| Summary: | <p>Perceiving and interacting with our world is an amazingly complicated computational problem. Our brains need to infer the best cause of action from a long chain of ambiguous and noisy signals reflecting properties from the environment. To solve this problem, our brains make use of multiple pieces of information available in a given context. In these thesis, we explore the influence of context on decision-making. First, we focus on decisions made in the context were three options are available to be chosen. Second, we focus on decisions made in the context of variable prior information about the probability of occurrence of a given event. A particular emphasis is placed on distinguishing mechanisms that optimize the accuracy of decisions from those that comply with axiomatic prescriptions of rational behaviour. We show that even as simple an assumption as having Gaussian noise distorting the representation of value is enough to drive irrational behaviour, calling into question the validity of the axioms of rationality as a yardstick for measuring human behaviour. We also show that uncertainty arising at different stages of the long chain of information processing, can drastically impact the nature of the inference problem that humans have to solve in different contexts. We show two mechanisms present when humans construct their value preferences between three available options. These mechanisms prioritize what information gets further processed in a way that minimizes the effects of noise arising at later stages of processing. Finally, we provide a novel computational explanation for the mismatch typically observed between near-optimal perceptual choices and suboptimal cognitive judgments. Namely, that people are metacognitive blind to noise arising when integrating of multiple pieces of evidence. Together, we show that the brain tends to make excellent use of contextual information to guide it decisions. Perhaps so much so, that we often trust our judgments blindly.
</p>
|
|---|