Tag Archives: abstraction

Content

The inevitable “layering” of models to extend the reach of our understanding

Leave a comment

By Bruce Edmonds

“Just as physical tools and machines extend our physical abilities, models extend our mental abilities, enabling us to understand and control systems beyond our direct intellectual reach” (Calder  & al. 2018)

Motivation

There is a modelling norm that one should be able to completely understand one’s own model. Whilst acknowledging there is a trade-off between a model’s representational adequacy and its simplicity of formulation, this tradition assumes there will be a “sweet spot” where the model is just tractable but also good enough to be usefully informative about the target of modelling – in the words attributed to Einstein, “Everything should be made as simple as possible, but no simpler”1. But what do we do about all the phenomena where to get an adequate model2 one has to settle for a complex one (where by “complex” I mean a model that we do not completely understand)? Despite the tradition in Physics to the contrary, it would be an incredibly strong assumption that there are no such phenomena, i.e. that an adequate simple model is always possible (Edmonds 2013).

There are three options in these difficult cases.

  • Do not model the phenomena at all until we can find an adequate model we can fully understand. Given the complexity of much around us this would mean to not model these for the foreseeable future and maybe never.
  • Accept inadequate simpler models and simply hope that these are somehow approximately right3. This option would allow us to get answers but with no idea whether they were at all reliable. There are many cases of overly simplistic models leading policy astray (Adoha & Edmonds 2017; Thompson 2022), so this is dangerous if such models influence decisions with real consequences.
  • Use models that are good for our purpose but that we only partially understand. This is the option examined in this paper.

When the purpose is empirical the last option is equivalent to preferring empirical grounding over model simplicity (Edmonds & Moss 2005).

Partially Understood Models

In practice this argument has already been won – we do not completely understand many computer simulations that we use and rely on. For example, due to the chaotic nature of the dynamics of the weather, forecasting models are run multiple times with slightly randomised inputs and the “ensemble” of forecasts inspected to get an idea of the range of different outcomes that could result (some of which might be qualitatively different from the others)4. Working out the outcomes in each case requires the computational tracking of a huge numbers of entities in a way that is far beyond what the human mind can do5. In fact, the whole of “Complexity Science” can be seen as different ways to get some understanding of systems for which there is no analytic solution6.

Of course, this raises the question of what is meant by “understand” a model, for this is not something that is formally defined. This could involve many things, including the following.

  1. That the micro-level – the individual calculations or actions done by the model each time step – is understood. This is equivalent to understanding each line of the computer code.
  2. That some of the macro-level outcomes that result from the computation of the whole model is understood in terms of partial theories or “rules of thumb”.
  3. That all the relevant macro-level outcomes can be determined to a high degree of accuracy without simulating the model (e.g. by a mathematical model).

Clearly, level (1) is necessary for most modelling purposes in order to know the model is behaving as intended. The specification of this micro-level is usually how such models are made, so if this differs from what was intended then this would be a bug. Thus this level would be expected of most models7. However, this does not necessarily mean that this is at the finest level of detail possible – for example, we usually do not bother about how random number generators work, but simply rely on its operation, but in this case we have very good level (3) of understanding for these sub-routines.

At the other extreme, a level (3) understanding is quite rare outside the realm of physics. In a sense, having this level of understanding makes the model redundant, so would probably not be the case for most working models (those used regularly)8. As discussed above, there will be many kinds of phenomena for which this level of understanding is not feasible.

Clearly, what many modelers find useful is a combination of levels (1) & (2) – that is, the detailed, micro-level steps that the model takes are well understood and the outcomes understood well enough for the intended task. For example, when using a model to establish a complex explanation9 (of some observed pattern in data using certain mechanisms or structures) then one might understand the implementation of the candidate mechanisms and verify that the outcomes fit the target pattern for a range of parameters, but not completely understand the detail of the causation involved. There might well be some understanding, for example how robust this is to minor variations in the initial conditions or the working of the mechanisms involved (e.g. by adding some noise to the processes). A complete understanding might not be accessible but this does not stop an explanation being established (although a  better understanding is an obvious goal for future research or avenue for critiques of the explanation).

Of course, any lack of a complete, formal understanding leaves some room for error. The argument here is not deriding the desirability of formal understanding, but is against prioritising that over model adequacy. Also the lack of a formal, level (3), understanding of a model does not mean we cannot take more pragmatic routes to checking it. For example: performing a series of well-designed simulation experiments that intend to potentially refute the stated conclusions, systematically comparing to other models, doing a thorough sensitivity analysis and independently reproducing models can help ensure their reliability. These can be compared with engineering methods – one may not have a proof that a certain bridge design is solid over all possible dynamics, but practical measures and partial modelling can ensure that any risk is so low as to be negligible. If we had to wait until bridge designs were proven beyond doubt, we would simply have to do without them.

Layering Models to Leverage some Understanding

As a modeller, if I do not understand something my instinct is to model it. This instinct does not change if what I do not understand is, itself, a model. The result is a model of the original model – a meta-model. This is, in fact, common practice. I may select certain statistics summarising the outcomes and put these on a graph; I might analyse the networks that have emerged during model runs; I may use maths to approximate or capture some aspect of the dynamics; I might cluster and visualise the outcomes using Machine Learning techniques; I might make a simpler version of the original and compare them. All of these might give me insights into the behaviour of the original model. Many of these are so normal we do not think of this as meta-modelling. Indeed, empirically-based models are already, in a sense, meta-models, since the data that they represent are themselves a kind of descriptive model of reality (gained via measurement processes).

