Over the course of its first 18 months the project has yielded a number of important developments and insights. A first goal of the project was to address a major limitation of current models of decision making which is that they do not account for the role that neural noise can play in shaping our decisions. This is particularly problematic because well known that neural noise levels change as we age and also differ in a number of prominent clinical populations. Consequently, models that fail to account for neural noise are likely missing critical parts of the picture when accounting for individual differences in decision making behaviour. Our team has conducted several investigations of new methods for directly estimating neural noise in such a manner that those estimates can then directly inform models. Our first effort in this direction was to apply a method originally developed in the field of psychophsics in which neural noise is estimated by progressively increasing the noisiness of a visual stimulus until a detectable impact on behaviour is observed - the logic being that higher levels of stimulus noise will be required to affect the behaviour of individuals with high neural noise. We successfully incorporated these estimates into a mathematical decision model and found the resultant model offered new insights into the brain changes that accompany extended training on a visual task. Specifically, we found that neural noise levels decreased as a function of training, resulting in improved accuracy and faster reaction times. In a separate study, we used an alternative, complementary approach in which we devised a novel model that could account for variations in neural noise across individuals and found that older subjects did not exhibit any differences in noise levels compared to younger subjects, a result that contradicted our expectations.
Another key goal of the project is to improve our understanding of how prior knowledge and expectations influence our decision making. According to predictive processing accounts, expectations fundamentally shape all neural activity yet several electrophysiological studies in humans have reported no effect of prior knowledge on early visual responses. We reasoned that this might be because the visual responses being analysed were not necessarily relevant to the task the participant was performing. We therefore devised a new study in which participants were required to discriminate the contrast levels of two visual stimuli and extracted a measure of contrast sensitivity in visual cortex using novel EEG methods. At the outset of certain trials, participants were presented with a cue which indicated what the correct choice was likely to be. In line with previous work, we found that the cues had a strong influence on choice behaviour, with participants more likely to report the cued direction and doing so more rapidly. Examining the visual contrast responses in the EEG, we found no effect of these cues in the early periods of testing but an effect did emerge after several hours of exposure to the task with visual contrast responses being biased in favour of the cued choice. This is an important observation because it suggests that early visual responses can be shaped by expectations while also providing an explanation for the failure of previous studies to observe such effects - our testing session was far longer than those of previous studies, thus allowing to pick out this slowly emerging effect.
Another important area of progress for this project has been in determining how we make decisions in contexts in which we cannot predict when relevant information will present itself. In particular, we have examined tasks in which we must continuously monitor the visual environment for an unpredictable target, such as when driving down a narrow country at night time and monitoring the bend ahead for oncoming headlamps. Our EEG and comuptational modelling analyses highlight that in these contexts we continually accumulate sensory information until a sufficient level of confidence has been reached that a target is present. To avoid frequent false alarms, our brains allow older samples of information to be forgotten and, in addition, the quantity of evidence required for us to report a target is continually adjusted in the brain such that we are willing to report targets based on less information during periods where targets are more expected (e.g. if a target has not been seen for a long time).
Finally, the project has made significant progress in elucidating the neural mechanisms that underpin metacognition - the cognitive operations that allow us to evaluate the accuracy of our choices. Our EEG methods make it possible to track the evidence accumulation process that informs our decisions on a milisecond by milisecond basis. Here, we have found that this same process actually continues to evolve even after we have committed to a choice and this allows newly encountered information to influence our subsequent confidence reports.