Investigating the Temporal Dynamics of Covert Visual Spatial Attention: Exploitative and Explorative Attentional Mechanisms

Abstract

This thesis explores how trial and error learning affects attentional processes. Previous research has shown that selective attention tends to be biased towards cues that accurately and consistently predict future events. Paying attention to predictive cues is adaptive because an animal can use the information that these cues convey to predict future events and change their behaviour accordingly. However, some research has shown that selective attention can be biased towards non-predictive, or uncertain, stimuli. Paying attention to non-predictive cues could also be adaptive because it could help establish the true nature of these currently uncertain stimuli. Although selective attention can be driven by both high and low levels of uncertainty, the factors that determine which driving force prevails are not fully understood. This thesis investigates whether time is one such factor. The experiments presented here involved training participants on a categorisation task where some stimuli were predictive (P) of the categorisation response while others were non-predictive (NP). These stimuli were then used as uninformative spatial cues to a target stimulus in a dot probe task. The time course of attention to the cues was investigated by manipulating the stimulus onset asynchrony (SOA) between the cues and the target. Behavioural and electrophysiological (EEG) data were collected. It was hypothesised that P cues would be preferentially processed early in a trial. However, after these cues were processed, we predicted that they would be inhibited, and that this inhibition would bias attention towards the currently NP cue. Experiments 1-3 (Chapter 2) explored the dot probe paradigm by using different stimuli and different dot probe tasks. Using two SOAs (250 and 1200 ms) and an intermixed dot probe and categorisation task, the reaction time (RT) and N2 posterior contralateral (N2pc) results showed that targets that appeared over P cues after short SOAs were easier to process compared to targets that appeared over NP cues. Therefore, P cues were preferentially processed early in a trial. However, no evidence of inhibition of the P cue was found at the longer SOA. Experiments 4-6 (Chapter 3) tested a wider and earlier range of SOAs and included additional behavioural measures (e.g., dot probe errors and premature responding). In these experiments, RTs were faster to targets that appeared over P cues compared to NP cues, and this advantage increased proportionally with SOA. This novel RT interaction suggested that the P cues were being strategically processed. The N2pc results also showed an interaction between predictiveness and SOA, but one that suggested a shift in attention from P to NP cues. The error data indicated that RTs in the localisation version of the dot probe task could be contaminated by a non-attentional response bias towards selecting a response that was congruent with the location of the P cues. Experiments 7-9 (Chapter 4) tested whether participants were strategically or automatically attending to the P and NP cues, and also investigated whether the attentional effects observed in the previous chapters could influence subsequent cue-outcome learning. Blocking the dot probe and categorisation task resulted in the loss of the interaction between predictiveness and SOA on RTs. Instead, the blocked design resulted in a small, but consistent, RT advantage towards the P cues across all SOAs. This suggested that the current task relevance of the cues is an important factor that determines whether they are strategically processed. In the final experiment, different target stimuli were used after short and long SOAs to investigate whether the changes in attention measured via the N2pc could impact subsequent cue-target learning. However, no evidence of biased learning was found. Chapter 5 presents a summary of the results, including meta-analyses of the behavioural data. The idea that time may be an important factor which moderates exploitative and explorative behaviour is discussed further. This discussion pivots around a real-time model that was modified to take attention into account. Ideas for future research are also presented.