Causality and its maths, is it too general to be useful ?

[rogerharris]

I have been asked to write a couple of papers with a mathematician who has made extensive use of casual theory in his career and done pretty well. All i will be providing is neuroscience knowhow but obviously as a co-author i need to accept some responsibility for his product.

Without getting into specific examples (although i can provide some if needed) here I am sure many here will be familiar with the concept of causality and how its applied in many fields.

It seems to me that causality is kind of a general circular way of "perceiving" events, their actions and relations such that anything can be stated to be causal..which is why it is popular. It also seems that its so general that we can take any system and assign it into casual sets.. then do as we like. So many areas of math and physics have been coded into such sets and manipulated correctly

But is causality a profound theoretical framework it proposes to be ? it seems to me very loose in the sense of having no specific means to falsify the validity of its use. We can use it sure, but it doesn't actually tell me anything specific.

 

[rogerharris]

actually my OP is a bit off.. I should say is it too general to have serious teeth for specific areas or the like. i.e. Its too general for a serious theory to be proposed as an aspect of causality. I guess that is the problem then.. I hear professors say "well this aspect of the natural world is a manifestation of causal theory"... but so can everything be..and the sets are just a framework we use..

 

[HallsofIvy]

What do you mean by "causal theory"?

 

[rogerharris]

What do you mean by "causal theory"?

as in

https://en.wikipedia.org/wiki/Causality_(physics )

and applied through

https://en.wikipedia.org/wiki/Causal_sets

https://en.wikipedia.org/wiki/Causal_structure

A typical application used for brain imaging but the system is being applied into almost every primary domain of natural sciences from the references and papers I am being sent.

http://en.wikipedia.org/wiki/Dynamic_causal_modelling

 

[verty]

From what you posted above, it is a theory of the structure of causal relationships, a sort of mereology of causes. I don't see how this is a system, or how it can be applied at all.

Can you give a concrete example? Before and after the application of the (so-called) system.

 

[rogerharris]

From what you posted above, it is a theory of the structure of causal relationships, a sort of mereology of causes. I don't see how this is a system, or how it can be applied at all.

Can you give a concrete example? Before and after the application of the (so-called) system.

this is the first neuroscience paper to apply.. since then fristons lab have produced dozens of applied works..the link to his lab is in the wiki entry for dynamic causal modelling.

http://www.fil.ion.ucl.ac.uk/~wpenny/publications/dcm.pdf

I have seen it successfully applied on other problems.. but i also am lost as to how its a "system" also. From what i can make out its an easier means to assign complex data into sets than we had previously. So for the complexity scientist its the equivalent of a spreadsheet, but they still have to program the functions for the data.

 

[verty]

I quote from that DCM paper:

Identification proceeds in a Bayesian framework given known, deterministic inputs and the observed responses of the system. We developed [our] approach for the analysis of effective connectivity using experimentally designed inputs and fMRI responses.

I wondered what effective connectivity is. I found this passage in another paper by Karl J. Friston, et al:

The term effective connectivity has been defined by various authors in convergent ways. A general definition is that effective connectivity describes the causal influences that neural units exert over [one] another. More specifically, ...

It seems this term effective connectivity is used only by this author and his peer group. And does he perhaps mean in divergent ways? And having given a general definition, he follows it with "more specifically" instead of just "specifically", though he has not yet been specific. Perhaps it is because English is not his native language, but he is in fact a born Englishman and member of the Royal Society. I wondered how he could publish papers with such obtuse writing. It seems he is editor-in-chief of the journal NeuroImage, the journal that this paper is from. So that explains that.

So we have extremely obtuse writing with terms that are difficult to pin down. Here is another example:

Functional integration in neuronal systems can be quantified in two ways, functional connectivity and effective connectivity. While functional connectivity only describes statistical dependencies between spatially segregated neuronal events, effective connectivity rests on a mechanistic model of how the data were caused.

I looked up functional integration:

In neurobiology, functional integration is the hypothesis that the integration within and among specialized areas of the brain is mediated by effective connectivity.

I can attribute no meaning to these circularly-defined terms. And what makes this modelling dynamic? I quote:

Dynamic causal modelling represents a fundamental departure from existing approaches to effective connectivity because it employs a more plausible generative model of measured brain responses that embraces their nonlinear and dynamic nature.

The implication is that earlier approaches that this approach supercedes did not embrace the dynamic and nonlinear nature of neural responses. Um, this is just utter rubbish.

But getting past the surface features, there is a more telling reason not to want anything to do with this researcher or his research which I will now describe.

Unlike previous approaches to neuroimaging, the causal model ascribes responses to designed deterministic inputs, as opposed to treating inputs as unknown and stochastic.

So rather than looking at the data with no preconceptions, we build into the model what our causes are. Deterministic means, what determines the effects is built into the model. And the model has parameters that are tuned by applying it to the time series (the data) in some probabilistic way.

We can think of it like this. Suppose I notice that when I work in the garden, it sometimes rains and I get wet. So I want to do an experiment. I will work in the garden every second weekend for a whole year, 26 trials. Each time, I will record the chance of rain as predicted by the weatherman as well as whether I got wet or not. I will use this data to tune the causal model with the input, the weatherman caused me to get wet by predicting it would rain. This is my deterministic input to the causal model, that the weatherman can cause me to get wet.

So I apply it to the data and there is a fairly good correlation because most of the time, if rain was predicted, it did rain and I got wet. And most of the time that no rain was predicted, I stayed dry. My causal model may tell me, it is 80% likely that the weatherman caused me to get wet.

It doesn't work. It isn't science. One can't plug in what the causes are and measure them. What can I say, it's bunk.

 

[rogerharris]

good review, or at least i will need to check if this is what the authors intended.. but overall the way i see casual logic applied makes me wonder if it is circular. Everytime I have a question about what the framework can do.. the answer is always "take this, put it into causal set format then derive your algorithms and go". it seems to be a formatting procedure, which is ok in principle.. but if its leading to circular process that is way more serious.. i need to go check though as its hard to believe all these scientists using casual frameworks could do something so self confirming.