Blogging* has been productive. Not only has it helped me collect my thoughts, but I’ve had some useful pointers from Ken and Richard. Today Richard suggested I look at Sukaina Bharwani et al fantastic paper “multi-agent modelling of climate outlooks and food security on a community garden scheme in Limpopo”. Beside making a compelling case for MABS/ABM as a tool/method for finding pathways out of poverty, it also solved a problem for me! The authors very cleverly decided to focus on just three crops: Maize (M), cabbage(C) and butternut squash (B), and two farmer agents: poor and better-off (based on 5-6 years of field data). M, C and B have quite different properties when it comes to their properties in the field and in the market. M is easy/cheap to grow but does not fetch a high price in a market. C is fairly easy/cheap to grow but its market price is high. B is quite difficult/expensive to grow but will reliably fetch a high price at the market (especially around Xmas time). The data shows that poor and better-off farmers have quite different strategies, but against the backdrop of increased climate variability (mostly in respect to rainfall) the two groups are likely to fare quite differently. This allows the authors to focus on the central question of their research: to what extent might seasonal [weather] forecasts ensure adaptation to longer-term climate change, linking learning at the seasonal scale with the evolution of climate on longer time-scales. In other words, how can vulnerable farmers adapt to the slow but relentless change in climate based on seasonal weather predictions. I won’t write the spoiler here (see the results section for some worrying graphics) but the paper makes a good case for how ABMs can be used to help whole communities (rather than just a minority of richer individuals) develop strategies that are more robust than their traditional response might be.
The paper has helped me imagine a more abstract and satisfying model. There will be less focus on gathering minutiae on farming practices e.g. “prepare-soil” in place of “weeding”, “tilling”, “digging furrows”. The real choice we want to explore is how much of a particular crop the farmer wants to grow. I can do this by giving each player one hectare per farming unit (I don’t know if a hectare would be farmed by an individual, family, or group of families…I imagine the latter) and asking how much space (which I will assume correlates roughly with effort) to dedicate to coffee, maize and manioc. I will then introduce challenges e.g. what will godel players do if wind flattens a maize crop just before harvest, or cows trample manioc, or one of numerous insects destroy a coffee crop.
Whilst thinking about this another benefit of out use of modelling4all came to mind. Not only will players be able to interact with the models by changing sliders, choosers and the other NL user interface elements, they will also be able to change micro-behaviours. This presents the intriguing possibility of creating agents that farm like real agents (as configured by the farmers). I could put these prototypical agents together into one model, and see how things play out. This will be more interesting if there is some kind of interaction between model agents e.g. a willingness to farm in groups, or if market prices are effected by over-abundance of a crop. The main point being, we could actually create a truly heterogeneous ABM, rather than a lot of agents that follow the same simple rules, but evolve differently because of the way they interact with the world and each other over time. The farmers might describe very different rules and so create very different kinds of farming agent.
In summary, my aim should be to support greater reflexivity – enabling farmers to build and explore farming models and discuss mitigation strategies. This is quite different from building models that predict the likely outcomes of different strategies. Ideally we’d have the latter, but as the paper by Sukaina et al puts it very nicely:
It should be acknowledged that many, if not most, complex socio-ecological systems will remain unpredictable even if an understanding of the influences of behaviour within the system is achieved. Social science is less concerned with prediction than with identifying how behaviour evolves and influences other processes. Since, the study of complex systems is an attempt to better understand systems which are difficult to grasp analytically, often the best available way to investigate such systems is through simulation (Gilbert & Troitzsch 1999). Purely deterministic prediction is difficult in matters of human thought and its relation
to action; stochastic prediction is likely to remain the norm. This is, in part, because ‘rules’ are not rules of action, but indicators of possible courses of action and are influenced by both the goals of the agent and the reliability of the agent’s categorization of the context, which relates to the different rules. MABS can explore social and environmental scenarios that do not exist at present, providing an experimental laboratory on the same level of sophistication as models of the global climate system.
(Multi-agent-based simulations (MABS) are roughly synonymous with ABMs).
*(talking aloud in the hope that someone out ‘there’ might post something more useful than an advert/something totally bizarre)