From KISS to KIDS an antisimplistic modelling approach

  • Slides: 18
Download presentation
From KISS to KIDS – an ‘anti-simplistic’ modelling approach Bruce Edmonds and Scott Moss

From KISS to KIDS – an ‘anti-simplistic’ modelling approach Bruce Edmonds and Scott Moss Centre for Policy Modelling, Manchester Metropolitan University From KISS to KIDS – an ‘anti-simplistic’ modelling approach, B. Edmonds & S. Moss, MAMABS 2004, July 2004, New York, http: //cfpm. org slide-1

The Small Print! • This is necessarily a brief overview of the relevant arguments

The Small Print! • This is necessarily a brief overview of the relevant arguments and examples (15 mins is not enough) • It abstracts the approach taken by a growing band of modellers (see JASSS, ESSA, MABS etc. ) • For many details (e. g. what is meant by: “useful model”, “complexity”, etc. ) you will have to read the paper and other papers of mine cited there • In particular I will not attempt to present my arguments concerning why simplicity is not truthindicative (argue with me later or read the report) • Really I should get off my butt and write a book about all of this but I haven’t (yet). Sorry. From KISS to KIDS – an ‘anti-simplistic’ modelling approach, B. Edmonds & S. Moss, MAMABS 2004, July 2004, New York, http: //cfpm. org slide-2

When KISS may be appropriate KISS may be sensible when: 1. One is trying

When KISS may be appropriate KISS may be sensible when: 1. One is trying to construct/build something with a particular purpose/function in mind • Because something more complex is more difficult to control and hence make behave as one wants 2. A design has not worked, in which case one should resist elaborating it with an ‘easy fix’ • I. e not to avoid a more fundamental re-evaluation of the failure which might require more work KISS makes sense as an engineering maxim From KISS to KIDS – an ‘anti-simplistic’ modelling approach, B. Edmonds & S. Moss, MAMABS 2004, July 2004, New York, http: //cfpm. org slide-3

But… • KISS is also often invoked to justify choices involved in representing complex

But… • KISS is also often invoked to justify choices involved in representing complex phenomena. • Whilst simplicity has many practical advantages in terms of: implementation, parameterisation; manipulation; verification; validation; etc. …it is sometimes claimed that the advantages of simplicity go beyond these pragmatic advantages, in other words that: (Somehow) a simpler theory is more likely to be true (or nearer to the truth etc. ) It is that we arguing against! From KISS to KIDS – an ‘anti-simplistic’ modelling approach, B. Edmonds & S. Moss, MAMABS 2004, July 2004, New York, http: //cfpm. org slide-4

“For the sake of simplicity” • The (in our view mistaken) philosophical tradition that

“For the sake of simplicity” • The (in our view mistaken) philosophical tradition that simplicity is somehow truth-indicative provides (false) legitimacy to the reason: “for the sake of simplicity”… • which otherwise would otherwise be expressed in more practical terms (e. g. “we had limited time” or “the simulation would have taken too long”) • These practical reasons are legitimate because they are inevitable – we are limited beings! • Masking true reasons hinders the scientific process by impeding communication as to assumptions and limitations in a simulation From KISS to KIDS – an ‘anti-simplistic’ modelling approach, B. Edmonds & S. Moss, MAMABS 2004, July 2004, New York, http: //cfpm. org slide-5

Indications that KISS might not to be useful for modelling complex MAS • Example

Indications that KISS might not to be useful for modelling complex MAS • Example 1: Social embedding – agents’ particular network of relationships is important – is widespread in observed social systems – is known to occur in MAS: e. g. with agents that: partially compete; can develop in an open-ended way; and can observe the actions of others (Edmonds 1999) • Example 2: Globally Coupled Chaotic Systems – in the presence of even weak non-local coupling – and chaos in the behaviour of separate parts – residue unpredictable “noise” can turn out not to obey the law of large numbers (Kaneko 1990) From KISS to KIDS – an ‘anti-simplistic’ modelling approach, B. Edmonds & S. Moss, MAMABS 2004, July 2004, New York, http: //cfpm. org slide-6

Counters to some arguments for the KISS approach applied to modelling • Many apparently

Counters to some arguments for the KISS approach applied to modelling • Many apparently complex systems have turned out to be modellable in fairly simple ways …this does not mean that any particular complex system will have an underlying simplicity, rather we should look at the evidence in each case • In ALife (and elsewhere) complex behaviours result from interaction of many simple behaviours …even if this is the case this does not mean that these simple behaviours are recoverable from the system now • MAS are deterministic computer programs so there must be a simple reason - their program and/or specification …local determinism does not lead to effective predictability of global long-term behaviour (Turing 1936) similar problems beset inference from a specification (Edmonds and Bryson 2004) From KISS to KIDS – an ‘anti-simplistic’ modelling approach, B. Edmonds & S. Moss, MAMABS 2004, July 2004, New York, http: //cfpm. org slide-7

Rather we suggest: KIDS – “Keep it Descriptive Stupid” Supposed to suggest a modelling

Rather we suggest: KIDS – “Keep it Descriptive Stupid” Supposed to suggest a modelling approach where: one starts with a model that is as descriptive as possible (which might be very complex) and is only simplified where this turns out to be justified This contrasts to the KISS approach where: one only tries a more complex model if a simpler one is inadequate The difference is in the starting point for model development and/or exploration – In both model behaviour compared to evidence should (!) drive any subsequent development From KISS to KIDS – an ‘anti-simplistic’ modelling approach, B. Edmonds & S. Moss, MAMABS 2004, July 2004, New York, http: //cfpm. org slide-8

