Commentary on "Functional MRI and the study of human consciousness" by Daniel Lloyd.

Abstract

Lloyd presents an exploratory investigation of the neural basis of consciousness that uses four data sets from the National fMRI Data Center. First, the author describes the fundamental aspects of consciousness as identified by analytical phenomenology—the philosophical subdiscipline that studies ‘‘the world as it appears.’’ Second, Lloyd uses these theoretical constructs to make empirical predictions about the functional imaging data and tests such predictions using multivariate techniques and artificial neural networks. This is an ambitious but entertaining and worthwhile contribution that differs from our original investigation in many ways. For instance, the research hypotheses that motivated the two analyses are very different. While in Mechelli, Friston, and Price (2000) the aim was to identify the neural substrates of specific cognitive functions (i.e., reading of words and pseudowords at different stimulus rates), Lloyd aims to characterize consciousness at a neuronal level in terms of a global and continuous process. Such divergent objectives influence the way the fMRI data are analyzed. In our original study, a standard cognitive subtraction approach that involved comparing neuronal activation for different experimental conditions (i.e., reading words vs. fixating a cross in the middle of the screen) was used. Fluctuations in signal unrelated to experimental manipulation were considered as noise or ‘‘nuisance variables,’’ and inferences were made on a voxelwise basis using univariate techniques. In contrast, Lloyd investigates whether neuronal activation is changing over time irrespective of experimental manipulation and tests whether the current neurophysiological state of the brain can predict or be predicted by subsequent or preceding states. Here fluctuations in signal unrelated to experimental manipulation provide potentially salient matrix for the continuous state inherent in consciousness. In addition, inferences are based on brain states considered globally using multivariate techniques and artificial neural networks. The application of artificial neural networks to functional imaging data represents one of the original aspects of Lloyd’s study. Although such application is relatively uncommon, the study by Lloyd illustrates that this approach may be very useful for characterizing changes in activity that cannot be predicted on the basis of experimental manipulation. Furthermore, the use of artificial neural networks may potentially be very useful for combining data from different methodologies that provide information on different levels of brain organization (i.e., single unit and system levels). The most original aspect of Lloyd’s study, however, is perhaps the exploitation of analytical phenomenology to make empirical predictions on the functional imaging data. This approach differs from previous neurophilosophical investigations that have mainly applied neuroscientific