Consciousness: converging insights from connectionist modeling and neuroscience

https://doi.org/10.1016/j.tics.2005.06.016Get rights and content

Over the past decade, many findings in cognitive neuroscience have resulted in the view that selective attention, working memory and cognitive control involve competition between widely distributed representations. This competition is biased by top-down projections (notably from prefrontal cortex), which can selectively enhance some representations over others. This view has now been implemented in several connectionist models. In this review, we emphasize the relevance of these models to understanding consciousness. Interestingly, the models we review have striking similarities to others directly aimed at implementing ‘global workspace theory’. All of these models embody a fundamental principle that has been used in many connectionist models over the past twenty years: global constraint satisfaction.

Introduction

We believe that a view that allows the integration of selective attention, working memory, cognitive control and consciousness is within sight. This view is based on the notion of biased competition: competition between representations that are widely distributed in the brain, with top-down influences, most notably from prefrontal cortex 1, 2, 3. This mechanism is now implemented in several connectionist models that collectively address a wide range of findings 4, 5, 6, 7, 8, 9, 10. Many of these models have not been directly targeted at explaining consciousness, but they have important implications for our understanding of conscious information processing. In this article, we review these developments, highlight key computational principles embodied in these models, and explore their implications for consciousness. We also emphasize that these models can be understood in terms of a fundamental principle that can be traced back to early connectionist efforts: global constraint satisfaction [11].

It is useful to distinguish between states of consciousness (e.g. being awake, asleep, in a coma, etc.) and the contents of consciousness (e.g. being conscious of the scene one is looking at) [12] (although it should be noted that these are not independent [13]). The present article is about the contents of consciousness: we will not address what might be called the ‘enabling factors’ for consciousness (e.g. appropriate neuromodulation from the brainstem, etc.).

Box 1 highlights some of the computational mechanisms that we hypothesize are associated with the contents of consciousness. These are developed in the remainder of this review.

Section snippets

Consciousness and active representations

When talking about biological or artificial neural networks, it is important to distinguish between two types of representation. First, there is long-term knowledge that is embedded (or latent) in the weights of the connections between units. This knowledge can drive behavior indirectly by eliciting specific firing patterns over ensembles of units, but does not seem to be available for direct inspection by, or transmission to, another system. Although some have criticized this aspect of

Global workspace theory

Baars has suggested that consciousness depends on access to a ‘global workspace’ [23] and Dehaene and collaborators have developed several neurocomputational models that implement this theory (e.g. 24, 25). A crucial feature of these models is that they work by biased competition (Box 2). The basic assumption is that the winning coalition of neurons determines conscious experience at a given moment. Many theorists (e.g. Crick and Koch [26], Edelman [27]) have made similar proposals, and this is

Attention, working memory and cognitive control

The idea of large-scale competition biased by top-down projections from prefrontal cortex (PFC) has recently been proposed as the mechanism underlying attention [1], working memory [3], and cognitive control 2, 6, 31. Braver, Cohen, O'Reilly and colleagues have implemented these ideas in several connectionist models of working memory and cognitive control (see Box 3). There have also been several connectionist implementations of biased competition models of selective attention (see Box 4).

The

Reentrant connections and consciousness

We have seen that the PFC probably influences the contents of consciousness via top-down projections. More broadly, the importance of feedback (or reentrant) connections for conscious experience has been emphasized by several authors 25, 27, 37, and there is evidence from neuroscience to support these views (see, e.g. [37] for review). For example, transcranial magnetic stimulation of visual area MT induces motion perception, but not if activity in V1 is disrupted some time after the

Consciousness and global constraint satisfaction

The idea that recurrent interactions at a nearly global scale are important for consciousness 25, 27, 37, 47, 48 makes perfect sense in light of the idea that neural networks implement global constraint satisfaction – an idea that goes back to the early days of connectionism [49]. Many early connectionist models used global constraint satisfaction to address findings in several domains (e.g. 11, 50, 51). The idea is that a network with recurrent connections arrives at an interpretation of a

Synchronization, binding and global competition

There has been much debate about the role of synchronized neuronal firing to solve the so-called ‘binding problem’ [56] and determine the contents of consciousness (e.g. 26, 57, 58, 59). This literature is too extensive to cover in detail here, but we would like to emphasize that the ‘synchrony hypothesis’ is fundamentally compatible with our proposals, as competition need not be implemented exclusively in terms of firing rate: synchronization may also play an important role.

