might be something like, you know, for adults,
5 foot 6 or something like this,
52
99:59:59,999 --> 99:59:59,999
(NJ) that's the average, and of course people have wide
variation; basketball players, and there's also
53
99:59:59,999 --> 99:59:59,999
(NJ) people who are much shorter, but nobody's a foot
tall in the adult population, and nobody's 20 feet tall.
54
99:59:59,999 --> 99:59:59,999
(NJ) Well, that's not how it works for the severity of
attacks.
55
99:59:59,999 --> 99:59:59,999
(NJ) You might think it would have, you know, in an
attack, people use an explosive device.
56
99:59:59,999 --> 99:59:59,999
(NJ) You might think it blows up a typical number of
people, you know, plus or minus 3.
57
99:59:59,999 --> 99:59:59,999
(NJ) But no, it's a completely different distribution.
58
99:59:59,999 --> 99:59:59,999
(NJ) It is the equivalent of having the 20 foot person,
and the 1 foot person happening pretty frequently.
59
99:59:59,999 --> 99:59:59,999
(HF) A power law curve looks like the downward slope
of a hill.
60
99:59:59,999 --> 99:59:59,999
(HF) At the top left, you have a large number of small
attacks that kill a few people,
61
99:59:59,999 --> 99:59:59,999
(HF) and at the bottom right, you have a tiny number of events that are very severe,
62
99:59:59,999 --> 99:59:59,999
(HF) with hundreds, or thousands of deaths.
63
99:59:59,999 --> 99:59:59,999
(HF) Behind this graph lies a very simple equation,
providing a clear mathematical link
64
99:59:59,999 --> 99:59:59,999
(HF) between smalll, frequent attacks, and rare,
large-scale strikes.
65
99:59:59,999 --> 99:59:59,999
(HF) Now this shape has been found time and time
again, over different decades, in different cities,
66
99:59:59,999 --> 99:59:59,999
(HF) and for different terrorist groups.
67
99:59:59,999 --> 99:59:59,999
(HF) And despite huge changes in global geopolitics,
68
99:59:59,999 --> 99:59:59,999
(HF) from the fall of the Soviet Union to the rise of
Islamic extremism,
69
99:59:59,999 --> 99:59:59,999
(HF) this simple mathematical pattern has persisted.
70
99:59:59,999 --> 99:59:59,999
(Dr) The remarkable thing about the power law
distribution in the sizes of terrorist events,
71
99:59:59,999 --> 99:59:59,999
(Dr) is that it seems to be very robust, so it suggests that
this may be a fundamental pattern,
72
99:59:59,999 --> 99:59:59,999
(Dr) it may be that the nature of the modern world
produces this pattern,
73
99:59:59,999 --> 99:59:59,999
(Dr) so from a policy perspective, there is an
implication that changing this pattern,
74
99:59:59,999 --> 99:59:59,999
(Dr) being able to reduce the likelihood of a large
terrorist even like 9/11 happening again
75
99:59:59,999 --> 99:59:59,999
(Dr) may be not as simple as finding the terrorists and
throwing them in jail.
76
99:59:59,999 --> 99:59:59,999
(Dr) It may be more subtle, it may be that the nature of
the global system
77
99:59:59,999 --> 99:59:59,999
(Dr) tends to produce the types of individuals that would then go about carrying out
78
99:59:59,999 --> 99:59:59,999
(Dr) these kinds of events.
79
99:59:59,999 --> 99:59:59,999
(Dr) And that's a much harder problem to solve from
a policy perspective.
80
99:59:59,999 --> 99:59:59,999
(HF) So how easy was it for you to get your work
published on this?
81
99:59:59,999 --> 99:59:59,999
(Dr) (laughs)
(HF) (laughs)
82
99:59:59,999 --> 99:59:59,999
(Dr) I'm laughing because it was not easy.
(HF) Yeah.
83
99:59:59,999 --> 99:59:59,999
(Dr) My colleague, Maxwell, and I started this work in
2003, when the Iraq invasion
84
99:59:59,999 --> 99:59:59,999
(Dr) was really getting going,
85
99:59:59,999 --> 99:59:59,999
(Dr) and the paper was not published until 2007.
86
99:59:59,999 --> 99:59:59,999
(Dr) There were many factors that made it difficult, but
I think one of them was that this was
87
99:59:59,999 --> 99:59:59,999
(Dr) such a weird take on a problem that people had
been thinking about for a long time.
