Computers, Society, and Nature

In O’Sullivan’s description of simple abstract models, he mentions Epstein and Axtell’s use of the term ‘generative’. It is discussed initially as a “new approach to social science” whereby if a model “replicates observed regularities in the real world”(542) the researcher can claim to have “explained the phenomenon”(542). Although this is mentioned in the context of social science, it strikes me as being completely against the use of the word ‘explanation’ when it is applied to science. To me, a scientific explanation involves being able to not only list factors that create the phenomenon and quantifying them but also understanding the interactions between them. In this case, ABMs are being used as a tool rather than a science and involve more trial and error button pushing than anything else. If one sets the parameters of the model in many different ways for each simulation and one happens upon the right combination of parameters, one has managed to explain a phenomenon. Thus, in my opinion, it is necessary to ensure one fully understands all the factors influencing a system and how these factors interact with one another before one can say they have explained the system even when an ABM produces results matching the system as seen in reality.

-Outdoor Addict

O’Sullivan posits that the complex nature of ABMs violates “one of the most common tenets of practical science, the imperative to prefer simplicity over elaboration” (p. 546). As stated in the post “ABM and Toolmaking,” many issues arise due to this complexity. It is difficult to image, however, a simplified model of human processes; there are so many important variables to take into consideration. But in developing these models—especially when they are used not just to understand a phenomena but also to predict—what happens when the predictions are incorrect? Perhaps this is an issue that Bonabeau does not delve into enough: it is possible that the over-simplification of a system or the inability to consider enough variables can lead to error and uncertainty. In class we discussed the Turcot interchange and the designing of the freeway system in general. Computer models, which may have not taken into consideration enough transportation demand and urban growth variables, may have let to inappropriate policy and planning decisions.

Again, this may be a problem that can be solved through technological advancements. Revisiting the freeway example, we can now model how expanded roads quickly reach capacity. But maybe there is another issue at play, that of scale. How can a model that represents the actions of agents be simplified and, therefore, more accurate? Bonabeau uses the very small-scale example of the fire escape simulation to assert the benefits of ABM. In this example, while there are relatively few and homogenous actors who all have a single aim, the idea to construct a column in front of the exit would likely not have been arrived at without the aid of ABM. An alternative way of problem solving and perceiving a process, in other words, was enabled. Therefore, perhaps the effectiveness of ABM lies in fully understanding its limitations.

According to O’Sullivan, “while simple, abstract models can be useful for exploring the implications of theories under particular assumptions, they cannot establish the truth of those theories” (p. 546). So, it’s great that we can now determine the optimal location of a column, but the current nature of ABM will mean that fully understanding a complex social phenomena will be riddled with uncertainty. Keeping this important aspect in mind will arguably be key to the success and future development of ABM.

– jeremy

As illustrated by O’Sullivan, individuals in an ABM are governed by a set of rules in order to provide a natural representation of a phenomena. I am intrigued by the notion of having such a flexible system, which allows agents to exhibit behaviours and reactions that differ from their counterparts. This reflects real-life situations, as the outcome of an event, for example, can completely depend on the actions of a single individual, who is influenced by their setting and social context and vice-versa.

Our in-class exercise of modeling the users of McGill’s outdoor walkways, however, revealed the incredibly complex nature of agent-based modeling. In attempting to represent how students, tourists, cyclists and vehicles use space, we quickly discovered that there are a wide range of users who use the space very differently, and also respond to certain events in very different ways.

This complexity brings up issues surrounding data collection/storage/usage limitations, which have arguably rendered the popularity of ABMs to be low in the geographic community to date. As noted by sah, I think that this issue has not been given enough attention regarding access to modeling capabilities. More importantly, however, because AMB is still in its infancy, I think that the way in which ABMs simulate human systems reinforces the notion of GIS as tool-making, a process whereby representations are constantly being improved upon. While GIS may currently struggle to represent processes, technological advancements—as O’Sullivan briefly illustrates—will perhaps enable ABMs (and GIS) to better incorporate the human element through increased public participation, for instance.

