Great Expectations: The Neural Mechanisms of Prediction and Attention in Visual Awareness

Abstract

Mechanisms of attention and prediction operate in the service of visual perception and subjective awareness. Attention enhances the processing of sensory stimuli that are relevant for guiding behaviour (Posner, 1994), whereas prediction uses prior information to guide the processing of new sensory input. Despite their convergent functions, attention and prediction have traditionally been cast as independent mechanisms and exhibit seemingly antagonistic effects on neural activity: attention enhances, whereas prediction suppresses, neural responses to sensory stimuli (Schroger, Marzecova, & SanMiguel, 2015). Recently, however, predictive coding theory has reconciled these effects under a new theoretical framework that casts attention and prediction as synergistic determinants of perceptual awareness (Feldman & Friston, 2010). Perceptual experience is proposed to reflect an inference about the causes of sensory inputs, whereas attention is the inference about the uncertainty of those causes. Mechanistically, attention is proposed to optimise precision expectations by increasing the gain of units encoding the difference between predicted and observed sensory events (so-called ‘prediction errors’).Predictive coding theory generates several important hypotheses about the interdependent relationships between attention, prediction, and awareness, which are explored in this thesis. First, attention is proposed to be an inference about the uncertainty of environmental states, rather than the states themselves, and therefore should be dissociable from awareness. Although recent research supports this hypothesis for certain types of attention, it remains unclear whether spatial attention can dissociate from awareness at the level of neural representations. Second, prediction is proposed to facilitate awareness by building a generative model of likely sensory events. Consistent with this theory, recent research has demonstrated that sensory templates are activated immediately prior to stimulus onset (Kok, Mostert, & de Lange, 2017), but it remains unclear how these templates influence visual awareness. Third, attention and prediction are based on interdependent neural mechanisms, and therefore should interact in their modulation of neural responses to stimuli. Recent studies have explored this hypothesis using a variety of attention and prediction manipulations, with some studies showing interactions (e.g., Czigler & Sulykos, 2010; Kok, Rahnev, Jehee, Lau, & de Lange, 2012; Marzecova, Widmann, SanMiguel, Kotz, & Schroger, 2017) and others showing only main effects (e.g., Garrido, Rowe, Halasz, & Mattingley, 2018; Kok, Jehee, & de Lange, 2012). Because these studies presented stimuli at attended locations or did not adequately control spatial attention with a difficult task, it remains unclear whether feature-based attention mechanisms interact with those of prediction. Finally, the interaction between attention and prediction is proposed to reflect an increase in prediction error, which is encoded as the mismatch between predicted and observed sensory information. Previous studies analysed the interaction between attention and prediction at the level of neural responses to stimuli (e.g., Czigler & Sulykos, 2010; Kok, Rahnev, et al., 2012; Marzecova et al., 2017), but such analyses cannot speak to the type of information underlying the interaction (Kriegeskorte & Bandettini, 2007). As such, it remains unclear whether the interactions observed in those studies were consistent with an increase in (mismatch) prediction errors.The overarching aim of this thesis is to investigate the underlying relationship between attention, prediction and visual awareness in humans, by combining behavioural methods, electroencephalography (EEG), magnetoencephalography (MEG) and computational modelling. In Chapter 1, I summarise the relevant literature on attention, prediction and awareness and introduce a new version of predictive coding theory that proposes to reconcile these three processes under a common framework. In Chapters 2-5, I describe four studies that test each of the four hypotheses described above. I conclude the thesis in Chapter 6 by discussing how the findings of these studies relate to previous literature and highlight important avenues for future research.