Elsevier

Consciousness and Cognition

Volume 47, January 2017, Pages 99-112
Consciousness and Cognition

Attention in the predictive mind

https://doi.org/10.1016/j.concog.2016.06.011Get rights and content

Highlights

  • Gives a non-technical exposition of the prediction-error coding theory of attention.

  • Explains the motivation for using precision-expectation to explain attention.

  • Argues that a precision-expectation-based theory of attention cannot be complete.

Abstract

It has recently become popular to suggest that cognition can be explained as a process of Bayesian prediction error minimization. Some advocates of this view propose that attention should be understood as the optimization of expected precisions in the prediction-error signal (Clark, 2013, 2016; Feldman & Friston, 2010; Hohwy, 2012, 2013). This proposal successfully accounts for several attention-related phenomena. We claim that it cannot account for all of them, since there are certain forms of voluntary attention that it cannot accommodate. We therefore suggest that, although the theory of Bayesian prediction error minimization introduces some powerful tools for the explanation of mental phenomena, its advocates have been wrong to claim that Bayesian prediction error minimization is ‘all the brain ever does’.

Section snippets

Prediction-error coding

An enormous amount of research has been devoted, over the last sixty years, to the discovery of efficient methods for gathering and disseminating information. The concepts that have emerged from this research have often been applied in theories of the mind. One of them enables us to draw a distinction between two quite different strategies for information gathering.

To see that distinction, suppose that I want to gather information about x from those of my sources that are in touch with x’s

Surprisal and selection

We have said that systems employing prediction error coding owe some of their efficiency to the fact that the only information to be propagated up through such systems is ‘surprising’, in the sense that it is information about those aspects of the incoming signal that have not already been predicted by the system’s prior hypotheses (see for example Barto et al., 2013, Friston, 2009). It is this that prevents resources from being taken up with the transmission of redundant information.

Some

Attention and precision

About half of the visual information that was presented in Neisser and Becklen’s experiment was consistent with the hypothesis that the filmed scene was one in which a hand game is being played. About half was consistent with the hypothesis that the scene was one in which a ball game is being played. Since these hypotheses are quite different, the information that is predicted by either one will tend not to be predicted by the other. The central claim of the prediction error theory is that the

The variety of attention shifts

When, following Friston, Hohwy writes that “attention is nothing but precision optimization in hierarchical inference”, his claim is explicitly a ‘nothing but’ claim. If true, it should apply with full generality. It should therefore apply to attention in all its forms. Psychologists have traditionally drawn distinctions between these forms. They have distinguished between feature-based and object-based attention; between endogenous and exogenous attention; and (what is not quite the same

Active inference

The argument above does not yet take account of an important complication, coming from Hohwy’s theory of action. That theory increases the complexity of the problem that we have just identified, without – we contend – doing anything to address the ultimate source of it.

In order for actions to be subsumed within the prediction-error theory they must be explained as attempts to minimize the prediction errors that are generated by our top-down hypotheses. The minimization of such errors is, Hohwy

Conclusions

Hohwy tell us that his “account of attention is interesting because it reduces attention to a simple matter of learning regularities of precision in nature” (p. 205). That account fails to apply to cases in which it is not regularities in nature that determine the way in which attention is allocated. There are instances of voluntary attention where the gain cannot be set by expectations of precision (whether or not these expectations are ‘learnt from nature’). When the relevant expectation is a

Acknowledgments

Madeleine Ransom and Sina Fazelpour are both supported by the Joseph Armand Bombardier Canada Graduate Scholarship Program. Christopher Mole’s research is funded by the Social Sciences and Humanities Research Council of Canada, Grant No.: 435-2015-0321.

The authors would like to thank Andy Clark, Adrian Downey, Dustin Stokes, Jona Vance, and the referees for this journal.

References (62)

  • C.M. Bishop

    Pattern recognition and machine learning

    (2006)
  • T.G.R. Bower

    The object in the world of the infant

    Scientific American

    (1971)
  • D. Broadbent

    Perception and communication

    (1958)
  • H. Brown et al.

    Active inference, sensory attenuation and illusions

    Cognitive Processing

    (2013)
  • B.P. Carlin et al.

    Empirical Bayes: Past, present and future

    Journal of the American Statistical Association

    (2000)
  • A. Clark

    Whatever next? Predictive brains, situated agents, and the future of cognitive science

    Behavioral and Brain Sciences

    (2013)
  • A. Clark

    Surfing uncertainty: Prediction, action, and the embodied mind

    (2016)
  • M. Colombo

    Social motivation in computational neuroscience. Or if brains are prediction machines, then the Humean theory of motivation is false

  • M. Colombo et al.

    Explanatory pluralism: An unrewarding prediction error for free energy theorists

  • P. Dayan et al.

    The Helmholtz machine

    Neural Computation

    (1995)
  • N. Eilan

    Perceptual intentionality, attention and consciousness

  • H. Feldman et al.

    Attention, uncertainty, and free-energy

    Frontiers in Human Neuroscience

    (2010)
  • K. Friston

    A theory of cortical responses

    Philosophical Transactions of the Royal Society B: Biological Sciences

    (2005)
  • K. Friston

    Hierarchical models in the brain

    PLoS Computational Biology

    (2008)
  • K. Friston

    The free-energy principle: A unified brain theory?

    Nature Reviews Neuroscience

    (2010)
  • K.J. Friston et al.

    Action and behavior: A free-energy formulation

    Biological Cybernetics

    (2010)
  • K. Friston et al.

    Active inference and agency: Optimal control without cost functions

    Biological Cybernetics

    (2012)
  • K. Friston et al.

    Free-energy minimization and the dark-room problem

    Frontiers in Psychology

    (2012)
  • T. Ganson et al.

    Visual prominence and representationalism

    Philosophical Studies

    (2012)
  • A. Gelman et al.

    Bayesian data analysis

    (2014)
  • Cited by (30)

    • Is predictive coding falsifiable?

      2023, Neuroscience and Biobehavioral Reviews
    • Affect-biased attention and predictive processing

      2020, Cognition
      Citation Excerpt :

      We should not have a high precision expectation that our decision to attend will lead to the sighting of the dog, given the infrequency with which it is actually in the yard. Second, following the arguments of Ransom, Fazelpour, and Mole (2017), PP theory does not provide a coherent account of voluntary attention. Any increased precision in the sensory input will be merely a consequence of attending, rather than what is driving attention.

    • Predictions, precision, and agentive attention

      2017, Consciousness and Cognition
      Citation Excerpt :

      In such cases variable precision weighting turns up the (post-synaptic) ‘volume’ on the selected signals. Attention, as Ransom, Fazelpour, and Mole (2017) rightly stress, is then identified with this mechanism (which may have multiple implementations in the brain, operating in different ways and at different timescales – see Clark (2013b)). Most generally, it is claimed that “attention is simply the process of optimizing precision during hierarchical inference” (Friston, 2009, p. 299).

    • Predictive Processing and Object Recognition

      2023, Expected Experiences: The Predictive Mind in an Uncertain World
    View all citing articles on Scopus
    View full text