KISS vs. KIDS illustrated KIDS More Complex in Aspect 2 etc. More Complex in

KISS vs. KIDS illustrated KIDS More Complex in Aspect 2 etc. More Complex in Aspect 1 KISS Simplest Possible From KISS to KIDS – an ‘anti-simplistic’ modelling approach, B. Edmonds & S. Moss, MAMABS 2004, July 2004, New York, http: //cfpm. org slide-9

However we need to distinguish intended theory from implementation What we are saying is

However we need to distinguish intended theory from implementation What we are saying is that: • It is the intended theory that should (as a starting point) be as descriptive as possible • And that a (single) simulation should be as direct a representation of this theory as is feasible Of course if there are two completely adequate ways of implementing the same theory as simulations then (for practical reasons) it is sensible to choose the simpler one And it is often infeasible to develop/use a single simulation to represent a theory – then one may need a series of partial models or at staged levels From KISS to KIDS – an ‘anti-simplistic’ modelling approach, B. Edmonds & S. Moss, MAMABS 2004, July 2004, New York, http: //cfpm. org slide-10

Building upwards towards validated general theory – an idealised picture Analytic Model General Theory

Building upwards towards validated general theory – an idealised picture Analytic Model General Theory Abstract Simulation Descriptive Simulation Data. Model Data Model Simple Model Applications From KISS to KIDS – an ‘anti-simplistic’ modelling approach, B. Edmonds & S. Moss, MAMABS 2004, July 2004, New York, http: //cfpm. org slide-11

MABS as part of a move towards descriptive modelling • MABS allows and facilitates

MABS as part of a move towards descriptive modelling • MABS allows and facilitates a more direct correspondence between what is observed and what is modelled • One benefit of this is that new sources of evidence become available for validation, e. g. – Anecdotal evidence from stakeholders and others – Histories of individual actor’s decisions and actions – Participatory approaches • Previously rejected as “unscientific” because it was not possible to include them in formal models • Now a good source of starting points which can be simulated and outcomes then cross-checked From KISS to KIDS – an ‘anti-simplistic’ modelling approach, B. Edmonds & S. Moss, MAMABS 2004, July 2004, New York, http: //cfpm. org slide-12

Example: Exploring Variations of a Model of Domestic Water Demand • Integrated assessment model

Example: Exploring Variations of a Model of Domestic Water Demand • Integrated assessment model of domestic water demand in the mid-Thames region of the UK • An agent-based simulation – agents are the households and the local water “authority” • Combines different kinds of models (climate, behavioural, network, influence, numeric, etc. ) • Is an attempt at a descriptive model based on all evidence available to us: stakeholder input; anecdotal accounts; surveys; other well-validated models (water runoff, etc. ); aggregate time series • Used to explore potential for variation and hence the assumptions behind statistics models of same From KISS to KIDS – an ‘anti-simplistic’ modelling approach, B. Edmonds & S. Moss, MAMABS 2004, July 2004, New York, http: //cfpm. org slide-13

Model Structure From KISS to KIDS – an ‘anti-simplistic’ modelling approach, B. Edmonds &

Model Structure From KISS to KIDS – an ‘anti-simplistic’ modelling approach, B. Edmonds & S. Moss, MAMABS 2004, July 2004, New York, http: //cfpm. org slide-14

Example range of runs from one set of parameter settings Shows 15 runs: aggregate

Example range of runs from one set of parameter settings Shows 15 runs: aggregate water demand (scaled so 1973=100; cross-shaped unwrapped neighbourhoods size 24) From KISS to KIDS – an ‘anti-simplistic’ modelling approach, B. Edmonds & S. Moss, MAMABS 2004, July 2004, New York, http: //cfpm. org slide-15

Summary of findings about this model • Following turned out to be important to

Summary of findings about this model • Following turned out to be important to the character of the aggregate model behaviour: – – – – the number and size of droughts; the distribution of biases of the households; the timing of innovations, the rate of forgetting in learning; the neighbourhood shape; the topology of the influence grid; and the size of the neighbourhoods • The shape of the initial distribution of the use of appliances in households turned out not to be important and hence could be simplified From KISS to KIDS – an ‘anti-simplistic’ modelling approach, B. Edmonds & S. Moss, MAMABS 2004, July 2004, New York, http: //cfpm. org slide-16

Conclusion: A Summary of the KIDS approach • A denial of the existence of

Conclusion: A Summary of the KIDS approach • A denial of the existence of ‘short cuts’ to useful general theory about complex social systems (incl MAS)… …which are not supported and led by a foundation of more descriptive models and evidence. • An assertion of the common-sense, default position that the complexity of a model should be adequate for the complexity of the phenomena one is concerned with …until we know what is relevant for a well-validated general theory as opposed to merely guessing (the case now) In the science of complex social systems (& MAS) (at least): truth comes before beauty From KISS to KIDS – an ‘anti-simplistic’ modelling approach, B. Edmonds & S. Moss, MAMABS 2004, July 2004, New York, http: //cfpm. org slide-17

The End Centre for Policy Modelling cfpm. org Bruce Edmonds cfpm. org/~bruce Scott Moss

The End Centre for Policy Modelling cfpm. org Bruce Edmonds cfpm. org/~bruce Scott Moss cfpm. org/~scott From KISS to KIDS – an ‘anti-simplistic’ modelling approach, B. Edmonds & S. Moss, MAMABS 2004, July 2004, New York, http: //cfpm. org slide-18