The interaction between attention and learning under conditions of uncertainty

Past research in animals has suggested that attention is distributed to exploit known relationships between stimuli (e.g., Mackintosh, 1975) and explore stimuli whose consequences are uncertain (e.g., Pearce & Hall, 1980). The resulting changes in attention influence how animals learn new information involving those stimuli. While there is strong support for exploitative attention and its effects on learning in humans, the evidence for exploratory attention is less well developed. Five experiments examined whether preferential allocation of attention (as measured by eye-gaze) to cues associated with uncertainty leads to more rapid learning of new associations involving these cues in the future. In each experiment, participants first learned about compounds containing one predictive cue and one non-predictive cue. The level of uncertainty during this first stage of training was also manipulated: cue-outcome relationships were either deterministic (certain) or probabilistic (uncertain). In a second stage, new cue-outcome relationships were trained, and the uncertainty of these relationships could be resolved by learning about an occasion setter (Experiments 1 & 2), or the previously non-predictive cues (Experiments 3 – 5). In Experiments 1 and 2 (Chapter 2), there was evidence suggesting that acquisition of Stage 2 knowledge was impaired for cues associated with uncertainty. This led to the suggestion that participants may form expectations for a given degree of uncertainty when trained with probabilistic contingencies. Experiments 3 and 4 (Chapter 3) showed that previous experience of uncertainty impacts prediction error processing. Participants who experienced an onset of uncertainty at the start of Stage 2 learned novel cue-outcome associations faster than participants for whom uncertainty was constant. Furthermore, participants experiencing unexpected uncertainty showed a greater increase in attention to cues in Stage 2. It was proposed that this effect may be caused by the noise present in probabilistic contingencies reducing sensitivity to subtle changes via an effect on cue associability. Experiment 5 (Chapter 4) provided a test of this idea using a within-subject replication of Experiments 3 and 4. The results suggested that theories of change detection (rather than changes in CS processing) provide an accurate account for the differing effects of expected and unexpected uncertainty