Integrated information, causal density and consciousness - By Adam Barret
09/03/2010 - 15:30
09/03/2010 - 17:00
Adam B. Barrett
For much of the twentieth century, research into consciousness was exclusively carried out as philosophical discourse. However, due to modern advances in neuroimaging, coupled with new multi-disciplinary approaches to neuroscience and psychology, a new science of consciousness has started to emerge. In my introduction, I discuss some of the aspects of consciousness that science can hope to explain. In particular I talk about how consciousness might possibly be measured. Regarding neural correlates of consciousness, I talk of the importance of uncovering "explanatory" correlates that actually account for certain phenomenal properties of consciousness.
I begin the technical part of the tutorial with a brief introduction to information theory. I then give a review of Tononi's integrated information theory of consciousness, which derives from the property that conscious experiences provide integrated representations of very large amounts of information. The theory posits that the quantity of consciousness present in a system correlates with the extent to which the system as a whole generates more information than the sum of its parts. I describe two new measures of integrated information, which, unlike previous formulations, can be applied to real neural systems.
In the second part, I give a review of Granger causality and describe how it can be used to map functional connectivity (as distinct from anatomical connectivity) in the brain. I then explain why the overall density of functional connections (i.e. causal density) might correlate with conscious level.
Finally, we will compare causal density and integrated information and discuss their relative merits and shortcomings as measures of consciousness.
As part of the tutorial I will demonstrate how to use Anil Seth’s Granger causality toolbox, as well as a new toolbox for computing integrated information. We will use these toolboxes to simulate data, as well as to compute causal density and integrated information for model neural systems, and hopefully also take a look at applying them to some real neuroimaging data.
Technical requirements: I will try to introduce all the necessary mathematics within the tutorial, although prior knowledge of information theory, linear algebra and probability theory would be useful.
1.Barrett, A.B. and Seth, A.K. (2010) “Practical measures of integrated information for