88
99:59:59,999 --> 99:59:59,999
(Dr) Political scientists have been studying terrorism for
decades,
89
99:59:59,999 --> 99:59:59,999
(Dr) so we were looking at the data in a new way,
90
99:59:59,999 --> 99:59:59,999
(Dr) we were talking about terrorism not as a
phenomenon of decision making,
91
99:59:59,999 --> 99:59:59,999
(Dr) but almost as a natural phenomenon.
92
99:59:59,999 --> 99:59:59,999
(Dr) This idea that we could look at the entire world
almost as a natural kind of system,
93
99:59:59,999 --> 99:59:59,999
(Dr) and characterise its patterns without having to refer
to the actual decisions that produced the events.
94
99:59:59,999 --> 99:59:59,999
(Dr) So the unpleasant aspect of trying to get this
work published is that we tried 10 different journals
95
99:59:59,999 --> 99:59:59,999
(HF) Wow.
(Dr) in order to get this published,
96
99:59:59,999 --> 99:59:59,999
(Dr) and sometimes the reviews we got back from
academics, they seemed to be
97
99:59:59,999 --> 99:59:59,999
(Dr) missing the point, in some ways.
98
99:59:59,999 --> 99:59:59,999
(Dr) They didn't understand what we were doing,
or what the results implied.
99
99:59:59,999 --> 99:59:59,999
(HF) So I understand one of the really important
implications is about really large events.
100
99:59:59,999 --> 99:59:59,999
(HF) What did you find there?
101
99:59:59,999 --> 99:59:59,999
(Dr) The fact that the power law pattern exists,
102
99:59:59,999 --> 99:59:59,999
(Dr) and the implication that these are all part of
the same fundamental process,
103
99:59:59,999 --> 99:59:59,999
(Dr) it does imply that the largest events, things like 9/11,
104
99:59:59,999 --> 99:59:59,999
(Dr) will occur with surprising frequency, and the
mathematical function is strong enough
105
99:59:59,999 --> 99:59:59,999
(Dr) that one can actually extrapolate, and then make a
forecast in much the same way
106
99:59:59,999 --> 99:59:59,999
(Dr) that forecasts are made for earthquakes.
107
99:59:59,999 --> 99:59:59,999
(Dr) In fact, the power law distribution is the same
distribution
108
99:59:59,999 --> 99:59:59,999
(Dr) that characterises the frequency of earthquakes,
and the frequency of terrorist events.
109
99:59:59,999 --> 99:59:59,999
(Dr) The shape of the distribution is slightly different,
but the pattern is the same.
110
99:59:59,999 --> 99:59:59,999
(Dr) And so, by applying this mathematical model to
forecasting, we can make an estimate
111
99:59:59,999 --> 99:59:59,999
(Dr) of the probability that we should see an event in the
next 10 years, for instance,
112
99:59:59,999 --> 99:59:59,999
(Dr) that kills as many, or more people as 9/11 did.
113
99:59:59,999 --> 99:59:59,999
(HF) And what did you find?
114
99:59:59,999 --> 99:59:59,999
(Dr) The likelihood across the set of models that we
fitted to the data, over the next 10 years,
115
99:59:59,999 --> 99:59:59,999
(Dr) is somewhere around 30%, which is not a
certainty,
116
99:59:59,999 --> 99:59:59,999
(Dr) but it's still an uncomfortably high number.
117
99:59:59,999 --> 99:59:59,999
(N1) At least a hundred people are reported killed, and
five hundred injured
118
99:59:59,999 --> 99:59:59,999
(N1) in a chain of car bomb attacks across Bombay.
119
99:59:59,999 --> 99:59:59,999
(HF) In 1993, 13 bombs exploded within 3 hours in
Bombay, and at the final count, over 1500 people
120
99:59:59,999 --> 99:59:59,999
(HF) were killed or injured.
121
99:59:59,999 --> 99:59:59,999
(N2) Russia observes two days of national mourning
for the many victims of Beslan.
122
99:59:59,999 --> 99:59:59,999
(N2) More than 100 funerals take place in one day,
attended by thousands of people.
123
99:59:59,999 --> 99:59:59,999
(HF) The Beslan school siege in 2004, carried out by
Chechen rebels,
124
99:59:59,999 --> 99:59:59,999
(HF) hit over 1000 children, parents, and teachers.
