An Invited talk at iEMSs, Leipzig 2012 (http://www.iemss.org/sites/iemss2012).
Abstract:
Behaviour in society and the responses from the environment are both highly context-dependent. There is a lot of evidence that hyman cognition and behaviour depends sharply on the percieved context. Human collective and social behaviour is even more so, indeed may be structured around co-determined contexts that are then entrenched within our training, infrastructure and habits. Similarly ecological niches, where species adapt to each other can be highly specific to a particular set of environmental affordences. The response to a pertabation (e.g. reduction of a resource or introduction of a new species) depends highly on the environmental context.
However, to a very large extent, our formal models of the environment and of our interaction with the environment are context-free. It is often simply assumed that the variations due to specific contexts can be dealt with as a kind of "noise" to a main trend or interaction. Whilst this maybe sometimes the case, this assumption is rarely justified by any evidence or indeed convincing argument . Often it seems that context is ignored simply because it seems too difficult to do otherwise, so work proceeds simply on the hope that context-dependency can be treated as a kind of noise. Other strategies to avoid the issue of context include keeping to within a single, very restricted context (which prevents any general conclusions) or remaining in the world of analogy and natural language discourse (where context-dependency is masked by the innate ability of humans to reapply analogies on the fly). I argue that this must often not be the case and that a collection of context dependent interactions if treated in this way, can result in very different outcomes, especially when one needs to scale any conclusions.
I then seek to show some possible ways forward, ways to include some of the context-dependency in our techniques and models. These include kinds of agent-based modelling that include context-awareness in the agents and actors, kinds of data-mining that could be used to search for patterns in a context-dependent manner, and new techniques from the field of visual analytics to visualise and interact with data via a visual interface in a context-friendly manner.
Context in Environmental Modelling– the room around the elephant
1. Context in Environmental Modelling
– the room around the elephant
Bruce Edmonds
Centre for Policy Modelling,
Manchester Metropolitan University
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 1
2. Acknowledgements
Many thanks to all those with whom I have
discussed these ideas, including: Emma Norling,
Nick Shryane, Jason Dykes, Scott Moss, those at
the Conference Series on “Modelling & Using
Context”, the regulars at the Manchester Complexity
Seminar and those in the SCID Project.
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 2
3. Some Questions about Context
• How important is the context when
modelling process/aspect/system X/Y/Z?
• How much can we ignore context…
• …or, conversely, how much of the context
do we have to include within our models?
• If we include context-dependency does that
stop us being scientific?
• How can we square the context-
dependency of the observed/involved world
with our models of that world?
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 3
4. Talk Outline
1. Context-dependency in the environment
2. Context-dependency in human behaviour
3. Some defensive responses to context-
dependency
4. Some possible ways forward
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 4
5. Note on Talking about Context
• The word “context” is used in many different
senses across different fields
• Somewhat of a “dustbin” concept resorted to
when more immediate explanations fail (like
the other “c-word”, complexity)
• Problematic to talk about, as it is not clear that
“contexts” are usually identifiably distinct
• Mentioning “context” is often a signal for a
more “humanities oriented” or
“participatory/involved” approach and hence
resisted by “scientists” who are seeking
general laws
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 5
6. Part 1:
Ecological Context-Dependency
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 6
7. Ecological Context
• A certain kind of environment might provide
certain affordances/difficulties
• Organisms adapt to exploit these but also
create new affordances/difficulties
• Migration between similar ecologies makes
organisms ready to exploit each type available
• The organisms are only fully understandable in
their ecological context – the web of other
organisms and their environment
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 7
8. A (colourful!) Picture of the World
• Each square (patch)
is a different, well-
mixed location
• There are 15 kinds
of location with
individuals in each (4
bit string)
• Small stars are
herbivores, circles
those who have
eaten another (the
bigger the more it
has eaten)
• Different colours
indicate different
species (not all
species are visually
distinguishable)
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 8
9. Brief (!) Model Outline
• Basic energy economy (life tax, 90%
transference, reproduction at 3, birth at 1 etc.)
• Patches and organisms have a binary vector
(lengths 4 and 100 respectively)
• Fixed 100x100 random matrix made at start
that broadly determines…
• …who can eat who (or who extract energy
from environment) determined by eater &
eaten’s binary strings (sum of entries in matrix
at rows and columns indicated by 1s)
• Slow processes of mutation, migration etc.