Consistent with

Access versus phenomenal consciousness

A decade ago, Block proposed a distinction between ‘access’ and ‘phenomenal’ consciousness [68]. In relation to the present framework, our working hypothesis is that the winning coalition determines both global accessibility and phenomenal experience, as a large-scale stable state seems an appropriate candidate for global access and would also explain the integrated character of phenomenal experience [45]. Block has recently suggested that losing coalitions might also determine the contents of

Conclusions

In summary, recent and current work in connectionist modeling and neuroscience is converging to provide an integrated view of attention, working memory, cognitive control and consciousness based on a single mechanism: global competition between representations, with top-down biases from PFC. This fosters an integrated understanding of these concepts in terms of the mechanisms and dynamics of global competition, rather than as reified processes with distinct neural instantiations.

Acknowledgements

This work was supported by a fellowship from the Calouste Gulbenkian Foundation (Portugal) to Tiago V. Maia and by an institutional grant from the Université Libre de Bruxelles (Belgium) and grant HPRN-CT-1999–00065 from the European Commission to Axel Cleeremans. Axel Cleeremans is a Senior Research Associate with the National Fund for Scientific Research (Belgium).

References (75)

  • N. Block

    Two neural correlates of consciousness

    Trends Cogn. Sci.

    (2005)
  • T.S. Braver et al.

    Dopamine, cognitive control, and schizophrenia: the gating model

    Prog. Brain Res.

    (1999)
  • R. Desimone et al.

    Neural mechanisms of selective visual attention

    Annu. Rev. Neurosci.

    (1995)
  • E.K. Miller et al.

    An integrative theory of prefrontal cortex function

    Annu. Rev. Neurosci.

    (2001)
  • R.C. O'Reilly

    A biologically based computational model of working memory

  • M.J. Frank

    Interactions between frontal cortex and basal ganglia in working memory: a computational model

    Cogn. Affect. Behav. Neurosci.

    (2001)
  • O'Reilly, R.C. (2003) Making Working Memory Work: A Computational Model of Learning in the Prefrontal Cortex and Basal...
  • T.S. Braver et al.

    On the control of control: The role of dopamine in regulating prefrontal function and working memory

  • G. Deco et al.

    Neurodynamics of biased competition and cooperation for attention: a model with spiking neurons

    J. Neurophysiol.

    (2005)
  • J.D. Cohen

    On the control of automatic processes: a parallel distributed processing account of the Stroop effect

    Psychol. Rev.

    (1990)
  • S. Corchs et al.

    Large-scale neural model for visual attention: Integration of experimental single-cell and fMRI data

    Cereb. Cortex

    (2002)
  • M.W. Spratling et al.

    A feedback model of visual attention

    J. Cogn. Neurosci.

    (2004)
  • D.E. Rumelhart

    Schemata and sequential thought processes in PDP models

  • S. Dehaene et al.

    Neural mechanisms for access to consciousness

  • S. Dehaene et al.

    Ongoing spontaneous activity controls access to consciousness: A neuronal model for inattentional blindness

    PLoS Biol.

    (2005)
  • Cleeremans, A. Computational correlates of consciousness. Prog. Brain Res. (in...
  • M. Velmans

    Is human information processing conscious?

    Behav. Brain Sci.

    (1991)
  • C. Koch

    The Quest for Consciousness: A Neurobiological Approach

    (2004)
  • Cleeremans, A. Being Virtual, Oxford University Press (in...
  • A. Cleeremans et al.

    Implicit learning and consciousness: A graded, dynamic perspective

  • B. Libet

    Production of threshold levels of conscious sensation by electrical stimulation of human somatosensory cortex

    J. Neurophysiol.

    (1964)
  • K. Moutoussis et al.

    The relationship between cortical activation and perception investigated with invisible stimuli

    Proc. Natl. Acad. Sci. U. S. A.

    (2002)
  • G. O'Brien et al.

    A connectionist theory of phenomenal experience

    Behav. Brain Sci.

    (1999)
  • B.J. Baars

    A Cognitive Theory of Consciousness

    (1988)
  • S. Dehaene

    A neuronal model of a global workspace in effortful cognitive tasks

    Proc. Natl. Acad. Sci. U. S. A.

    (1998)
  • S. Dehaene

    A neuronal network model linking subjective reports and objective physiological data during conscious perception

    Proc. Natl. Acad. Sci. U. S. A.

    (2003)
  • F. Crick et al.

    A framework for consciousness

    Nat. Neurosci.

    (2003)
  • Cited by (0)

    View full text