125
99:59:59,999 --> 99:59:59,999
(HF) And, after 9/11 - the largest terrorist attack in
human history, which killed almost 3000 people -
126
99:59:59,999 --> 99:59:59,999
(HF) funding was ploughed into new, scientific ways
to try and tackle terrorism,
127
99:59:59,999 --> 99:59:59,999
(HF) from using Game Theory in airport security
to analysing the social networks of terror suspects.
128
99:59:59,999 --> 99:59:59,999
(HF) But now, researchers are using data on terrorism
in insurgencies to try and forecast future attacks.
129
99:59:59,999 --> 99:59:59,999
(HF) Physicist, Neil Johnson, from the University of
Miami, studied the timings between
130
99:59:59,999 --> 99:59:59,999
(HF) terrorist events in one area, and, strangely enough,
131
99:59:59,999 --> 99:59:59,999
(HF )the same kind of power law pattern lay behind this data too.
132
99:59:59,999 --> 99:59:59,999
(NJ) Our initial study was looking at a few regions in
Afghanistan, and Iraq.
133
99:59:59,999 --> 99:59:59,999
(NJ) We've now done the study across basically every
country where we could find data -
134
99:59:59,999 --> 99:59:59,999
(NJ) that includes Africa, includes suicide bombings
of Hezbollah
135
99:59:59,999 --> 99:59:59,999
(NJ) we've looked at suicide attacks in Pakistan,
136
99:59:59,999 --> 99:59:59,999
(NJ) they all seem to follow this relationship.
137
99:59:59,999 --> 99:59:59,999
(HF) So if this happened then, for example, in a UK
city, how confident could you be
138
99:59:59,999 --> 99:59:59,999
(HF) in predicting the timing of the next attack.
139
99:59:59,999 --> 99:59:59,999
(NJ) Yeah, of course it will never be down to the day;
140
99:59:59,999 --> 99:59:59,999
(NJ) it will be, in some situa- like, you know- if it's down
to the week.
141
99:59:59,999 --> 99:59:59,999
(NJ) But what can be very useful is predicting the trend.
142
99:59:59,999 --> 99:59:59,999
(NJ) Are we going to see the events get less
frequent in time?
143
99:59:59,999 --> 99:59:59,999
(NJ) Are we going to see them get more frequent?
144
99:59:59,999 --> 99:59:59,999
(NJ) And if we see them get more frequent, roughly
how many are we going to be getting
145
99:59:59,999 --> 99:59:59,999
(NJ) every few weeks?
146
99:59:59,999 --> 99:59:59,999
(NJ) It's not meant to be like predicting, you know
whether it's going to rain in the next three days,
147
99:59:59,999 --> 99:59:59,999
(NJ) it's meant to be some trend, and I very much see it
in that analogy with say, weather prediction.
148
99:59:59,999 --> 99:59:59,999
(NJ) The medium-term weather forecasts are getting
better and better,
149
99:59:59,999 --> 99:59:59,999
(NJ) but it's still very hard to say that at 5 o' clock,
in three days,
150
99:59:59,999 --> 99:59:59,999
(NJ) rain will fall on a particular place in London.
151
99:59:59,999 --> 99:59:59,999
(NJ) But we're all very interested, and very keen, to
know when some front is moving through,
152
99:59:59,999 --> 99:59:59,999
(NJ) and over the next three days, there'll be some
increase in the trend of rain.
153
99:59:59,999 --> 99:59:59,999
(NJ) That is a very useful statement.
154
99:59:59,999 --> 99:59:59,999
(NJ) And it's exactly that level [of] prediction that we
believe that this type of work is good for.
155
99:59:59,999 --> 99:59:59,999
(NJ) It's never going to say that in five days' time,
[in] a particular place, there'll be a particular attack,
156
99:59:59,999 --> 99:59:59,999
(NJ) but what it is saying is: wait a minute, this group,
in this region of a country,
157
99:59:59,999 --> 99:59:59,999
(NJ) they are escalating.
158
99:59:59,999 --> 99:59:59,999
(HF) So I suppose it goes without saying then, that
these ideas would be incredibly useful
159
99:59:59,999 --> 99:59:59,999
(HF) to counter-terrorism agencies, but how much
interest have they shown in your research?
160
99:59:59,999 --> 99:59:59,999
(NJ) Well, I'm currently funded by the Office of Naval
Research,
161
99:59:59,999 --> 99:59:59,999
(NJ) they've been very supportive, they're very
interested in this.
162
99:59:59,999 --> 99:59:59,999
(NJ) I've also had funding through another agency
which was interested in counter-IED strategies.