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 9
10. Simulation at (up to) Reference Point
First Successful Carnivores Simulation
Herbivore Appear “Frozen”
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 10
11. From this point on…
50 times for each of 16 different “aspects” (as
well as none, the base case)…
• Reset world to this point
• “Block” interaction on one of the dimensions
(the entries in the matrix indicated by 1s in that
column/row number are not summed)
• Simulate the world for a further 100 ticks (with
different random seed each time)
• Measure the genetic diversity of the population
overall and by each niche type (average
hamming distance between all distinct agents)
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 11
12. Affect of Blocking Different Aspects of Interaction
(av. over 20 runs after 100 ticks, ±2SD)
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 12
13. Effect of Blocking Aspects of
Interaction by Aspect
Base Case
(no blocking)
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 13
14. Implications of Environmental
Context-Dependency
• Whilst there are some underlying universals
that affect the environment (water, genetics,
energy…)
• What characterises “the” environment is
that it is not singular but a complex,
overlapping patchwork of different
ecological contexts
• We can gain some understanding of what is
happening within each context, but generic
understandings across these can be weak
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 14
15. Part 2:
Context-Dependency in Human
Behaviour
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 15
16. A (simplistic) illustration of context from the
point of view of an actor
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 16
17. Situational Context
• The situation in which an event takes place
• This is indefinitely extensive, it could include
anything relevant or coincident
• The time and place specify it, but relevant
details might not be retrievable from this
• It is almost universal to abstract to what is
relevant about these to a recognised type
when communicating about this
• Thus the question “What was the context?”
often effectively means “What about the
situation do I need to know to understand?
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 17
18. Cognitive Context (CC)
• Many aspects of human cognition are context-
dependent, including: memory, visual perception,
choice making, reasoning, emotion, and language
• The brain somehow deals with situational context
effectively, abstracting kinds of situations so
relevant information can be easily and preferentially
accessed
• The relevant correlate of the situational context will
be called the cognitive context
• It is not known how the brain does this, and
probably does this in a rich and complex way that
might prevent easy labeling/reification of contexts
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 18
19. The Context Heuristic
• The kind of situation is recognised in a rich,
fuzzy, complex and unconscious manner
• Knowledge, habits, norms etc. are learnt for
that kind of situation and are retrieved for it
• Reasoning, learning, interaction happens with
respect to the recognised kind of situation
• Context allows for the world to be dealt with by
type of situation, and hence makes
reasoning/learning etc. feasible
• It is a fallible heuristic…
• …so why do we have this kind of cognition?
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 19
20. Social Intelligence Hypothesis
• Kummer, H., Daston, L., Gigerenzer, G. and Silk, J. (1997)
• The crucial evolutionary advantages that
human intelligence gives are due to the
social abilities it allows
• Explains specific abilities such as imitation,
language, social norm instinct, lying,
alliances, gossip, politics etc.
• Social intelligence is not a result of general
intelligence, but at the core of human
intelligence, “general” intelligence is a side-
effect of social intelligence
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 20
21. An Evolutionary Perspective
Social intelligence implies that:
• Groups of humans can develop their own
(sub)cultures of technologies, etc. (Boyd and
Richerson 1985)
• These allow the group with their culture to
inhabit a variety of ecological niches (e.g.
the Kalahari, Polynesia) (Reader 1980)
• Thus humans, as a species, are able to
survive catastrophes that effect different
niches in different ways (specialisation)
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 21
22. Implications of SIH
• That different complex “cultures” of knowledge
are significant
• An important part of those cultures is how to
socially organise, behave, coordinate etc.
• One should expect different sets of social
knowledge for different groups of people
• That these might not only be different in terms
of content but imply different ways of
coordinating, negotiating, cooperating etc.
