28 May 2013 | doi: 10.3389/fpsyg.2013.00200
Global workspace dynamics: cortical "binding and propagation" enables conscious contents
Bernard J. Baars, Stan Franklin and Thomas Zoega Ramsoy
Regardless of whether there is a concise definition of the Dynamic Global Workspace or not, this paper has all of the concepts needed to build a satisfying mental model of how consciousness works. This is a collection of quotes from the paper that I think are very powerful.
There is a rather careless error. Figure 4. should be from Network architecture of the long-distance pathways in the macaque brain (2010) by Dharmendra S. Modha and Raghavendra Singh. (cheat sheet), but really the image is from: Circular representation of human cortical networks for subject and population-level connectomic visualization Andrei Irimia, Micah C. Chambers, Carinna M. Torgerson, and John D. Van Horn. The caption is captivating and describes both the Irimia and the Modha image.
Figure 4. The C-T complex supports any-to-any signaling. The cortico-thalamic system allows any spatiotopic activity (SA) array to signal any other. Combined with adaptive resonance, this allows an open set of cortical and thalamic coalitions to bind and broadcast information from any region to any other. The left half represents the left hemisphere of the brain, whereas the right half represents the right hemisphere. The brain stem is shown at the bottom. Circular color bars at the bottom describe the scale of the corresponding anatomical ring.
This paper is not for the lay reader. It jumps into the deep end and stays there. For example, the acronyms undefined. So, you need to read with a Wikipedia window open! Then you will know what TMS & PAG stand for, I hope.
TMS - http://en.wikipedia.org/wiki/Transcranial_magnetic_stimulation
And, perhaps you need to have read a couple of books on consciousness. But, if you love the concepts line "connective arborizations" and are struggling with consciousness and you are willing to cogitate, this is very, very rewarding! This is one of my favorite paragraphs:
Starting with LGN, all spatiotopic pathways become bidirectional. Successive arrays pick up visual features like spatial frequency, contrast, edge orientation, gestalt properties, hue, motion, and object identity. Higher level properties like object permanence, size constancy, color constancy, shape from shading, face and object recognition, scene analysis, movements, causality, and event organization, all require complex interactions among 40 or more spatiotopic arrays. The sight of a red traffic light must remain stable in spite of differences in reflectance, observer motion, background clutter, and changes in sunlight.I am not sure Spatiotopic is the best word here vs. Retinotopic or Visuotopic. I'm not sure it matters. Here is the basic idea Retinotopic Maps.
Entire visuotopic arrays can play inhibitory or excitatory roles, so that SA maps can take on all positive or all negative values. This is useful when the visual system needs to focus on a single level of analysis, like a single colored dot in a Seurat painting, rather than a gestalt of many colored dots.
Feeling of knowing are not imprecise in their underlying contents. They are simply subjectively vaguer than the sight of a coffee cup (Baars, 1988), lacking clear figure-ground contrast, differentiated details and sharp temporal boundaries. However, concepts, judgments, and semantic knowledge can be complex, precise, and accurate. The history of science and mathematics is filled with examples of accurate, but subjectively vague, insights that could not be articulated and tested until much later.
From the section: A Changing Stream of Bound Moments
The stream of consciousness is constantly changing, but momentary events are perceived as stable and consistent. The stream’s contents emerge from ongoing winner-take-all competitions between activity streams.
The current theory suggests that microstates represent binding and broadcasting equilibria involving dynamic coalitions of adaptively resonant populations of neurons.