163
99:59:59,999 --> 99:59:59,999
(NJ) So I would say that there's been a lot of interest,
164
99:59:59,999 --> 99:59:59,999
(NJ) and we've even had interest, I remember getting
an email from Marines
165
99:59:59,999 --> 99:59:59,999
(NJ) in a forward-operating base in Afghanistan, telling
me that they'd been trying -
166
99:59:59,999 --> 99:59:59,999
(NJ) because the formula is so simple, the mathematics
is actually so simple -
167
99:59:59,999 --> 99:59:59,999
(NJ) that they'd been trying out this particular analysis
of successive events that
168
99:59:59,999 --> 99:59:59,999
(NJ) they'd actually been seeing, and experiencing
around them.
169
99:59:59,999 --> 99:59:59,999
(HF) Wow. So Marines in the field have been in contact
with you, talking about your mathematical models, then?
170
99:59:59,999 --> 99:59:59,999
(NJ) Yeah, the intelligence officer from one of the
units, yes.
171
99:59:59,999 --> 99:59:59,999
(HF) But not everyone is convinced that we can make
such precise predictions.
172
99:59:59,999 --> 99:59:59,999
(HF) Dr Karsten Donnay is a social scientist from
the Swiss Federal Institute of Technology,
173
99:59:59,999 --> 99:59:59,999
(HF) who specialises in conflict modelling and
simulation.
174
99:59:59,999 --> 99:59:59,999
(HF) He thinks there's a danger that these models can
be taken too far.
175
99:59:59,999 --> 99:59:59,999
(KD) I think one has to be very careful about using
this for direct prediction,
176
99:59:59,999 --> 99:59:59,999
(KD) I mean we do distinguish between a prediction
and forecasting.
177
99:59:59,999 --> 99:59:59,999
(KD) If you talk about long-term predictions, then it's
feasible to make statements like this:
178
99:59:59,999 --> 99:59:59,999
(KD) There's (??) this chance that in the next 10 years, an event of a certain size will occur,
179
99:59:59,999 --> 99:59:59,999
(KD) that's definitely possible.
180
99:59:59,999 --> 99:59:59,999
(KD) But in the sense of forecasting; actually telling
when and where events will occur,
181
99:59:59,999 --> 99:59:59,999
(KD) actually it's not really possible.
182
99:59:59,999 --> 99:59:59,999
(KD) The more we try to narrow down predictions,
183
99:59:59,999 --> 99:59:59,999
(KD) the more we're running the risk of false positives.
184
99:59:59,999 --> 99:59:59,999
(HF) You need a few ingredients to create a
mathematical model:
185
99:59:59,999 --> 99:59:59,999
(HF) Start with some good data, and then spot the
underlying pattern.
186
99:59:59,999 --> 99:59:59,999
(HF) Finally, pinpoint which vital features in the data
create that pattern,
187
99:59:59,999 --> 99:59:59,999
(HF) and everything else, you can strip out.
188
99:59:59,999 --> 99:59:59,999
(HF) When it comes to terrorism, this if fine if all you
want to do is explain the big relationships,
189
99:59:59,999 --> 99:59:59,999
(HF) or look at the long term,
190
99:59:59,999 --> 99:59:59,999
(HF) but, according to Karsten Donnay, this
simplification limits the kind of predictions
191
99:59:59,999 --> 99:59:59,999
(HF) that you can make.
192
99:59:59,999 --> 99:59:59,999
(KD) What you can use this model effectively for
at the moment -
193
99:59:59,999 --> 99:59:59,999
(KD) and I think this is what a lot of the research is
about -
194
99:59:59,999 --> 99:59:59,999
(KD) [is] to understand fundamental dynamics that
are ongoing.
195
99:59:59,999 --> 99:59:59,999
(KD) And this is always looking into the past.
196
99:59:59,999 --> 99:59:59,999
(KD) But of course, if we understand these dynamics,
we can make at least qualitative assessments
197
99:59:59,999 --> 99:59:59,999
(KD) of what might happen in the future,
198
99:59:59,999 --> 99:59:59,999
(KD) but making a quantitative prediction requires much
more than getting the general gist right;
199
99:59:59,999 --> 99:59:59,999
(KD) you have to really understand what are the driving
motives of when and where things happen,
200
99:59:59,999 --> 99:59:59,999
(KD) and some of it is actually governed by chance - it's
coincidence, the way it plays out -
201
99:59:59,999 --> 99:59:59,999
(KD) and this is something which is systematically
extremely hard to forecast. (14:50)