• That these will relate as a complete “package”
to a significant extent
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 22
23. Social Embedding
• Granovetter (1985)
• Contrasts with the under- and over-socialised
models of behaviour
• That the particular patterns of social
interactions between individuals matter
• In other words, only looking at individual
behaviour or aggregate behaviour misses
crucial aspects
• That the causes of behaviour might be spread
throughout a society – “causal spread”
• Shown clearly in some simulation models
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 23
24. Illustration of Causal Complexity
Lines indicate causal link in behaviour, each box an agent
(Edmonds 1999)
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 24
25. Implications of Social Embedding
• In many circumstances agents can learn to
exploit the computation and knowledge in their
society, rather than do it themselves (invest in
what Warren Buffet invests in)
• Knowledge is often not explicit but is
something learned – this takes time
• This is particularly true of social knowledge –
studying guides as to living in a culture are not
the same as living there for a time
• Social embedding means that human
behaviour can not be understood well separate
from its cultural context
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 25
26. The Social Co-Development of Shared
Recognised Context
• Over time, due to their similarities, certain kinds of
situation become recognised as similar by
participants
• This facilitates the development of a set of shared
habits, norms, knowledge, language etc. that is
specific to the context
• The more this happens the more distinctive that
kind of situation becomes and hence more
recognisable by newcomers
• Eventually these may become institutionalised in
terms of infranstructure, training etc. (e.g. how to
behave in a lecture theatre)
• This co-development of context may be the reason
for its social/evolutionary value
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 26
27. Implications of the Context-
Dependency of Human Behaviour
• Behaviour of observed actors might change sharply
across different social contexts
• The relevant behaviour, norms, kinds of interaction
etc. might also change
• Social contexts are co-developed and changing
• They may be different for different groups
• Some kinds of social behaviour seem to be
inherently context-dependent (compliance)
• It is unlikely that a lot of key social knowledge,
norms, behaviour etc. will be generic
• Models that assume a cross-context engine of
human behaviour may be deeply misleading!
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 27
28. My Central Point
• Given the sharp context-dependency
of both human behaviour and the
environment…
• …how is it that a lot of our models use
generic models of human behaviour
and/or the environmental response?
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 28
29. Part 3:
Defensive Responses
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 29
30. Some Possible Responses
• Its too difficult, I’ll ignore it
• I am looking at the wider/more general picture,
what is common across contexts
• I treat intra-context variation as random noise
• I have included context, it is the variables a, b, c
etc. which vary with the context
• I am acting within context only
• I am only modelling a single context
• It is not scientific
• I need an analytic expression for my model
• Use natural language/analogical models only
• I don’t have enough data
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 30
31. Ignoring Context
• Much modelling happens with a single
context in mind, in which case it can be
ignored but only if
– everyone is using the same idea of this context
– there is no significant “leakage” of causation
from outside the background, that is the scope
is wide enough to include all significant
influencing factors
– The actors/organisms don’t deal with the same
situation as different cognitive contexts
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 31
32. The “Simple is more General” Fallacy
• If one has a general model one can make it
more specific (less general) by adding more
processes/aspects…
• …in which case it can become more complex
• However, the reverse is no true…
• If one simplifies/abstracts then you don’t get a
more general model (well almost never)!
– there may be no simpler model that is good
enough for your purpose
– But, even if there is, you don’t know which aspects
can be safely omitted – if you remove an essential
aspect if will be wrong everywhere (no generality)
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 32
33. Context-Dependency and
Randomness
Lots of
information
lost if
randomness
used to
“model”
contextual
variation
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 33
34. Scaling by Size
• Look at variance as system size increases…
• Does variance as a proportion of size disappear?
• In this case Law of large numbers does not apply
• Simple examples:
• Kaneko (1990): parallel globally coupled chaotic processes
• Edmonds (199?): scaling Brian Arthur’s “El Farol Bar” Model
Contextual variation
Variance
(scaled by size)
Model with random noise
Size
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 34
35. Context-Dependency
and “Being Scientific”
• If the relevant context can be reliably
indentified then…
• …context-dependency is not the same as
subjectivity (even if there are a some hard
cases that escape definition)
• Generality is nice if you can get it, but its no
good pretending to have it if you can’t
• Science should adapt to what it wishes to
understand, not the other way around
• It does mean (often) an acceptance that
general/generic approaches are not useful
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 35
36. Analogical Thinking
• Humans are good at using analogies, relating an idea or
example from one context to another in a rich, relevant
and flexible manner – it is a powerful method of thought
