A Theory of Consciousness

C_MagicThis post attempts to summarize: “Toward a computational theory of conscious processing” by Stanislas Dehaene, Lucie Charles, Jean-Remi King and Sebastien Marti that appeared in Current Opinion in Neurobiology 2014, 25:76–84. Even more than normally in my posts, I should note that if this were a research paper, everything should probably be in quotations or have a footnote.  None of the ideas are mine except for my mistakes. The paper is a review of the research done so far.  The post The Global Neuronal Workspace is based on Dehaene’s work and might also be of interest.  Connectome–How the Brain’s Wiring Makes Us Who We Are, by Sebastian Seung is extremely readable.

The paper begins with this quotation from Vladimir Nabokov: “Consciousness is the only real thing in the world and the greatest mystery of all.”  Bend Sinister (1947)

The content of consciousness refers to the specific information that I am aware of at a given moment. Dehaene states: “For instance, I am currently aware of reading these words, but not of the music playing in the background (until I attend to it).”   Conscious report is the process by which a conscious content can be described, verbally or by various gestures. Such reportability remains the main criterion for whether a piece of information is or is not conscious since I can report something if and only if I am aware of it. Many things can be consciously accessed, including perceptual states, abstract knowledge, memories, plans, and other internal states (e.g. feelings, confidence, and errors). Self-consciousness is a particular instance of conscious access where the conscious ‘spotlight’ is oriented toward internal states.

To clarify the nature of conscious processing, a first step consists in determining what it is not. Using masking, crowding, inattention or binocular rivalry, images can be presented under conditions so they remain strictly invisible. Behavioral priming and brain imaging can then reveal how deep these stimuli are processed.  Research indicates that subliminal processing can be quite deep. The brain non-consciously recognizes the abstract identity of pictures, words and faces, the quantity attached to a number symbol, the fact that two words are related or synonymous, the emotional meaning of a word, or the reward value of a coin or an arbitrary symbol. Beyond even this, in chess experts, a brief non-conscious flash of a chessboard suffices to determine whether the king is in check. Transitive inferences can also be deployed non-consciously: after non-conscious exposure to arbitrary word pairs such as ‘winter-red’ and ‘red-computer’, word association effects generalize to non-adjacent pairs (‘winter-computer’), a transitive link.  All in all, these findings refute the idea that non-conscious processing stops at an early perceptual level: meaning and value can clearly be assigned non-consciously. There is also considerable evidence that attention can be deployed and enhance processing even if its target remains non-conscious.

Recent findings also invalidate the idea that the central executive, which controls our strategies and inhibits unwanted behaviors, always operates consciously. Dehaene et al indicate that a series of experiments with the go/no-go paradigm show that an unseen visual cue can trigger inhibitory control circuits. Error detection and task switching, which are typical executive functions, can be triggered non-consciously. Overall, these findings support the view that virtually any cerebral processor may operate in a non-conscious mode.

What, if anything, remains unique to conscious processing? A subliminal stimulus may induce above-chance performance, behavioral priming, and a small amount of brain activity in narrowly defined brain networks, but these measures often increase dramatically as soon as the subject reports seeing the stimulus, especially in high-level areas. Accumulation of evidence (see post Evidence Accumulation Model) has been demonstrated with non-conscious stimuli, but only conscious stimuli cross the threshold beyond which an overt strategy can be flexibly deployed. Several theorists propose that conscious perception occurs when the stimulus allows the accumulation of sufficient sensory evidence to reach a threshold, at which point the brain ‘decides’ whether it has seen anything, and what it is. The mechanisms of conscious access would then be comparable to those of other decisions, involving an accumulation toward a threshold — with the difference that conscious perception would correspond to a global high-level ‘decision to engage’ many of the brain’s internal resources.  Recurrent NMDA (NMDA is the predominant molecular device for controlling synaptic plasticity and memory function.) connections impose slow accumulation dynamics and multi-stable ‘all-or-none’ behavior, whereby the incoming evidence either quickly dies out (corresponding to subliminal processing) or is accumulated and amplified non-linearly into a full-blown state of high-level activity. This global ‘ignition’ has been proposed as a marker of conscious perception.  Late ignition seems to provide a robust signature of conscious access. The contrast between an early linear variation in brain activity and a very late non-linear ignition has even been observed in 5, 12 and 15-month-old infants, leading to the tentative suggestion that infants too enjoy a conscious perception of visual stimuli.

Global Neuronal Workspace (GNW) theory proposes that conscious access stems from a cognitive architecture with an evolved function: the flexible sharing of information throughout the cortex. While non-conscious stimuli are processed in parallel by specialized cortical processors, conscious perception would be needed in order to flexibly route a selected stimulus
through a series of non-routine information processing stages. According to Dehaene, global information sharing and routing would rely on a set of interconnected high-level cortical regions forming a ‘global workspace.’

Only conscious primes allow for the development of subsequent serial strategies. The brain’s routing system is capacity-limited, and this feature may explain the frequent failure of conscious perception in a dual-task setting. Conscious processing of a first target T1 causes a bottleneck on the routing of a subsequent target T2, either by dramatically postponing its processing (a phenomenon known as the ‘psychological refractory period’) or by preventing its conscious perception altogether (‘attentional blink’).  Perception of T2 is not only delayed, but also temporarily unperceived, such that its subjective onset is displaced to the moment when T1 processing finishes. The minimal condition for creating these effects is that T1 is consciously perceived. These effects have been partially captured in simulations of  spiking neurons (Figure 1).

Experimentally, mathematical measures of the complexity and global integration of brain signals do provide solid markers of the state of consciousness, particularly when contrasting wakefulness with sleep or anesthesia. Intracranial recordings in humans undergoing propofol anesthesia indicate a dramatic and sudden fragmentation of neural activity, which remains locally organized but globally disintegrated, possibly because prefrontal cortices are invaded by an alpha-like rhythm.  Most importantly, integration and long-distance cortical communications provide signatures of residual consciousness that are clinically applicable to patients recovering from coma. These observations suggest that global cortical communication provides an excellent index of conscious processing.

Several candidate markers of conscious processing are now available that tentatively relate conscious processing, respectively, to global ignition, long distance broadcasting, and information integration. These ideas are not necessarily incompatible. Dehaene concludes that, on the contrary, considerable convergence exists to suggest that firstly, conscious access triggers an all-or-none change in the state of distributed cortical networks; secondly, conscious processing corresponds to a massive cortico-cortical exchange of information, allowing flexible routing and therefore the slow serial performance of novel and arbitrary tasks; and finally, the state of consciousness, that is the brain’s very ability to host a ceaseless stream of such all-or-none conscious episodes, rests upon the integrity of long-distance cortico-cortical exchanges.