Assumptions in the scientific process
As in the affordance essay, we start by examining the assumptions explicitly and implicitly held when we go about doing science. In particular, we focus on the assumptions held with regard to causation. Because of the action-at-a-distance problem, science has predominantly operated under the rule that for a certain explanation to plausibly account for a cause-and-effect relationship, it must be of the mechanical, efficient kind. This requires causation to be local, where the variable causing makes direct contact with the variable caused. With the widely-used example of an-arm-moving-a cue-stick-hitting-a-cue-ball-hitting-an-eight-ball to illustrate efficient cause, explanations naturally appealed too to linear causation, where causes influence intermediary factors that lead to the observed phenomena in an unbranching, unbroken chain.
These initially theoretical assumptions on causation have gone on to influence the statistical analyses and methodological designs we use in order to study human behaviour. Linear causation assumes that changes in the alleged cause lead to proportional changes in the observed effect. This has led use statistical tools that reflect this assumption, like the generalised linear models underlying the regressions and ANOVAs so common in psychological research. Besides this, assumptions that causal factors are additive and independent of each other have led to methodological designs which manipulate these factors separately, as well as statistics that analyse them in isolation.
Is this a fair description of the world?
This might be all well and good, except for the fact that the world does not work like this! While this might be the case for man-made machinery, it certainly isn’t for the biological systems we study. For example, there is often no proportionate, linear relationship between cause and effect. Big changes in the manipulated variables might see small behavioural effects and plateaus, while tiny changes in these variables can also precede a disproportionately large change in behaviour. One way that linear methods have tried to account for this is by using polynomial and logarithmic curves to better fit the data. However, Heino et al. (2021) argue that this is still an inadequate move, as this does not account for other common features in complex, nonlinear systems, such as irregular changes, periodic peaks and plateaus, and recoveries after negative shocks.
Furthermore, the complex nature of biological systems makes it difficult to believe that any one faculty or component within these systems operates independently from others. This calls into question the practice of studying causal variables separately before combining them in a simple and additive fashion later on. Wallot & Kelty-Stephen (2018) call this component-dominant dynamics, where an observed trait can be attributed to some mental construct in the brain that is time-invariant and that applies to all situations. According to this view, there are hard-assembled faculties in the brain that correspond to functional activities like memory, perception, and motivation, all of which operate independently of one another and that are added upon each other linearly to give rise to a certain behaviour. On the contrary, the authors advocate what they call interaction-dominant dynamics, which I take to be a description of the features of dynamical systems. Here, lower-order components are interdependent upon each other, and operate using simple local rules that give rise to a higher-order behaviour through self-organisation. This emergence of behaviour means that causality cannot be attributed to a single sufficient cause, but is rather a result of many different parts and processes coming together in a synergistic manner.
The main point here is, the way we do science is very much influenced by both the assumptions we make about the reality of the world and those we make by virtue of using a particular methodological design or statistical analyses. Far from being objective, the methods we choose to explore human behaviour constrain the types of results we obtain, and the types of explanations we derive. If we want to actually account for the complex nature of reality, scientists should explore alternative methods that give us a way to quantify and study these complexities. Given that it is difficult to predict the global behaviour of a system just by knowing the components involved, there is a need to understand the interactions and processes involved, as well as how they change over time or in different contexts. The need to identify both the composition and organisation of a system, and also the way in which the system evolves over time, is where dynamical systems theory comes in.
Dynamical Systems
Roughly, a dynamical system is one where the state of the system at one time point is dependent on the past states of that same system. Right off the bat, this introduces the notion of time-dependent processes in understanding human behaviour from a dynamical systems perspective. It also highlights a crucial limitation of current methods that take pre- and post-test measurements before finding averages among large groups of participants. These methods take snapshots of behaviour at a limited number of time points, and are only informative under the assumption that behaviour is static and time-independent. Another limitation of these methods is the assumption that group-level outcomes are informative of intra-individual processes, known as ergodicity. Instead of measuring large numbers of people at fewer time points, Heino et al. (2021) argue that the underlying interaction-dominant dynamics are better elucidated by assuming non-ergodicity and measuring smaller groups but with substantially more measurement time points. This way, we move from group-level averages to individual-level time series data that allows us to see the trajectory of behaviour.