This meta-modelling strategy can be iterated to produce meta-meta-models etc. resulting in “layers” of models, with each layer modelling some aspect of the one “below” until one reaches the data and then what the data measures. Each layer should be able to be compared and checked with the layer “below”, and analysed by the layer “above”.

An extended example of such layering was built during the SCID (Social Complexity of Immigration and Diversity) project10 and illustrated in Figure 1. In this a complicated simulation (Model 1) was built to incorporate some available data and what was known concerning the social and behavioural processes that lead people to bother to vote (or not). This simulation was used as a counter-example to show how assumptions about the chaining effect of interventions might be misplaced (Fieldhouse et al. 2016). A much simpler simulation was then built by theoretical physicists (Model 2), so that it produced the same selected outcomes over time and aa range of parameter values. This allowed us to show that some of the features in the original (such as dynamic networks) were essential to get the observed dynamics in it (Lafuerza et al. 2016a). This simpler model was in turn modelled by an even simpler model (Model 3) that was amenable to an analytic model (Model 4) that allowed us to obtain some results concerning the origin of a region of bistability in the dynamics (Lafuerza et al. 2016b).

Layering fig 1

Figure 1. The Layering of models that were developed in part of the SCID project

Although there are dangers in such layering – each layer could introduce a new weakness – there are also methodological advantages, including the following. (A) Each model in the chain (except model 4) is compared and checked against both the layer below and that above. Such multiple model comparisons are excellent for revealing hidden assumptions and unanticipated effects. (B) Whilst previously what might have happened was a “heroic” leap of abstraction from evidence and understanding straight to Model 3 or 4, here abstraction happens over a series of more modest steps, each of which is more amenable to checking and analysis. When you stage abstraction the introduced assumptions are more obvious and easier to analyse.

One can imagine such “layering” developing in many directions to leverage useful (but indirect) understanding, for example the following.

  • Using an AI algorithm to learn patterns in some data (e.g. medical data for disease diagnosis) but then modelling its working to obtain some human-accessible understanding of how it is doing it.
  • Using a machine learning model to automatically identify the different “phase spaces” in model results where qualitatively different model behaviour is exhibited, so one can then try to simplify the model within each phase.
  • Automatically identifying the processes and structures that are common to a given set of models to facilitate the construction of a more general, ‘umbrella’ model that approximates all the outcomes that would have resulted from the set, but within a narrower range of conditions.

As the quote at the top implies, we are used to settling for partial control of what machines do because it allows us to extend our physical abilities in useful ways. Each time we make their control more indirect, we need to check that this is safe and adequate for purpose. In the cars we drive there are ever more layers of electronic control between us and the physical reality it drives through which we adjust to – we are currently adjusting to more self-drive abilities. Of course, the testing and monitoring of these systems is very important but that will not stop the introduction of layers that will make them safer and more pleasant to drive.

The same is true of our modelling, which we will need to apply in ever more layers in order to leverage useful understanding which would not be accessible otherwise. Yes, we will need to use practical methods to test their fitness for purpose and reliability, and this might include the complete verification of some components (where this is feasible), but we cannot constrain ourselves to only models we completely understand.

Concluding Discussion

If the above seems obvious, then why am I bothering to write this? I think for a few reasons. Firstly, to answer the presumption that understanding one’s model must have priority over all other considerations (such as empirical adequacy) so that sometimes we must accept and use partially understood models. Secondly, to point out that such layering has benefits as well as difficulties – especially if it can stage abstraction into more verifiable steps and thus avoid huge leaps to simple but empirically-isolated models. Thirdly, because such layering will become increasingly common and necessary.

In order to extend our mental reach further, we will need to develop increasingly complicated and layered modelling. To do this we will need to accept that our understanding is leveraged via partially understood models, but also to develop the practical methods to ensure their adequacy for purpose.

Notes

[1] These are a compressed version of his actual words during a 1933 lecture, which were: “It can scarcely be denied that the supreme goal of all theory is to make the irreducible basic elements as simple and as few as possible without having to surrender the adequate representation of a single datum of experience.” (Robinson 2018)
[2] Adequate for whatever our purpose for it is (Edmonds & al. 2019).
[3]The weasel words I once heard from a Mathematician excusing an analytic model he knew to be simplistic were: that, although he knew it was wrong, it was useful for “capturing core dynamics” (though how he knew that they were not completely wrong eludes me).
[4] For an introduction to this approach read the European Centre for Medium-Range Weather Forecasts’ fact sheet on “Ensemble weather forecasting” at: https://www.ecmwf.int/en/about/media-centre/focus/2017/fact-sheet-ensemble-weather-forecasting
[5] In principle, a person could do all the calculations involved in a forecast but only with the aid of exterior tools such as pencil and paper to keep track of it all so it is arguable whether the person doing the individual calculations has an “understanding” of the complete picture. Lewis Fry Richardson, who pioneered the idea of numerical forecasting of weather in the 1920s, did a 1-day forecast by hand to illustrate his method (Lynch 2008), but this does not change the argument.
[6] An analytic solution is when one can obtain a closed-form equation that characterises all the outcomes by manipulating the mathematical symbols in a proof. If one has to numerically calculate outcomes for different initial conditions and parameters this is a computational solution.
[7] For purely predictive models, whose purpose is only to anticipate an unknown value to a useful level of accuracy, this is not strictly necessary. For example, how some AI/Machine learning models work may not clear at the micro-level, but as long as it works (successfully predicts) this does not matter – even if its predictive ability is due to a bug.
[8] Models may still be useful in this case, for example to check the assumptions made in the matching mathematical or other understanding.
[9] For more on this use see (Edmonds et al. 2019).
[10] For more about this project see http://cfpm.org/scid

Acknowledgements

Bruce Edmonds is supported as part of the ESRC-funded, UK part of the “ToRealSim” project, 2019-2023, grant number ES/S015159/1 and was supported as part of the EPSRC-funded “SCID” project 2010-2016, grant number EP/H02171X/1.