– jeremy

The papers discussed bring up some very important issues about ABMs – how useful are they when models are too simple, how can we extract real causal relationships when they become more complex and start to mirror the real world. Problems of equifinality and how to evaluate ABMs are also important – how are we to verify and validate an ABM that we use for predictive purposes? Even though there are many problems concerning the use and interpretation of ABMs, I think it is still important to acknowledge that these are very cost-effective and quick forms of social experimentation that do not require large amounts of time, manpower and money to perform. There was perhaps one problem that we may see with models such as Schelling’s segregation model – the fact that his model had a final steady state. It seems that if an ABM is too simple, it is very possible to end up with a final steady state, such as with the case of segregation. This is hardly the case, depending on the temporal scale one is looking at, and the accuracy of the programming of agents. Most systems in the world tend to be in flux rather than unchanging – this is what makes the world complicated. Therefore, if conclusions of steady states arise from ABMs, it is perhaps better to use a more complicated model. There tend to always be some kind of exogenous factors that will affect the output of a real world system, and this is what makes ABMs so hard to work with. However, does that mean that Schelling’s model should take into account income levels, land values, rent values, available services etc.? I do not think so, as it would convolute the question and focus of the model away from ‘individual preferences for like individuals’. Having too many variables in a model just serves to blur away any possible causality. This is a problem with all sciences, but especially problematic for ABMs since they tend to deal in complexity rather than simplicity. Some previous comments here have noted the high demand ABMs have for computational power. This is increasingly becoming less of an issue, and becoming more of a data transfer, and data structure problem in my opinion. Actual processing power will not be the limit in the future, only our own data and programming structures with which we create ABMs.

Another very limiting factor to ABMs is calculating error and uncertainty. How should this be done, especially when used for ‘predicting’ or ‘forecasting’, and when we cannot truly model every single possible action of real-life agents? I think this is one of the problems of ABMs that holds it back from mainstream science or even GIS. Whereas in say, hyperspectral imaging, you can attribute your error to the sensor and other factors such as your calibration and correction, in ABMs it would be difficult to assign any sort of error value to conclusions, especially those that do not have a real-world comparison.

Finally, I would like to draw us to the question of: are there alternatives to ABMs? I believe the answer is no. Social experimentation involving large numbers of individuals is too difficult to control in the real world, and much more consuming in terms of time and money. A large-scale real life social experiment is just not as efficient. Additionally, ABMs have the important feature of being able to be re-run very easily – but real-life social experiments cannot just be ‘reset’, especially when the researcher doesn’t memory of the previous experiment to influence agent behaviour.

-Peck

I particularly liked how O’Sullivan’s introduced the various types of ABMs by separating them into three categories depending on how realistic they are because it reminds me that, although we can create incredibly complex ABMs that resemble reality very closely, the value in “simple abstract models as thoughts experiments” (542) should not be underestimated. Bearing in mind that “complicated models may remain just as baffling as the world they purport to represent” (546), perhaps for many research questions, extremely realistic stimulations are not necessary. Simple models that explore the interactions between only a few theories can no doubt shed new light on problems even if the stimulated scenarios are not observed in reality. Thus, building ABMs to stimulate thought experiments could prove to be a useful tool at the exploratory stages of research and theory building.

The concept of equifinality and model verification also got me thinking. In complex and flexible systems, isn’t it more common to be able to reach a certain outcome through different means than through only one means? I think learning to accept the fact that many models may be “valid” and evaluating model outcomes in terms of “… the trajectory by which those outcomes are reached” (546-547) must go hand-in-hand. For instance, in a game of chess, the same final outcome (e.g. checkmate) is reached through thousands of different sets of moves. Thus, when comparing two highly skilled chess players, it is much more convincing to evaluate how each player executes his moves than to see whether or not he/she can deliver a checkmate.

-Ally_Nash

Bonabeau’s article certainly gives us a good introduction to the potential of agent-based modeling and the wide-reaching social phenomenon that it is able to explore. The key contribution that ABM offers social science is the ability to study an issue from bottom-up by looking at interactions between agents rather than overall processes produced by agent. The concept of “growing” an explanation to social phenomenon is both catchy and intriguing.

I especially liked the last example in Bonabeau’s paper. It showed how an ABM is capable of incorporating the fact that each individual is situated in the social world and that we can most influenced by the people we know both in terms of our behaviors and our location (http://gizmodo.com/5879504/how-your-friends-locations-give-yours-away-online). In an age where social network sites allow us to communicate with more people than ever before, ABM can be powerful in studying the adoption of new ideas, dissemination of information, and power/influence.

However, there were two points which I thought deserved further explanation. First, when arguing for the flexibility exhibited in an ABM, the author points to its “… ability to change levels of description and aggregation: one can easily play with aggregate agents, sub-groups of agents, and single agents, with different levels of description coexisting in a given model” (7281). I think this is an attractive feature that warrants more description regarding how groups are created, whether a single agent keeps its description when aggregated with other agents, and can an agent move dynamically in/out of a group? Secondly, Bonabeau mentions the ability for ABM to capture “individual behaviors [that] exhibit memory… learning and adaptation” (7281), but fails to mention the type or the complexity of knowledge that can be learnt by agents (are logical inferences possible?). It would have been interesting to see an example of how this artificial intelligence plays out in an ABM and a brief discussion about the power and recent developments in the capability of this type of algorithms.