In the brain, the event-related potential (ERP) occurs when a significant or intense stimulus is processed, causing a stereotypical wave pattern to sweep through the brain. The event ERP shows a series of negative and positive deviations over a period of ∼600 ms, corresponding to early sensory processing (N100), stimulus recognition, and attentional modulation (P200), decision making, and mismatch detection (P300 a and b), and stimulus meaning (N400 and longer). Revonsuo and colleagues have found that a conscious visual stimulus contributes a small negativity near the P3b wave (see Figure 8). ERP peaks and valleys are sensitive to numerous cognitive variables, but their overall shape tends to remain stable. ERP waves show both stimulus-triggered resetting and ongoing background activity.Logothetis et al.’s research program continues to yield spectacular results including the strongest direct evidence for global distribution of conscious visual stimulation to date (Panagiotaropoulos et al., 2012). More recent studies indicate involvement of the dorsolateral prefrontal cortex, which also has visuotopic maps. However, the bilateral removal of the prefrontal cortex does not abolish binocular rivalry, suggesting, again, that binding of visual input is not localized in a single region, but varies dynamically depending on stimulus, task, anticipatory set, emotional valence, memory load, distractors, and perhaps cortical resource allocation.
Since “binding” and “broadcasting” involve adaptive resonance, the distinctive type of signaling in the C-T core, dGW suggests “binding resonance” to define the winner-take-all gestalt that becomes conscious and “broadcasting resonance” to propagate the winning gestalt to receiving networks.
Dynamic Global Workspace theory implies a directional signal flow from binding to receiving coalitions. For each conscious event there is a dominant source and a set of receivers, where the propagated signal is interpreted, used to update local processes, and refreshed via reentrant signaling to the source (Edelman, 1989). Conscious sensations arise in a different center of binding and propagation than “feelings of knowing (FOK),” like the tip-of-the-tongue (TOT) experience, as demonstrated by brain imaging studies (Maril et al., 2001). Directional broadcasting of bound conscious contents is one testable distinction from other proposals (Edelman et al., 2011). Supportive evidence has been reported by Doesburg et al. (2009) and others.
The first GW architecture developed by Newell and coworkers was explicitly designed to resolve the multi-level ambiguities of auditory words spoken in a normal noisy space.
Adaptation to novelty has been proposed to be one of a small set of necessary conditions for conscious experience (Baars, 1988, Chapter 12).
Figure 9 shows evidence for stable microstates in the EEG for both the rabbit and humans, exhibiting rapidly changing phase after ∼100–200 ms, the rate of theta oscillations (Freeman, 2007). Other laboratories, using quite different methods, have also reported momentarily stable, content-sensitive microstates. The current theory suggests that microstates represent binding and broadcasting equilibria involving dynamic coalitions of adaptively resonant populations of neurons. This view seems quite compatible with sophisticated theoretical work by Freeman and Kozma. Crick and Koch (2005) suggested the term “coalitions” for this general concept.
Bullet Points from Baars2013 - http://journal.frontiersin.org/Journal/10.3389/fpsyg.2013.00200/full Further below you will find 5 lists of the key points of the paper. These are the: ================================================= 1) Headings (I) 2) Figures (F) 3) Table 2 - Summary of Hypotheses (H) 4) Table 3 Predictions ( P ) 5) Summary (S) P1. The cortico-thalamic system supports any-to-many binding and broadcasting of conscious contents. A bound conscious gestalt may emerge from anywhere in the cerebrum, and spread globally to all other regions for —100 ms. * Depends on CT complex - bidirectional connections * Depends on brain regions * Depends on long distance network I4, I5, I10, F1, F2, F4, F5, F6, H1, P1, S1, S2, S3, S4 H3) Dominant coalitions I8, I21 H9) adaptive resonance - The generality of adaptive resonance in C-T biocomputation I5, F2, P4, S5, F3) access - Conscious contents enable access to cognitive functions, including sense modalities, working memory, long term memories I7, I21) Microstates - Winner-Take-All Coalitions: Microstates F9, I22, -- streams? I23) Sensory Percepts vs. Feelings of Knowing, ?? - attention, voluntary effort. I27, I26) Conscious Events Evoke Widespread Adaptation or Updating - Novelty I29) The Hippocampus and Conscious Contents: A Novel Prediction F12, P8, I8) Spatiotopic Activity Maps, -- ??) phase coupling?) F6, H4, ==================================================