References

Calder, M., Craig, C., Culley, D., de Cani, R., Donnelly, C.A., Douglas, R., Edmonds, B., Gascoigne, J., Gilbert, N. Hargrove, C., Hinds, D., Lane, D.C., Mitchell, D., Pavey, G., Robertson, D., Rosewell, B., Sherwin, S., Walport, M. and Wilson, A. (2018) Computational modelling for decision-making: where, why, what, who and how. Royal Society Open Science, DOI:10.1098/rsos.172096.

Edmonds, B. (2013) Complexity and Context-dependency. Foundations of Science, 18(4):745-755. DOI:10.1007/s10699-012-9303-x

Edmonds, B. and Moss, S. (2005) From KISS to KIDS – an ‘anti-simplistic’ modelling approach. In P. Davidsson et al. (Eds.): Multi Agent Based Simulation 2004. Springer, Lecture Notes in Artificial Intelligence, 3415:130–144. DOI:10.1007/978-3-540-32243-6_11

Edmonds, B., le Page, C., Bithell, M., Chattoe-Brown, E., Grimm, V., Meyer, R., Montañola-Sales, C., Ormerod, P., Root H. & Squazzoni. F. (2019) Different Modelling Purposes. Journal of Artificial Societies and Social Simulation, 22(3):6. DOI:10.18564/jasss.3993

Fieldhouse, E., Lessard-Phillips, L. & Edmonds, B. (2016) Cascade or echo chamber? A complex agent-based simulation of voter turnout. Party Politics. 22(2):241-256.  DOI:10.1177/1354068815605671

Lafuerza, LF, Dyson, L, Edmonds, B & McKane, AJ (2016a) Simplification and analysis of a model of social interaction in voting, European Physical Journal B, 89:159. DOI:10.1140/epjb/e2016-70062-2

Lafuerza L.F., Dyson L., Edmonds B., & McKane A.J. (2016b) Staged Models for Interdisciplinary Research. PLoS ONE, 11(6): e0157261. DOI:10.1371/journal.pone.0157261

Lynch, P. (2008). The origins of computer weather prediction and climate modeling. Journal of Computational Physics, 227(7), 3431-3444. DOI:10.1016/j.jcp.2007.02.034

Robinson, A. (2018) Did Einstein really say that? Nature, 557, 30. DOI:10.1038/d41586-018-05004-4

Thompson, E. (2022) Escape from Model Land. Basic Books. ISBN-13: 9781529364873

Edmonds, B. (2023) The inevitable “layering” of models to extend the reach of our understanding. Review of Artificial Societies and Social Simulation, 9 Feb 2023. https://rofasss.org/2023/02/09/layering

© The authors under the Creative Commons’ Attribution-NoDerivs (CC BY-ND) Licence (v4.0)

Content

Socio-Cognitive Systems – a position statement

Leave a comment

By Frank Dignum1, Bruce Edmonds2 and Dino Carpentras3

1Department of Computing Science, Faculty of Science and Technology, Umeå University, frank.dignum@umu.se
2Centre for Policy Modelling, Manchester Metropolitan University, bruce@edmonds.name
3Department of Psychology, University of Limerick, dino.carpentras@gmail.com

In this position paper we argue for the creation of a new ‘field’: Socio-Cognitive Systems. The point of doing this is to highlight the importance of a multi-levelled approach to understanding those phenomena where the cognitive and the social are inextricably intertwined – understanding them together.

What goes on ‘in the head’ and what goes on ‘in society’ are complex questions. Each of these deserves serious study on their own – motivating whole fields to answer them. However, it is becoming increasingly clear that these two questions are deeply related. Humans are fundamentally social beings, and it is likely that many features of their cognition have evolved because they enable them to live within groups (Herrmann et al. 20007). Whilst some of these social features can be studied separately (e.g. in a laboratory), others only become fully manifest within society at large. On the other hand, it is also clear that how society ‘happens’ is complicated and subtle and that these processes are shaped by the nature of our cognition. In other words, what people ‘think’ matters for understanding how society ‘is’ and vice versa. For many reasons, both of these questions are difficult to answer. As a result of these difficulties, many compromises are necessary in order to make progress on them, but each compromise also implies some limitations. The main two types of compromise consist of limiting the analysis to only one of the two (i.e. either cognition or society)[1]. To take but a few examples of this.

  1. Neuro-scientists study what happens between systems of neurones to understand how the brain does things and this is so complex that even relatively small ensembles of neurones are at the limits of scientific understanding.
  2. Psychologists see what can be understood of cognition from the outside, usually in the laboratory so that some of the many dimensions can be controlled and isolated. However, what can be reproduced in a laboratory is a limited part of behaviour that might be displayed in a natural social context.
  3. Economists limit themselves to the study of the (largely monetary) exchange of services/things that could occur under assumptions of individual rationality, which is a model of thinking not based upon empirical data at the individual level. Indeed it is known to contradict a lot of the data and may only be a good approximation for average behaviour under very special circumstances.
  4. Ethnomethodologists will enter a social context and describe in detail the social and individual experience there, but not generalise beyond that and not delve into the cognition of those they observe.
  5. Other social scientists will take a broader view, look at a variety of social evidence, and theorise about aspects of that part of society. They (almost always) do not include individual cognition into account in these and do not seek to integrate the social and the cognitive levels.

