Do Markov Blankets Give Us Real Neural Parts?

In my last post, I laid out what I think the rules are for developing a mechanistic model of the neural scale contribution to behaviour. I ended there with a question: what counts as a real neural part? How can we successfully decompose neural activity supporting a given perception-action loop into parts? 

In this post, I want to discuss one potential option: the hypothesis that Markov blankets, a key feature of the free-energy principle approach to neuroscience, can identify and pick out real neural parts. I'll discuss some recent ecological critiques of Markov blankets and some potential answers to the challenges.

Mechanistic Models of the Brain

I'm getting increasingly interested in neuroscience, and how to make it ecological. I also think that the ecological approach is capable of supporting mechanistic explanatory models of behaviour and is the correct scale* at which to ground these models. This means that my current plan is to find a way to add neuroscience as a lower scale part of a model grounded at the scale of the organism-environment system. 

There's a lot going on in that sentence, though, so I want to unpack it a bit to lay out the rules. and the things I currently don't know. 

(*NB I am using scale rather than level throughout because the concept of a level is complicated and currently, I am convinced that scale is a better term. The argument continues, however). 

Mechanisms for Cognitive and Behavioural Science (#MechanismWeek 5)

This week we have reviewed what a mechanism is, various ways to model mechanisms, and talked about the kinds of functional and dynamical models cognitive science currently relies on. We then rejected the argument that cognitive science cannot have causal mechanistic models that refer to real parts and processes. We claim that if we ground our models at the ecological level of perceptual information, a truly mechanistic analysis is possible, and we walked through a causal mechanistic model of a perception-action task as proof of concept for our claims. Sabrina then presented these ideas this week at a mechanism conference in Warsaw to an audience that included Bechtel and Craver (with great success, hurray! :)

The message we want you to go home with after #MechanismWeek is this:

Despite the fact that psychology has been trucking on very-nicely-thank-you developing various kinds of functional models, these remain extremely limited in their explanatory scope and they are not moving us towards explanatory mechanistic models. We have demonstrated that explanatory, causal mechanistic modelling of cognitive and behavioural systems is possible, so long as that analysis is grounded at the level of ecological information. These models are powerful scientific tools for exploring and understanding the behaviour of systems, and if we can get them, we should definitely be trying to. 

The research programme for getting mechanistic models is that laid out in Bechtel & Abrahamsen (2010) who used the development of mechanistic models of circadian rhythms as an exemplar for cognitive science. That programme involves spending time empirically decomposing and localising the real parts and processes of that actual mechanism. This requires going into the mechanism at a useful level of analysis; if you are struggling to find real parts and processes, you might be working at the wrong level. Only once you know the composition and organisation of the mechanism do you try to model it, typically using dynamical equations containing terms serving as representations of each component, placed in the appropriate relation to one another.

We have risen to their challenge by identifying the ecological level of analysis as the correct place to ground our work, and by identifying a cognitive science model that parallels the biological exemplar. It is our hope that this work will help others move in the direction of mechanistic research and modelling in the cognitive and behavioural sciences, so that we all gain the many benefits of causal mechanistic explanations.

This work will form the centre piece of a large scale paper we are currently writing. We've posted this part of the work on the blog in part to stake a claim to this analysis, but also to try and garner useful feedback from interested parties. If you have questions, comments or feedback, please contact us by commenting on any relevant post (we'll see it, even if it's on an older post), emailing us or finding us on Twitter. 

Thanks for reading along with us! We hope you enjoyed #MechanismWeek :)

Ecological Mechanisms and Models of Mechanisms (#MechanismWeek 4)

Mechanistic models are great, but so far cognitive science doesn't have any. We have functional models (of, for example, memory or categorisation) and dynamical models (of, for example, neural networks) but none of these can support the kind of explanations mechanistic models can. Is that it for psychology, or can we do better?

Here we propose that it's possible to do psychology in a way that allows for the development of explanatory, mechanistic models. The trick, as we have discussed, is to identify the correct level of analysis at which to ground those models. These models will definitely end up being multi-level (Craver, 2007), but the form of these final models will be dictated and constrained by the nature of the real parts and operations at the grounding level.

The correct level of analysis, we propose, is the ecological level. Specifically, ecological information is going to be the real component whose nature will place the necessary constraints on both our empirical investigations of psychological mechanisms as well as the mechanistic models we develop.

Let's see how this might work.

Cognitive Models Are Not Mechanistic Models (#MechanismWeek 2)

So far we have talked about what mechanisms are and what sort of model counts as properly mechanistic. The next question is to have a look in more detail at the models of cognitive science and see how far they can take us towards mechanistic explanations.

Last time I discussed the examples of research on memory, visual object recognition and categorisation. This kind of functional modelling work is the rule, not the exception in cognitive science - it's how we're taught to work and how the field moves along.

@PsychScientists A mechanism is a graph with at least three boxes and two arrows.

— Tim van der Zee (@Research_Tim) May 24, 2016

This kind of program feels like it's heading towards mechanism . Every division into new sub-capacities comes from work showing the two sub-capacities function differently and are therefore the result of different mechanisms. Every new representational model adds a new component (part or process) that handles another part of the capacity. There is one basic problem, however. None of these models make any explicit reference to any real parts or processes that have been empirically identified by other work - for example, 'working memory' still refers to a capacity, not a component. This means there is no reason to think this new capacity maps onto any particular parts and processes or if it does, to which parts and processes.

We are, in effect, doing science backwards: modelling first, running experiments later, and the result is that we are not actually on a trajectory towards mechanistic models, just better functional ones. This is a problem to the extent you want access to the many real benefits mechanistic models offer, in particular the ability to explain rather than simply describe a mechanism (see Bechtel & Abrahamsen, 2010 and the last post). This post reviews whether functional models explain or whether they can be part of a trajectory towards an explanation. The answer, unsurprisingly, will be no.