IntroductionI1) Dynamic GW Vis-À-Vis Other Theoretical Proposals I2) Dynamic GW as a Local-Global Theory I3) The Reportability of Conscious Events I4) Cortex and Thalamus I5) Bidirectional Pathways and Adaptive Resonance I6) Feature and Frame Binding I7) Consciousness Enables Many Kinds of Access I8) Spatiotopic Activity Maps, Streams, and Coalitions I9) Any-to-Any Signaling I10) Broadcasting: Any-to-Many Signaling I11) States and Contents I12) Waking and Deep Sleep I16) Global Chatting, Chanting, and Cheering I17) Task-Related Frequency Coupling I18) Conscious Contents I19) A Changing Stream of Bound Moments I20) Conscious and Unconscious Task Elements I21) Winner-Take-All Coalitions: Microstates I22) Conscious Moments are Embedded in Cycles I23) Sensory Percepts vs. Feelings of Knowing I24) Mental Effort I25) Voluntary Attention I26) Conscious Events Evoke Widespread Adaptation or Updating I27) Consciously Mediated Cognition I28) Voluntary Reports of Conscious Events I29) The Hippocampus and Conscious Contents: A Novel Prediction I30) Summary FIGURES ======= F1) Binding and broadcasting from many locations F2) Feature binding by adaptive resonance F3) Conscious contents enable access to cognitive functions, including sense modalities, working memory, long term memories F4) The C-T complex supports any-to-any signaling F5) Any-to-many signaling in conscious vision F6) Long-distance phase-locking in the waking state F7) Intracranial recordings in epileptic patients have 1,000 times the signal-to-noise ration of scalp recordings... F8) VAN: the visual awareness negativity wave F9) EEG microstates at theta rates in humans and rabbits F10) Perceptual experiences vs. feelings of knowing (FOKs) F11) Cortical adaptation as a novel task becomes automatic F12) Hippocampal-neocortical binding and broadcasting --------------- Table 2. A summary of hypotheses ============================= H1) Any-to-many binding and broadcasting in the C-T system H2) Perceptual consciousness vs. feelings of knowing H3) Dominant coalitions H4) Task-related chatting via phase coupling and decoupling H5) Feelings of effort interpreted H6) Voluntary control involves binding and broadcasting in the prefrontal lobe H7) Voluntary attention H8) The relationships to states of consciousness H9) The generality of adaptive resonance in C-T biocomputation Summary Numbering. Full Summary appears below. ========================== S1) A GW is a functional hub of signal binding and propagation... S2) Conscious experiences may reflect a GW function in the brain. . . S3) In humans the C-T complex underlies reportable conscious percepts, . . S4) Global workspace theory follows the historic distinction between the “focus” . . . S5) Cortico-thalamic core is a great mosaic of multi-layered two-dimensional neuronal arrays. S6) The C-T nexus appears to be the most parallel-interactive structure in the brain, . . . S7) Cortico-thalamic pathways run in all canonical directions and follow small-world organization, . . . S8) Global workspace dynamics interprets the traditional distinction between the “object” and “ground” . . . S9) For example, while sensory experiences are proposed to bind and broadcast from posterior cortex,. . . S10) Finally, dGW suggests the counter-intuitive idea that in intact humans, the MTL supports conscious episodes.. .
2. Receiving networks adapt to novel information from broadcast sources. After widespread receiver adaptation (updating), broadcasts are driven out of consciousness by competing inputs.
3. Posterior cortex generates perceptual conscious contents, while explicit feelings of knowing are generated from non-sensory cortex. (Frontal, anterior temporal, parietal). Many cognitive tasks involve both a conscious perceptual and a reportable semantic broadcast, as shown in the extended E RP
4. Since nearly all cortico-thalamic links are bidirectional, the cerebrum supports very widespread adaptive resonance (reentrant signaling). Signaling of conscious contents is superimposed on baseline resonant activity in the C-T core. Because of spatiotopic array organization in the cortex and thalamic nuclei, content signaling in the cerebrum is simultaneously spatial and temporal.