Each of these in the different ways separate the internal mechanisms of thought from the wider mechanisms of society or limits its focus to a very specific topic. This is understandable; what each is studying is enough to keep them occupied for many lifetimes. However, this means that each of these has developed their own terms, issues, approaches and techniques which make relating results between fields difficult (as Kuhn, 1962, pointed out).

SCS Picture 1

Figure 1: Schematic representation of the relationship between the individual and society. Individuals’ cognition is shaped by society, at the same time, society is shaped by individuals’ beliefs and behaviour.

This separation of the cognitive and the social may get in the way of understanding many things that we observe. Some phenomena seem to involve a combination of these aspects in a fundamental way – the individual (and its cognition) being part of society as well as society being part of the individual. Some examples of this are as follows (but please note that this is far from an exhaustive list).

  • Norms. A social norm is a constraint or obligation upon action imposed by society (or perceived as such). One may well be mistaken about a norm (e.g. whether it is ok to casually talk to others at a bus stop), thus it is also a belief – often not told to one explicitly but something one needs to infer from observation. However, for a social norm to hold it also needs to be an observable convention. Decisions to violate social norms require that the norm is an explicit (referable) object in the cognitive model. But the violation also has social consequences. If people react negatively to violations the norm can be reinforced. But if violations are ignored it might lead to a norm disappearing. How new norms come about, or how old ones fade away, is a complex set of interlocking cognitive and social processes. Thus social norms are a phenomena that essentially involves both the social and the cognitive (Conte et al. 2013).
  • Joint construction of social reality. Many of the constraints on our behaviour come from our perception of social reality. However, we also create this social reality and constantly update it. For example, we can invent a new procedure to select a person as head of department or exit a treaty and thus have different ways of behaving after this change. However, these changes are not unconstrained in themselves. Sometimes the time is “ripe for change”, while at other times resistance is too big for any change to take place (even though a majority of the people involved would like to change). Thus what is socially real for us depends on what people individually believe is real, but this depends in complex ways on what other people believe and their status. And probably even more important: the “strength” of a social structure depends on the use people make of it. E.g. a head of department becomes important if all decisions in the department are deferred to the head. Even though this might not be required by university or law.
  • Identity. Our (social) identity determines the way other people perceive us (e.g. a sports person, a nerd, a family man) and therefore creates expectations about our behaviour. We can create our identities ourselves and cultivate them, but at the same time, when we have a social identity, we try to live up to it. Thus, it will partially determine our goals and reactions and even our feeling of self-esteem when we live up to our identity or fail to do so. As individuals we (at least sometimes) have a choice as to our desired identity, but in practice, this can only be realised with the consent of society. As a runner I might feel the need to run at least three times a week in order for other people to recognize me as runner. At the same time a person known as a runner might be excused from a meeting if training for an important event. Thus reinforcing the importance of the “runner” identity.
  • Social practices. The concept already indicates that social practices are about the way people habitually interact and through this interaction shape social structures. Practices like shaking hands when greeting do not always have to be efficient, but they are extremely socially important. For example, different groups, countries and cultures will have different practices when greeting and performing according to the practice shows whether you are part of the in-group or out-group. However, practices can also change based on circumstances and people, as it happened, for example, to the practice of shaking hands during the covid-19 pandemic. Thus, they are flexible and adapting to the context. They are used as flexible mechanisms to efficiently fit interactions in groups, connecting persons and group behaviour.

As a result, this division between cognitive and the social gets in the way not only of theoretical studies, but also in practical applications such as policy making. For example, interventions aimed at encouraging vaccination (such as compulsory vaccination) may reinforce the (social) identity of the vaccine hesitant. However, this risk and its possible consequences for society cannot be properly understood without a clear grasp of the dynamic evolution of social identity.

Computational models and systems provide a way of trying to understand the cognitive and the social together. For computational modellers, there is no particular reason to confine themselves to only the cognitive or only the social because agent-based systems can include both within a single framework. In addition, the computational system is a dynamic model that can represent the interactions of the individuals that connect the cognitive models and the social models. Thus the fact that computational models have a natural way to represent the actions as an integral and defining part of the socio-cognitive system is of prime importance. Given that the actions are an integral part of the model it is well suited to model the dynamics of socio-cognitive systems and track changes at both the social and the cognitive level. Therefore, within such systems we can study how cognitive processes may act to produce social phenomena whilst, at the same time, as how social realities are shaping the cognitive processes. Caarley and Newell (1994) discusses what is necessary at the agent level for sociality, Hofested et al. (2021) talk about how to understand sociality using computational models (including theories of individual action) – we want to understand both together. Thus, we can model the social embeddedness that Granovetter (1985) talked about – going beyond over- or under-socialised representations of human behaviour. It is not that computational models are innately suitable for modelling either the cognitive or the social, but that they can be appropriately structured (e.g. sets of interacting parts bridging micro-, meso- and macro-levels) and include arbitrary levels of complexity. Lots of models that represent the social have entities that stand for the cognitive, but do not explicitly represent much of that detail – similarly much cognitive modelling implies the social in terms of the stimuli and responses of an individual that would be to other social entities, but where these other entities are not explicitly represented or are simplified away.

Socio-Cognitive Systems (SCS) are: those models and systems where both cognitive and social complexity are represented with a meaningful level of processual detail.