-Ally_Nash

Agent-based models are an interesting concept, with much potential, and as O’Sullivan describes, many limitations.  As exciting as they are, they have not gained the popularity in geography one may have expected, either.  Several reasons for this stood out to me amid the papers and class discussion.

First, agent-based models of increasing complexity are expensive, but if we want them to put out detailed results, they require this commitment of time and money to collect and input extensive amounts of data.  How many companies can afford the required commitment?

Second, as O’Sullivan mentioned, ABMs “frequently violate one of the most common tenets of practical science, the imperative to prefer simplicity over elaboration”.  This, combined with the fact that complex results can sometimes be just as hard to fathom as the complex data entered into the models, can be off-putting.

Third, and I would argue, most relevant, is the current availability of real-time data.  In class, we discussed how a lot of the results ABMs are attempting to discover can currently be monitored in real-time.  If we can have access to real people outputting results, as opposed to agents endowed with real qualities, why would we not choose the former?  Surely they can best model the complexities and counter-intuitive nature of real life?

All that being said, ABMs seem incredibly exciting, and an interesting way to model problems that we don’t have answers and data for already.  With continued improvements into the future, they seem like a technology to watch, maybe for emergence in unexpected ways.

-sah

O’Sullivan, David. “Geographical Information Science: Agent-Based Models.” Progress in Human Geography. 32.4 (2008): 541-550. Print.

I would like to respond and build on to sidewalk ballet’s post about the contextual limitations of abstract Agent Based Models (ABMs). While your critique may be levelled at abstract ABMs specifically, I believe that ABMs are capable of capturing the complex specificities of any local area. Coming at a computational cost, I would imagine users of ABM would typically tailor their models to accurately reflect their situation. Another thing to consider is that an ABM’s environment does not need to represent a physical location. ABM can be used with environments that are very common to many agents, such as time or virtual environments like the stock market. I think of it as a way to model simultaneous decision-making, game theorying. In these types of non-place specific environments, more generalized ABMs may still be appropriate.

An important reminder is that agents need not be individuals as well. Frankly, I find this a more challenging concept for ABM. O’Sullivan touches on this point as well; how do we represent agents that are not individuals and who are also mobile? Is it ever appropriate to assume an entire family is one agent when a census tracks movement of the population? Can we assume a pride of lions in the savannah is one agent since it is clearly led by the dominant male? Tackling these issues of generalization and how to represent their movements allows for reduced computation cost and a scaling up of these models.

O’Sullivan, David. (2008). “Geographical information science: agent-based models.” Progress in Human Geography, 32(4) 541-550.

– Madskiier_JWong

Multiple, differing agent-based models, being “simulation[s] that [use] agents to represent actors in the real world,” (O’Sullivan, 542), can produce results “that would match with empirical observations equally well” (546). Many basic, abstract models can reach the same conclusion, but how well can these be applied to specific occurrences in different locations?

In geography we know about the importance of locational specificity; the particular environmental, social, economic, political etc. factors that influence—and ultimately shape—the place. Abstract ABMs disregard any detail of real world situations, and therefore the emergent phenomena is incredibly abstract itself—making it impractical for it be applied to a wide variety of different places. Things occurring in one space cannot be blindly applied to a different one without acknowledging the different factors which comprise and inhabit the space. The substitutability of space is an inherent problem.

Abstract ABMs may not be seeking to generate results explaining phenomena for a particular location, but then what is their use in GIScience? Without incorporating any specific spatial entities, micro-level factors cannot be discerned and, ultimately, the phenomena cannot be explained. In the real world, the local has a great influence on agent behavior and interactions which is being overlooked in abstract ABMs. ABMs focus on the heterogeneity of individual agents, but basic models don’t consider the heterogeneity of different environments and how it influences agent behavior.

From a GIScience perspective, it seems that basic models don’t hold much weight in the explanation of phenomena for specific locations. Knowing that circumstances change in different areas, the behaviour depicted in abstract ABMs is incredibly superficial without model concerns for the particular.

O’Sullivan, David. (2008). “Geographical information science: agent-based models.” Progress in Human Geography, 32(4) 541-550.

-sidewalk ballet