Naar een mechanistisch begrip van zelfevaluatie van beslissingen
Year: 2025
Type: dissertation
Abstract: Feelings of confidence are ubiquitous in decision-making, representing the uncertainty surrounding decisions. Confidence guides behavior, such as seeking more information to support our decisions when confidence is low, yet it can sometimes be miscalibrated. According to the dominant framework, confidence reflects the estimated probability of making a correct decision, based on sampled information supporting the decision. Importantly, appropriately mapping sampled information to the corresponding probability that a decision is correct requires holding task-level representations of confidence. All else being equal, I should be more confident in an easy than in a difficult task context, simply because my decisions are on average more correct in the easy context. In this thesis, I investigate how task-level representations of confidence are formed and updated in humans. In Chapter 2, I manipulated people's prior beliefs in their task performance to provide direct evidence that humans use task-level representations of confidence to inform local confidence judgments, a finding that replicated in all subsequent chapters. In Chapter 3, I developed a new model of confidence that describes how cognitive agents may efficiently approximate their probability of being correct with their confidence estimates. I validated the parameterization of this model using a novel model-generated feedback manipulation approach. In Chapter 4, I provided experimental evidence that confidence computation is dynamically updated from one decision to the next. I introduce a learning extension of my model developed in Chapter 3 that successfully accounts for these dynamics. In Chapter 5, I explored how humans inform their confidence computation in novel tasks with their experience on similar tasks. In two experiments, I biased confidence via manipulated feedback in one task to observe a transfer of this bias to similar tasks only. I concluded that humans compositionally organize their confidence representations. Together, these results increase our knowledge on the involvement of confidence representations in the computation of local decisions, as well as how those representations are constructed and learned. Additionally, the selective influence of feedback manipulation on confidence holds great promise for future studies of the mechanisms underlying the computation of confidence, their neural implementation, and how confidence relates to behavioral adaptations.
Source: Lirias
Author: Le Denmat, Pierre
Language: English
Subfield: General Decision Sciences
Field: Decision Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7113322006