A good example of an application where this appeared of the biggest importance was in simulations for the covid-19 crisis. The spread of the corona virus on macro level could be given by an epidemiological model, but the actual spreading depended crucially on the human behaviour that resulted from individuals’ cognitive model of the situation. In Dignum (2021) it was shown how the socio-cognitive system approach was fundamental to obtaining better insights in the effectiveness of a range of covid-19 restrictions.

Formality here is important. Computational systems are formal in the sense that they can be unambiguously passed around (i.e. unlike language, it is not differently re-interpreted by each individual) and operate according to their own precisely specified and explicit rules. This means that the same system can be examined and experimented on by a wider community of researchers. Sometimes, even when the researchers from different fields find it difficult to talk to one another, they can fruitfully cooperate via a computational model (e.g. Lafuerza et al. 2016). Other kinds of formal systems (e.g. logic, maths) are geared towards models that describe an entire system from a birds eye view. Although there are some exceptions like fibred logics Gabbay (1996), these are too abstract to be of good use to model practical situations. The lack of modularity and has been addressed in context logics Giunchiglia, F., & Ghidini, C. (1998). However, the contexts used in this setting are not suitable to generate a more general societal model. It results in most typical mathematical models using a number of agents which is either one, two or infinite (Miller and Page 2007), while important social phenomena happen with a “medium sized” population. What all these formalisms miss is a natural way of specifying the dynamics of the system that is modelled, while having ways to modularly describe individuals and the society resulting from their interactions. Thus, although much of what is represented in Socio-Cognitive Systems is not computational, the lingua franca for talking about them is.

The ‘double complexity’ of combining the cognitive and the social in the same system will bring its own methodological challenges. Such complexity will mean that many socio-cognitive systems will be, themselves, hard to understand or analyse. In the covid-19 simulations, described in (Dignum 2021), a large part of the work consisted of analysing, combining and representing the results in ways that were understandable. As an example, for one scenario 79 pages of graphs were produced showing different relations between potentially relevant variables. New tools and approaches will need to be developed to deal with this. We only have some hints of these, but it seems likely that secondary stages of analysis – understanding the models – will be necessary, resulting in a staged approach to abstraction (Lafuerza et al. 2016). In other words, we will need to model the socio-cognitive systems, maybe in terms of further (but simpler) socio-cognitive systems, but also maybe with a variety of other tools. We do not have a view on this further analysis, but this could include: machine learning, mathematics, logic, network analysis, statistics, and even qualitative approaches such as discourse analysis.

An interesting input for the methodology of designing and analysing socio-cognitive systems is anthropology and specifically ethnographical methods. Again, for the covid-19 simulations the first layer of the simulation was constructed based on “normal day life patterns”. Different types of persons were distinguished that each have their own pattern of living. These patterns interlock and form a fabric of social interactions that overall should satisfy most of the needs of the agents. Thus we calibrate the simulation based on the stories of types of people and their behaviours. Note that doing the same just based on available data of behaviour would not account for the underlying needs and motives of that behaviour and would not be a good basis for simulating changes. The stories that we used looked very similar to the type of reports ethnographers produce about certain communities. Thus further investigating this connection seems worthwhile.

For representing the output of the complex socio-cognitive systems we can also use the analogue of stories. Basically, different stories show the underlying (assumed) causal relations between phenomena that are observed. E.g. seeing an increase in people having lunch with friends can be explained by the fact that a curfew prevents people having dinner with their friends, while they still have a need to socialize. Thus the alternative of going for lunch is chosen more often. One can see that the explaining story uses both social as well as cognitive elements to describe the results. Although in the covid-19 simulations we have created a number of these stories, they were all created by hand after (sometimes weeks) of careful analysis of the results. Thus for this kind of approach to be viable, new tools are required.

Although human society is the archetypal socio-cognitive system, it is not the only one. Both social animals and some artificial systems also come under this category. These may be very different from the human, and in the case of artificial systems completely different. Thus, Socio-Cognitive Systems is not limited to the discussion of observable phenomena, but can include constructed or evolved computational systems, and artificial societies. Examination of these (either theoretically or experimentally) opens up the possibility of finding either contrasts or commonalities between such systems – beyond what happens to exist in the natural world. However, we expect that ideas and theories that were conceived with human socio-cognitive systems in mind might often be an accessible starting point for understanding these other possibilities.

In a way, Socio-Cognitive Systems bring together two different threads in the work of Herbert Simon. Firstly, as in Simon (1948) it seeks to take seriously the complexity of human social behaviour without reducing this to overly simplistic theories of individual behaviour. Secondly, it adopts the approach of explicitly modelling the cognitive in computational models (Newell & Simon 1972). Simon did not bring these together in his lifetime, perhaps due to the limitations and difficulty of deploying the computational tools to do so. Instead, he tried to develop alternative mathematical models of aspects of thought (Simon 1957). However, those models were limited by being mathematical rather than computational.

To conclude, a field of Socio-Cognitive Systems would consider the cognitive and the social in an integrated fashion – understanding them together. We suggest that computational representation or implementation might be necessary to provide concrete reference between the various disciplines that are needed to understand them. We want to encourage research that considers the cognitive and the social in a truly integrated fashion. If by labelling a new field does this it will have achieved its purpose. However, there is the possibility that completely new classes of theory and complexity may be out there to be discovered – phenomena that are denied if either the cognitive or the social are not taken together – a new world of a socio-cognitive systems.