5. Goal-directed signaling in the C-T core is waking state dependent. Waking, dreaming and slow-wave sleep reflect distinct global modes. However, even slow-wave sleep may support waking-like activity during the UP phase of the slow oscillation.
6. While many spatiotemporal codes may exist, cross-frequency phase coupling is thought to integrate the full range of C-T rhythms. Because conscious sensory events are integrated within 100 ms periods, 4-12 Hz rhythms may underlie conscious moments.
7. Effortful voluntary control involves binding and broadcasting from frontoparietal regions. Mental effort is an FOK that is associated with major cognitive styles like persistence and general intelligence.
8. The hippocampal complex supports conscious event perception, as well as serving to encode episodic memory traces in multiple brain regions. Hippocampal lesions often lead to cortical reorganization of conscious sensory functions.
A GW is a functional hub of signal binding and propagation in a population of loosely coupled signaling agents. Neurons and neuronal cell assemblies can be defined as such agents when they respond selectively to input.
Conscious experiences may reflect a GW function in the brain. The brain has many anatomical hubs, but conscious percepts are unitary and internally consistent at any given moment. This suggests that a brain-based GW capacity cannot be limited to only one anatomical hub. Rather, it should be sought in a dynamic and coherent binding capacity – a functional hub – for neural signaling over multiple networks. A number of findings are consistent with the theory.
In humans the C-T complex underlies reportable conscious percepts, concepts, FOK, visual images and executive functions. While subcortical areas are often sometimes proposed to specify conscious contents, the human evidence is slight and disputed. Because cortex and thalamus are interleaved so densely as to constitute a single functional system, we will refer here to the C-T system as a whole. C-T pathways permit constant reentrant signaling, so that multiple spatiotopic maps can sustain or inhibit each other. The daily states of the core are controlled by basal brain nuclei.
Global workspace theory follows the historic distinction between the “focus” of experience vs. the largely implicit background of experience. Extensive evidence shows that visual and auditory consciousness flows from the respective sensory cortices to frontoparietal regions. This directionality differentiates GW dynamics from information integration theory and dynamic core theory.
Cortico-thalamic core is a great mosaic of multi-layered two-dimensional neuronal arrays. Each array of cell bodies and neurites projects to others in topographically systematic ways. Since all C-T pathways are bidirectional, signaling is “adaptively resonant” (reentrant). In this complex, layered two-dimensional arrays are systematically mirrored between cortex and thalamus, region by region.
The C-T nexus appears to be the most parallel-interactive structure in the brain, allowing for efficient signal routing from any neuronal array to any other. This connectivity is different from other structures that do not directly enable conscious contents, like the cerebellum. The cerebellum is organized in modular clusters that can run independently of each other, in true parallel fashion. But in the C-T core any layered array of cortical or thalamic tissue can interact with any other, more like the world-wide web than a server farm.
Cortico-thalamic pathways run in all canonical directions and follow small-world organization, so that each array is efficiently linked to many others. The entire system acts as an oscillatory medium, with markedly different global regimes in conscious and unconscious states.
Global workspace dynamics interprets the traditional distinction between the “object” and “ground” of experiences as a directional flow between the moment-to-moment focus of conscious experience vs. the implicit background and sequelae of focal contents. The proposed directionality of broadcasting suggests a testable distinction with information integration theory and dynamic core theory.
For example, while sensory experiences are proposed to bind and broadcast from posterior cortex, “fringe conscious” FOK plausibly emerge from non-sensory cortices, and may therefore broadcast rostrocaudally.
Finally, dGW suggests the counter-intuitive idea that in intact humans, the MTL supports conscious episodes. The classic lesion case of HM seems to contradict that idea, but traditional psychophysical testing does not show episodic organization. We therefore suggest that the positive evidence from brain recording methods for conscious experience in MTL may outweigh lesion evidence against it, given the well-established tendency toward cortical reorganization after injury.
2014.07.18 draft jch