One common way of visualising and describing behaviour in dynamical systems theory is by using the notions of attractors and attractor layouts. The attractor layout is most easily imagined as a 2D sheet filled with peaks and valleys. If you imagine behaviour as a ball in contact with this sheet, the attractor layout then represents all possible behavioural states at that time point, in that context. The possible movements of the ball across the sheet then represent all the different behavioural trajectories possible, that is, the different possible evolutions of behaviour with respect to time. The valleys are called attractors, and represent stable regions on the sheet that the ball tends to fall towards. In other words, these are the behaviour states a biological system tends towards as time goes by.
Other terms include order parameters, which are macroscopic behavioural variables that can be multiply realised from lower-order variables, and that also serve as parsimonious ways to distinguish different behavioural states from each other. Control parameters, on the other hand, are variables which change the shape of the attractor layout (which I take to represent the order parameter), which can lead to the emergence, death, or movement of different attractors. Finally, a state space is the whole region in which a behaviour lives, and is defined by some kind of variable. What kind, I’m not entirely clear. At the moment, it feels like the state space refers to the region containing all the possible behavioural states a perception-action system can take on. If so, how is this different from the attractor layout described above? And if the attractor layout represents the order parameter, is the state space simply a reflection of the order parameter? Intuitively, something doesn’t sound right here, and I’ve clearly missed something. Help is very much appreciated, but I have plans to read more in these areas with these questions in mind.
A small (but nontrivial) note on dynamical systems
So far, it seems like using dynamical systems to accurately model and describe the behavioural phenomenon of biological and social systems has been a success. It accounts for nonlinearity, the time- and history-dependent evolution of systems, and treats noise as a richly nuanced source of flexibility instead of noise. However, it is NOT a theory of behaviour (Wilson, 2022). It is better characterised as a mathematical tool to quantify and describe complex behaviours, but it does not necessarily explain the emergence of such attractor layouts and behaviours. To do this, one needs a theory of behaviour (hint: ecological psychology (Golanka & Wilson, 2012)) that uses dynamics to serve its scientific exploration. This might actually be hinted from how so many disciplines (and also many subdisciplines within psychology) while having different theoretical motivations, have come to commonly embrace the use of the methodological tool that is dynamical systems.
Gaps and future steps
My knowledge of dynamical systems still has much to be desired. I am definitely not as confident in it as ecological psychology theory, and this is probably a mixture of my mathematical background and general avoidance of material that includes anything dynamical systems related. My next steps are pretty clear. Other than generally reading within the complexity literature, I intend to (a) continue making progress in my college math endeavours, (b) start this book on complexity by Melanie Mitchell, (c) and actually try to read some of the CESPA dynamical systems stuff. I am also still coming to terms with the concepts of nonlinearity and the role of noise in complex systems, and have a bunch of papers that I’ll dig into during the coming weeks.
References
Golonka, S., & Wilson, A. D. (2012). Gibson’s ecological approach – a model for the benefits of a theory driven psychology. DOAJ (DOAJ: Directory of Open Access Journals). https://doaj.org/article/bb2e3f79777544d7bbed6e2cc8edfe24
Heino, M. T. J., Knittle, K., Noone, C., Hasselman, F., & Hankonen, N. (2021). Studying Behaviour Change Mechanisms under Complexity. Behavioral Sciences, 11(5), 77. https://doi.org/10.3390/bs11050077
Wallot, S., & Kelty-Stephen, D. G. (2017). Interaction-Dominant causation in mind and brain, and its implication for questions of generalization and replication. Minds and Machines, 28(2), 353–374. https://doi.org/10.1007/s11023-017-9455-0
Wilson, A. D. (2022). Ecological Mechanistic research and modelling. Ecological Psychology, 34(1–2), 48–70. https://doi.org/10.1080/10407413.2022.2050912
No comments:
Post a Comment