Notes

[1] Some economic models claim to bridge between individual behaviour and macro outcomes, however this is traditionally notional. Many economists admit that their primary cognitive models (varieties of economic rationality) are not valid for individuals but are what people on average do – i.e. this is a macro-level model. In other economic models whole populations are formalised using a single representative agent. Recently, there are some agent-based economic models emerging, but often limited to agree with traditional models.

Acknowledgements

Bruce Edmonds is supported as part of the ESRC-funded, UK part of the “ToRealSim” project, grant number ES/S015159/1.

References

Carley, K., & Newell, A. (1994). The nature of the social agent. Journal of mathematical sociology, 19(4): 221-262. DOI: 10.1080/0022250X.1994.9990145

Conte R., Andrighetto G. and Campennì M. (eds) (2013) Minding Norms – Mechanisms and dynamics of social order in agent societies. Oxford University Press, Oxford.

Dignum, F. (ed.) (2021) Social Simulation for a Crisis; Results and Lessons from Simulating the COVID-19 Crisis. Springer.

Herrmann E., Call J, Hernández-Lloreda MV, Hare B, Tomasello M (2007) Humans have evolved specialized skills of social cognition: The cultural intelligence hypothesis. Science 317(5843): 1360-1366. DOI: 10.1126/science.1146282

Hofstede, G.J, Frantz, C., Hoey, J., Scholz, G. and Schröder, T. (2021) Artificial Sociality Manifesto. Review of Artificial Societies and Social Simulation, 8th Apr 2021. https://rofasss.org/2021/04/08/artsocmanif/

Gabbay, D. M. (1996). Fibred Semantics and the Weaving of Logics Part 1: Modal and Intuitionistic Logics. The Journal of Symbolic Logic, 61(4), 1057–1120.

Ghidini, C., & Giunchiglia, F. (2001). Local models semantics, or contextual reasoning= locality+ compatibility. Artificial intelligence, 127(2), 221-259. DOI: 10.1016/S0004-3702(01)00064-9

Granovetter, M. (1985) Economic action and social structure: The problem of embeddedness. American Journal of Sociology 91(3): 481-510. DOI: 10.1086/228311

Kuhn, T,S, (1962) The structure of scientific revolutions. University of Chicago Press, Chicago

Lafuerza L.F., Dyson L., Edmonds B., McKane A.J. (2016) Staged Models for Interdisciplinary Research. PLoS ONE 11(6): e0157261, DOI: 10.1371/journal.pone.0157261

Miller, J. H., Page, S. E., & Page, S. (2009). Complex adaptive systems. Princeton university press.

Newell A, Simon H.A. (1972) Human problem solving. Prentice Hall, Englewood Cliffs, NJ

Simon, H.A. (1948) Administrative behaviour: A study of the decision making processes in administrative organisation. Macmillan, New York

Simon, H.A. (1957) Models of Man: Social and rational. John Wiley, New York

Dignum, F., Edmonds, B. and Carpentras, D. (2022) Socio-Cognitive Systems – A Position Statement. Review of Artificial Societies and Social Simulation, 2nd Apr 2022. https://rofasss.org/2022/04/02/scs

© The authors under the Creative Commons’ Attribution-NoDerivs (CC BY-ND) Licence (v4.0)

Content

The Poverty of Suggestivism – the dangers of “suggests that” modelling

2 Comments

By Bruce Edmonds

Vagueness and refutation

A model[1] is basically composed of two parts (Zeigler 1976, Wartofsky 1979):

  1. A set of entities (such as mathematical equations, logical rules, computer code etc.) which can be used to make some inferences as to the consequences of that set (usually in conjunction with some data and parameter values)
  2. A mapping from this set to what it aims to represent – what the bits mean

Whilst a lot of attention has been paid to the internal rigour of the set of entities and the inferences that are made from them (1), the mapping to what that represents (2) has often been left as implicit or incompletely described – sometimes only indicated by the labels given to its parts. The result is a model that vaguely relates to its target, suggesting its properties analogically. There is not a well-defined way that the model is to be applied to anything observed, but a new map is invented each time it is used to think about a particular case. I call this way of modelling “Suggestivism”, because the model “suggests” things about what is being modelled.

This is partly a recapitulation of Popper’s critique of vague theories in his book “The Poverty of Historicism” (1957). He characterised such theories as “irrefutable”, because whatever the facts, these theories could be made to fit them. Irrefutability is an indicator of a lack of precise mapping to reality – such vagueness makes refutation very hard. However, it is only an indicator; there may be other reasons than vagueness for it not being possible to test a theory – it is their disconnection from well-defined empirical reference that is the issue here.

Some might go as far as suggesting that any model or theory that is not refutable is “unscientific”, but this goes too far, implying a very restricted definition of what ‘science’ is. We need analogies to think about what we are doing and to gain insight into what we are studying, e.g. (Hartman 1997) – for humans they are unavoidable, ‘baked’ into the way language works (Lakoff 1987). A model might make a set of ideas clear and help map out the consequences of a set of assumptions/structures/processes. Many of these suggestivist models relate to a set of ideas and it is the ideas that relate to what is observed (albeit informally) (Edmonds 2001). However, such models do not capture anything reliable about what they refer to, and in that sense are not part of the set of the established statements and theories that is at the core of science  (Arnold 2014).

The dangers of suggestivist modelling

As above, there are valid uses of abstract or theoretical modelling where this is explicitly acknowledged and where no conclusions about observed phenomena are made. So what are the dangers of suggestivist modelling – why am I making such a fuss about it?

