Learned changes in outcome associability

The ability to learn associations between events is of crucial importance for human and non-human animals. Without this ability we would be unable to identify when danger is imminent (e.g., the signs of a looming predator) or how to process the causal structure of our environment (e.g., understanding the role of causes and their corresponding effects). As such, it is essential to understand the factors which influence the formation of learned associations. Indeed, a plethora of studies have been conducted with this goal in mind and many models have been devised to elucidate the factors which influence how associative learning between events arises. A critical factor that a number of these models have identified concerns how the predictiveness of a cue, or a conditioned stimulus, can influence associative learning. For instance, studies have identified that a cue that has reliably predicted an outcome in the past, enters into subsequent associations better than a cue that has been an unreliable predictor of an outcome (e.g., the learned predictiveness effect; see Le Pelley & McLaren, 2003). The predictive ability of a cue can also interact with and influence learning about other cues. For example, prior learning about a cue and an outcome can block leaning between an additional cue and the same outcome, when these cues are presented together (i.e., blocking; see Kamin, 1968; 1969). Furthermore, simply presenting a cue in isolation or in compound – with no prior exposure to either – can also influence learning, with the elements of a cue compound entering into an association with an outcome to a lesser extent than a cue presented in isolation (i.e., overshadowing, see Pavlov, 1927). As such, a considerable amount of evidence has clearly identified the importance of the prior predictive ability of a cue and the extent to which cues can interact with one another to influence learning. However, much less research has explored the extent to which the predictability of an outcome can influence learning and the extent to which outcomes can interact with one another. Interestingly, key associative learning models assume that the prior predictability of an outcome should have no bearing on subsequent learning about this outcome, once separated from the cue which it was first paired with (e.g. Esber & Haselgrove, 2011; Le Pelley, 2004; Mackintosh, 1975; Pearce & Hall, 1980; Rescorla & Wagner, 1972). These models also assume that outcomes are unable to interact with other outcomes, when presented in compound, in the same manner as cues. This approach appears logical: cues can provide information which allows an organism to change its behaviour in anticipation of an event, therefore providing the organism with an evolutionary advantage. Consequently, it appears reasonable to suggest that the associability of cues can be modified based on their associative history or due to the presence of other cues. In contrast, however, when an outcome is presented the event of significance has already happened. As such, there appears to be no evolutionary advantage to the associability of an outcome being modified based on its prior associability, or the presence of other outcomes: once the event has happened it is too late for any advantage to be gained. Nevertheless, there is evidence to suggest that the prior predictability of an outcome can influence subsequent learning about this outcome. In light of this, Experiments 1 – 4 of Chapter 2 sought to further investigate whether the prior predictability of an outcome can influence subsequent learning about this outcome when separated from the cue which the outcome was first paired with. Experiments 1 and 2 revealed that participants learned better about a previously well-predicted outcome than a previously less well-predicted outcome, a finding which is inconsistent with key associative learning models. Experiments 3 and 4 respectively revealed that the concurrent predictability of the outcomes and Stage 1 training is important in obtaining this effect. Chapter 3 sought to assess whether the prior predictability of an outcome could influence subsequent learning about an additional outcome, when these outcomes were presented in compound. The experiments in this chapter employed an outcome blocking procedure using a food allergist task. In Experiment 5 an outcome blocking procedure was employed where an outcome compound was presented consisting of a previously well-predicted outcome and a novel outcome. In addition, an outcome compound was presented featuring a novel outcome and a surprising outcome. In this experiment an interaction effect was observed providing evidence to suggest that the prior predictability of an outcome can influence learning about other outcomes, as has been suggested by some other studies. In contrast to some previous studies in animals, however, an outcome facilitation effect was observed as opposed to the blocking effect which is typically – although not exclusively – seen in cues (cf. Batson & Batsell, 2000). Experiment 6 replicated this effect using a diagnostic task where the outcomes were causes and the cues were effects, thus demonstrating that the effect was present regardless of the type of judgement task participants completed. These findings are inconsistent with the predictions of causal model theory (e.g. Waldmann & Holyoak, 1992). Experiment 7 employed an additional type of control trial within the outcome blocking procedure employed in Experiments 5 – 6. When employing this additional type of control, which received exposure only in Stage 2, an outcome interaction effect was again observed. However, when this additional control trial was employed an outcome competition effect was observed in contrast to the outcome facilitation effect. Experiment 8 sought to tease apart two theoretical accounts derived from the Mackintosh model (1975) and the Standard Operating Procedures (SOP) model proposed by Wagner (1981) to account for the findings of Experiment 7. According to the Mackintosh account the effects reported in Experiments 5-7 should only be present following the second trial when the outcomes are presented in compound (i.e., when the associability of the cues have had an opportunity to be modified). The SOP account, however, stipulates that the effect should be present following the first trial. As such, Experiment 8 employed a one-trial outcome blocking procedure to test these accounts. This experiment provided partial support for the SOP account of the experiments reported in this chapter (whilst facilitation was observed amongst the outcomes competition was not). To further explore whether outcome interaction effects can be observed when additional types of procedures are employed, Chapter 4 sought to assess whether interaction effects could also be observed when outcome overshadowing and outcome relative validity procedures were employed. These experiments also used the food allergist task. In the outcome overshadowing experiment, both a single (i.e., elemental) outcome and a compound outcome (i.e., two outcomes) were reliably predicted. In the outcome relative validity task outcome compounds were presented which contained a common outcome (i.e., an outcome which was predicted by more than one cue) and unique outcomes. Some of these outcome compounds were reliably predicted and some were partially predicted. When an outcome relative validity procedure was employed no interaction effects were observed. One possible account of this finding is to suggest that both competition and facilitation effects were operating simultaneously, thus obscuring any differences. Yet, when an outcome overshadowing procedure was employed a competition effect was observed which is typically seen in cues. That is, an outcome which was presented alone was learned about better than when it was presented in compound with an additional outcome. Taken together these results provide evidence to suggest that the associative history of an outcome can influence subsequent learning about this outcome when separated from the cue which the outcome was first paired with. The results of Chapter 3 suggest that both facilitation and competition effects can be observed amongst outcomes, thus being similar to cues, which have also displayed both competition and facilitation effects. Although there was an absence of an interaction effect when a relative validity procedure was employed in Experiments 10 and 11 in Chapter 4, Experiment 9 of Chapter 4 also provides evidence to suggest that the associability of an outcome can interact with and influence subsequent learning about additional outcomes when an outcome overshadowing procedure is employed. As such, these experiments suggest that outcomes might be more similar to cues than originally conceived of in associative learning models. There are, however, several important factors to consider and methodological issues which arise when employing outcome associability and interaction procedures. The challenge of future work will be to identify the key factors which influence outcome associability, to understand the impact of this on key associative learning models, and to consider how outcomes are conceived of and distinguished from cues