Automaticity and Cognitive Control in the Learned Predictiveness Effect

Abstract

This thesis explores changes in attention that occur in response to predictive learning. Learned predictiveness is a bias in learning towards cues with prior predictive utility, or indeed a bias away from cues that have poor prior utility. In a typical learned predictiveness experiment, predictive validity is first manipulated such that cues are either predictive or non-predictive of a set of outcomes. In a subsequent and seemingly unrelated task involving novel outcomes, new learning is biased in favour of the previously predictive cues. This bias has been interpreted as a shift in attention towards items with predictive utility. However, the relative contribution of automatic and controlled selection mechanisms in producing the bias remain to be fully characterised. The studies reported here examine the expression of learned predictiveness using cognitive tasks that involve competition for stimulus processing. Chapter 2 measured the processing of predictive and non-predictive items in the attentional blink, a visual detection task that limits the availability of controlled attention. There was no evidence that the predictive history of targets was associated with variations in their detection. However, predictiveness did influence target detection indirectly by virtue of the predictive history of critical distractors, such that targets were easier to identify when they were immediately flanked by non-predictive stimuli. In contrast, novelty was a potent source of processing bias for both targets and distractors: Novel targets were easier to detect than familiar targets. In Chapter 3, controlled attention was manipulated by issuing instructions about the causal nature of cues, and differentiating between measures of associative memory and causal reasoning. Experiment 7 – Experiment 8 attempted to tease apart inferential and automatic contributions to the bias by presenting instructed causes that were predictive and non-predictive in initial training, as well as instructed non-causes that were predictive and non-predictive in initial training. The results showed that the predictive history of cues influenced subsequent learning over and above the effect of explicit instruction, suggesting an automatic bias that persists in the presence of top-down control. However, the results for cues instructed as non-causes suggest that the relationship between explicit instruction and predictiveness was interactive rather than additive. Chapter 4 extended this procedure to include the presence of novel stimuli. Learned predictiveness in the presence of novelty was first assessed in the absence of explicit instruction about which cues were causal. There was no evidence for the bias when predictive and non-predictive cues appeared alongside novel items at the start of the second task. When instructions about the causal structure of the task were reintroduced, novelty was once again a potent source of selection bias, such that learning favoured novel items instructed as causal. In Experiment 11, there was evidence that the predictive history of items was influencing associative memory for cues known to be non-causal of an outcome, though this pattern of results was slightly different to those observed in Chapter 3. This suggests that the presence of novelty changes the way in which the automatic effects of predictive history interact with top-down control. The results are discussed in Chapter 5 in relation to the interaction between automatic and controlled selection mechanisms, as well as theories of learning and attention.