Firstly, that people often seem to confuse a model as an analogy – a way of thinking about stuff – and a model that tells us reliably about what we are studying. Thus they give undue weight to the analyses of abstract models that are, in fact, just thought experiments. Making models is a very intimate way of theorising – one spends an extended period of time interacting with one’s model: developing, checking, analysing etc. The result is a particularly strong version of “Kuhnian Spectacles” (Kuhn 1962) causing us to see the world though our model for weeks after. Under this strong influence it is natural to confuse what we can reliably infer about the world and how we are currently perceiving/thinking about it. Good scientists should then pause and wait for this effect to wear off so that they can effectively critique what they have done, its limitations and what its implications are. However, often in the rush to get their work out, modellers often do not do this, resulting in a sloppy set of suggestive interpretations of their modelling.

Secondly, empirical modelling is hard. It is far easier (and, frankly, more fun) to play with non-empirical models. A scientific culture that treats suggestivist modelling as substantial progress and significantly rewards modellers that do it, will effectively divert a lot of modelling effort in this direction. Chattoe-Brown (2018) displayed evidence of this in his survey of opinion dynamics models – abstract, suggestivist modelling got far more reward (in terms of citations) than those that tried to relate their model to empirical data in a direct manner. Abstract modelling has a role in science, but if it is easier and more rewarding then the field will become unbalanced. It may give the impression of progress but not deliver on this impression. In a more mature science, researchers working on measurement methods (steps from observation to models) and collecting good data are as important as the theorists (Moss 1998).

Thirdly, it is hard to judge suggestivist models. Given their connection to the modelling target is vague there cannot be any decisive test of its success. Good modellers should declare the exact purpose of their model, e.g. that is analogical or merely exploring the consequences of theory (Edmonds et al. 2019), but then accept the consequences of this choice – namely, that it excludes  making conclusions about the observed world. If it is for a theoretical exploration then the comprehensiveness of the exploration, the scope of the exploration and the applicability of the model can be judged, but if the model is analogical or illustrative then this is harder. Whilst one model may suggest X, another may suggest the opposite. It is quite easy to fix a model to get the outcomes one wants. Clearly, if a model makes startling suggestions – illustrating totally new ideas or making a counter-example to widely held assumptions – then this helps science by widening the pool of theories or hypotheses that are considered. However most suggestivist modelling does not do this.

Fourthly, their sheer flexibility of as to application causes problems – if one works hard enough one can invent mappings to a wide range of cases, the limits are only those of our imagination. In effect, having a vague mapping from model to what it models adds in huge flexibility in a similar way to having a large number of free (non-empirical) parameters. This flexibility gives an impression of generality, and many desire simple and general models for complex phenomena. However, this is illusory because a different mapping is needed for each case, to make it apply. Given the above (1)+(2) definition of a model this means that, in fact, it is a different model for each case – what a model refers to, is part of the model. The same flexibility makes such models impossible to refute, since one can just adjust the mapping to save them. The apparent generality and lack of refutation means that such models hang around in the literature, due to their surface attractiveness.

Finally, these kinds of model are hugely influential beyond the community of modellers to the wider public including policy actors. Narratives that start in abstract models make their way out and can be very influential (Vranckx 1999). Despite the lack of rigorous mapping from model to reality, suggestivist models look impressive, look scientific. For example, very abstract models from the Neo-Classical ‘Chicago School’ of economists supported narratives about the optimal efficiency of markets, leading to a reluctance to regulate them (Krugman 2009). A lack of regulation seemed to be one of the factors behind the 2007/8 economic crash (Baily et al 2008). Modellers may understand that other modellers get over-enthusiastic and over-interpret their models, but others may not. It is the duty of modellers to give an accurate impression of the reliability of any modelling results and not to over-hype them.

How to recognise a suggestivist model

It can be hard to detangle how empirically vague a model is, because many descriptions about modelling work do not focus on making the mapping to what it represents precise. The reasons for this are various, for example: the modeller might be conflating reality and what is in the model in their minds, the researcher is new to modelling and has not really decided what the purpose of their model is, the modeller might be over-keen to establish the importance of their work and so is hyping the motivation and conclusions, they might simply not got around to thinking enough about the relationship between their model and what it might represent, or they might not have bothered to make the relationship explicit in their description. Whatever the reason the reader of any description of such work is often left with an archaeological problem: trying to unearth what the relationship might be, based on indirect clues only. The only way to know for certain is to take a case one knows about and try and apply the model to it, but this is a time consuming process and relies upon having a case with suitable data available. However, there are some indicators, albeit fallible ones, including the following.

  • A relatively simple model is interpreted as explaining a wide range of observed, complex phenomena
  • No data from an observed case study is compared to data from the model (often no data is brought in at all, merely abstract observations) – despite this, conclusions about some observed phenomena are made
  • The purpose of the model is not explicitly declared
  • The language of the paper seems to conflate talking about the model with what is being modelled
  • In the paper there are sudden abstraction ‘jumps’ between the motivation and the description of the model and back again to the interpretation of the results in terms of that motivation. The abstraction jumps involved are large and justified by some a priori theory or modelling precedents rather than evidence.

How to avoid suggestivist modelling

How to avoid the dangers of suggestivist modelling should be clear from the above discussion, but I will make them explicit here.