• They build the mapping from the analogy to the a
context “on the fly”, largely unconsciously
• The mappings are different each time an analogy is
applied, thus not a reliable source of transmittable
knowledge – each person might build a different
mapping unless they inhabit the same context
• Many published models do not have an explicit mapping
to a domain, but are used more as analogy
• This is sometimes hidden, so when a simulation (or
analytic model) does not directly map to observations
but to an idea which then applies as an analogy to the
domain and not directly, this gives a spurious
impression of generality
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 36
37. Part 4:
Some Ways Forward
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 37
38. Some ways forward
• Keeping the data and simply NOT summarising it (at least not
prematurely)
• Data mining local patterns to detect commonality of multiple
models/measurements across similar contexts
• More complex simulation models with context-dependent
cognitive models
• Context-sensitive microsimulation models
• Context-oriented visualisation techniques
• Use of “mundane”, context-specific models of human behavior
rather than ambitious generic ones
• Integrating personal/anecdotal accounts of behaviour –
making use of qualitative evidence
• Not leaving the context(s) – acting within the normal sphere of
shared and relevant situations
• Staging abstraction more gradually
• Clusters of related models
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 38
39. Cleveland Heart Disease Data Set – the
processed sub-set used
In processed sub-set:
• 281 entries
• 14 numeric or numerically coded attributes
• Attribute 14 is the outcome (0, 1, 2, 3, 4)
• Some attributes: age, sex, resting blood
pressure (trestpbs), cholesterol (chol),
fasting blood sugar (fbs), maximum heart
rate (thalach), number of major vessels (0-
3) colored by flourosopy (ca)
• From the Machine Learning Repository
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 39
40. Fitting a Global Model (R=56%)
Num = -0.01*age + 0.17*sex + 0.20*cp + 0.00*trestbps + 0.10*restecg + -
0.01*thalach + 0.23*exang + 0.18*oldpeak + 0.16*slope + 0.43*ca + 0.14*thal + -
0.60 (+/- 0.83)
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 40
41. Looking for Clusters in HD Data Set
(Start of Process)
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 41
42. Final Set of Clustered Solutions
• Final solution
set after some
time.
• Still complex but
some structure
is revealed
• Note presence
of “fbs” despite
not being
globally
correlated and
that “chol”
helped define
the context
space
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 42
43. Clusters of Model Scopes suggest a
Context M 1
M1 M2
suggests a context
A useful context is one that:
– includes related models with different
goals/predictions but similar scope
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 43
44. Basic Cognitive Model
Reasoning/plan
Context
ning/belief
Recognition
revision/etc.
Context-Structured
Memory
• Rich, automatic, imprecise, messy cognitive
context recognition using many inputs
(including maybe internal ones)
• Crisp, costly, conscious, explicit cognitive
processes using material indicated by
cognitive context
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 44
45. Example – models in the cognition of
a trading agent
950
Volatility - past 5 periods
900
850
800
750
700
750 850 950
Volume - past 5 periods
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 45
46. The model contents in snapshot of
one trader
model-256 priceLastWeek [stock-4]
model-274 priceLastWeek [stock-5]
model-271 doneByLast [normTrader-5] [stock-4]
model-273 IDidLastTime [stock-2]
model-276 IDidLastTime [stock-5]
minus
[divide
[priceLastWeek [stock-2]]
model-399 [priceLastWeek [stock-5]]]
[times
[priceLastWeek [stock-4]]
[priceNow [stock-5]]]
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 46
47. Total Assets in a Typical Run
30000
Total Value of Assets
25000
20000
15000
10000
5000
0
0 100 200 300 400 500
Time
Black=context, White= non-context
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 47
48. Some Simulation Work addressing
Context-Dependency in Cognition
• (Schlosser & al 2005) argue that reputation is
context dependent
• (Edmonds & Norling 2007) looks at difference
that context-dependent learning and reasoning
makes in an artificial stock market
• (Andrighetto & al 2008) show context-
dependent learning of norms is different form a
generic method
• (Tykhonov & al 2008) argue that trust is
context dependent
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 48
49. Conclusions
• Ignoring it and simply hoping it won’t matter is not
an option (if we are serious about our project)
• There are ways forward to meaningfully make
progress in dealing with context-dependency
• And some of these involve the integration of
qualitative/in situ approaches with
quantitative/formal modelling
• We will need a LOT more data both
multi-dimensional and at a finer-granularity, but this
is starting to come on stream
• Context seems to be an important factor impeding
the integration of both: action-oriented and model-
based approaches, as well as quantitative and
qualitative approaches
• Please help
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 49
50. Ad for a workshop!
The End
Bruce Edmonds
http://bruce.edmonds.name
Centre for Policy Modelling
http://cfpm.org
Context in Environmental Modelling - the room around the elephant, Bruce Edmonds, iEMSs, Leipzig, July 2012, slide 50
Editor's Notes
AI, NL, Sociology, Philosophy, Mobile devices, Psychology, Cognitive ScienceFor detailed argument seem my previous papers on thisDustbin Like complexitywill talk about this problem later
Imagine a professor of physics in a wild place – does his intelligence help him to survive?
Reader 1980, Man on Earth
leakage noisenot the case where un-modelled aspects are effectively randomdiscuss random gas example