  • Be clear about the model purpose – that is does the model aim to achieve, which indicates how it should be judged by others (Edmonds et al 2019)
  • Do not make any conclusions about the real world if you have not related the model to any data
  • Do not make any policy conclusions – things that might affect other people’s lives – without at least some independent validation of the model outcomes
  • Document how a model relates (or should relate) to data, the nature of that data and maybe even the process whereby that data should be obtained (Achter et al 2019)
  • Be explicit as possible about what kinds of phenomena the model applies to – the limits of its scope
  • Keep the language about the model and what is being modelled distinct – for any statement it should be clear whether it is talking about the model or what it models (Edmonds 2020)
  • Highlight any bold assumptions in the specification of the model or describe what empirical foundation there is for them – be honest about these

Conclusion

Models can serve many different purposes (Epstein 2008). This is fine as long as the purpose of models are always made clear, and model results are not interpreted further than their established purpose allows. Research which gives the impression that analogical, illustrative or theoretical modelling can tell us anything reliable about observed complex phenomena is not only sloppy science, but can have a deleterious impact – giving an impression of progress whilst diverting attention from empirically reliable work. Like a bad investment: if it looks too good and too easy to be true, it probably isn’t.

Notes

[1] We often use the word “model” in a lazy way to indicate (1) rather than (1)+(2) in this definition, but a set of entities without any meaning or mapping to anything else is not a model, as it does not represent anything. For example, a random set of equations or program instructions does not make a model.

Acknowledgements

Bruce Edmonds is supported as part of the ESRC-funded, UK part of the “ToRealSim” project, grant number ES/S015159/1.

References

Achter, S., Borit, M., Chattoe-Brown, E., Palaretti, C. & Siebers, P.-O. (2019) Cherchez Le RAT: A Proposed Plan for Augmenting Rigour and Transparency of Data Use in ABM. Review of Artificial Societies and Social Simulation, 4th June 2019. https://rofasss.org/2019/06/04/rat/

Arnold, E. (2014). What’s wrong with social simulations?. The Monist, 97(3), 359-377. DOI:10.5840/monist201497323

Baily, M. N., Litan, R. E., & Johnson, M. S. (2008). The origins of the financial crisis. Fixing Finance Series – Paper 3, The Brookings Institution. https://www.brookings.edu/wp-content/uploads/2016/06/11_origins_crisis_baily_litan.pdf

Chattoe-Brown, E. (2018) What is the earliest example of a social science simulation (that is nonetheless arguably an ABM) and shows real and simulated data in the same figure or table? Review of Artificial Societies and Social Simulation, 11th June 2018. https://rofasss.org/2018/06/11/ecb/

Edmonds, B. (2001) The Use of Models – making MABS actually work. In. Moss, S. and Davidsson, P. (eds.), Multi Agent Based Simulation, Lecture Notes in Artificial Intelligence, 1979:15-32. http://cfpm.org/cpmrep74.html

Edmonds, B. (2020) Basic Modelling Hygiene – keep descriptions about models and what they model clearly distinct. Review of Artificial Societies and Social Simulation, 22nd May 2020. https://rofasss.org/2020/05/22/modelling-hygiene/

Edmonds, B., le Page, C., Bithell, M., Chattoe-Brown, E., Grimm, V., Meyer, R., Montañola-Sales, C., Ormerod, P., Root H. & Squazzoni. F. (2019) Different Modelling Purposes. Journal of Artificial Societies and Social Simulation, 22(3):6. http://jasss.soc.surrey.ac.uk/22/3/6.html.

Epstein, J. M. (2008). Why model?. Journal of artificial societies and social simulation, 11(4), 12. https://jasss.soc.surrey.ac.uk/11/4/12.html

Hartmann, S. (1997): Modelling and the Aims of Science. In: Weingartner, P. et al (ed.) : The Role of Pragmatics in Contemporary Philosophy: Contributions of the Austrian Ludwig Wittgenstein Society. Vol. 5. Wien und Kirchberg: Digi-Buch. pp. 380-385. https://epub.ub.uni-muenchen.de/25393/

Krugman, P. (2009) How Did Economists Get It So Wrong? New York Times, Sept. 2nd 2009. https://www.nytimes.com/2009/09/06/magazine/06Economic-t.html

Kuhn, T.S. (1962) The Structure of Scientific Revolutions. Chicago: University of Chicago Press.

Lakoff, G. (1987) Women, fire, and dangerous things. University of Chicago Press, Chicago.

Morgan, M. S., & Morrison, M. (1999). Models as mediators. Cambridge: Cambridge University Press.

Moss, S. (1998) Social Simulation Models and Reality: Three Approaches. Centre for Policy Modelling  Discussion Paper: CPM-98-35, http://cfpm.org/cpmrep35.html

Popper, K. (1957). The poverty of historicism. Routledge.

Vranckx, An. (1999) Science, Fiction & the Appeal of Complexity. In Aerts, Diederik, Serge Gutwirth, Sonja Smets, and Luk Van Langehove, (eds.) Science, Technology, and Social Change: The Orange Book of “Einstein Meets Magritte.” Brussels: Vrije Universiteit Brussel; Dordrecht: Kluwer., pp. 283–301.

Wartofsky, M. W. (1979). The model muddle: Proposals for an immodest realism. In Models (pp. 1-11). Springer, Dordrecht.

Zeigler, B. P. (1976). Theory of Modeling and Simulation. Wiley Interscience, New York.

Edmonds, B. (2022) The Poverty of Suggestivism – the dangers of "suggests that" modelling. Review of Artificial Societies and Social Simulation, 28th Feb 2022. https://rofasss.org/2022/02/28/poverty-suggestivism

© The authors under the Creative Commons’ Attribution-NoDerivs (CC BY-ND) Licence (v4.0)