Decision Confidence.pdf

Decision Confidence 18 January 2024 13:04 Main Ideas Notes Important research question How is sense of confidence formed during the decision process? Why study confidence? Often make decision in situations of uncertainty or absence of immediate feedback Confidence estimates facilitate adaptive behav...

Decision Confidence 18 January 2024 13:04 Main Ideas Notes Important research question How is sense of confidence formed during the decision process? Why study confidence? Often make decision in situations of uncertainty or absence of immediate feedback Confidence estimates facilitate adaptive behaviour: ○ Changes of mind ○ Error recognition ○ Learning ○ Social interaction Confidence research aims to identity computational neural substrates supporting confidence in simple decisions, employing different species and techniques Notes Limitation of DDM approach to confidence ○ Kiani & Shadlen = DDM model works well for opt-out choices sampled after fixed durations ○ But does not model confidence judgments given after a decision has been made Post-decisional mechanisms for confidence Decision itself always reaches same level of confidence (fixed bound) Evidence continues to accumulate after choice to inform confidence judgments Can capture both error monitoring and confidence Studying post-decisional processing Studying confidence: Type 1 and Type 2 decisions Post-decisional extension of the drift-diffusion model can capture changes of mind and changes of confidence ▪ Non-verbal type 2 decisions ○ ○ Confidence measured as willingness to wait for a reward ▪ How is confidence informed by evidence accumulation signals: ○ Pre-decisional mechanisms for confidence ▪ Macaque monkeys were trained 'opt out' of the random dot motion task when it was difficult - version of post-decision wagering ▪ Performance is better when opt-out option is available ○ Linking the DDM (drift diffusion model) to confidence ▪ Used DDM as a computational model □ Stimulate lots of diffusion paths using a drift-diffusion model □ Record probability correct associated with making a decision at each point in the space □ When stimulus is extinguished, decide whether to opt-out or choose left or right based on expected level of choice accuracy (decision confidence) ○ Confidence signals in monkey LIP ▪ ○ Relating model to data ▪ Model (lines) provide a good fit to the behavioural data ▪ Sorting drift-diffusion model trajectories by whether the monkey had high or low confidence matched neural data Summary Confidence is a ubiquitous feature of decision-making can measure confidence using both explicit measures (type 2 decisions) and non -verbal, implicit metrics such as willingness -to-wait Both pre-decisional and post-decisional aspects of evidence accumulation contribute to confidence Post-decisional evidence accumulation is linked to mechanisms involved in error monitoring (eg pMFC activity) When interpreting the neural basis of confidence, it is important to keep different computations separate, e,g, (sensory) cer tainty vs. decision confidence The ventromedial prefrontal cortex in the human brain may act as a hub for the computation of (domain -general) decision confidence (also rodent OFC) PSYC0032 The Brain in Action Page 1 Relating post-decision processing to error monitoring - Primary task: make a speeded response (A) to all incongruent colour/word stimuli ▪ Withhold response to congruent or repeated stimuli - If error made, subjects should signal the error by pressing a secondary response button - Found detected errors associated with post-decisional EEG signal - Post-decisional EEG signal shows hallmarks of a second drift-diffusion process - Note that the diffusion process here is metacognitive: it is “deciding” whether an error has been made, rather than what the stimulus is Gaining experimental control over post-decisional processing - - Subjects made an initial decision about pre-decision motion, before being given another chance to see the stimulus and the rating their confidence - Coherence titrated individually for each subject - 3 (pre-decision coherence) x 3 (post-decision coherence) factorial design - Bayesian model predicts that confidence should increase with post-decision evidence strength after correct decisions (blue lines; new evidence confirms your original choice) ▪ but decrease with post-decision evidence strength after incorrect decisions (red lines; new evidence disconfirms original choice) - Behavioural data follows model predictions (solid lines) - when post-decision evidence increases, confidence diverges as a function of decision accuracy (blue vs. red) Notes Sensory uncertainty informs confidence decoded sensory uncertainty (from visual cortex) negatively predicts reported confidence (people are less confident when the neural representation of stimulus orientation is more uncertain Conceptual Understanding of Decision Confidence Definition and Importance ○ Decision Confidence: A person's belief in the accuracy of their own decisions. ○ Role in Decision-Making: Confidence guides future actions, helps in evaluating decisions, and influences the willingness to learn from outcomes. Implications ○ In Everyday Decisions: Confidence levels affect choices in daily life, from simple to complex scenarios. ○ In Learning and Adaptation: High confidence often leads to sticking with a choice, while low confidence may trigger reassessment and learning. Neural and Computational Substrates of Confidence Brain Regions Involved ○ Prefrontal Cortex: Key area for integrating information and forming confidence judgments. ○ Posterior Regions: Involved in processing sensory information, contributing to confidence formation. Computational Models ○ Bayesian Models: Explain how the brain might compute confidence as a probability, assessing the likelihood of a decision bein g correct. ○ Signal Detection Theory: Offers a framework for understanding the trade-off between sensitivity (ability to detect) and specificity (avoiding false alarms) in decision-making. Sensory Evidence and Decision Making Influence of Sensory Input ○ Gathering Evidence: How sensory information is accumulated to make a decision. ○ Quality of Sensory Evidence: Affects confidence level; more reliable evidence usually leads to higher confidence. Decision Thresholds ○ Setting Thresholds: The brain sets a 'threshold' for decision-making, influenced by the quality of sensory evidence. ○ Threshold Adjustment: In uncertain situations, thresholds may be adjusted, impacting both decision accuracy and confidence. Adaptive Behavior and Confidence Behavioral Implications ○ Error Recognition: Confidence helps in recognizing and correcting mistakes. ○ Social Interactions: Confidence levels can affect social behaviors and perceptions by others. Confidence in Learning ○ Feedback and Adaptation: Confidence judgments influence how individuals respond to feedback and adapt their strategies. Experimental Studies and Techniques Key Studies ○ Neuroimaging and Confidence: Studies using fMRI and EEG to explore brain regions involved in confidence judgments. ○ Behavioral Experiments: Tasks designed to assess how different variables affect confidence levels. Methodologies ○ Signal Detection Tasks: Used to measure the accuracy and confidence in sensory discrimination tasks. ○ Computational Modelling: Helps in understanding the underlying processes of confidence formation. - - Neural activity in posterior medial frontal cortex consistent with encoding extent to which confidence should be lowered following new evidence Separating decision confidence from sensory uncertainty There are lots of different types of uncertainty being represented by the brain Decision confidence is specific type of uncertainty/certainty about a decision can have uncertainty about multiple quantities (eg perceptual, cognitive) Confidence additionally depends on committing to a (hypothetical) choice By fitting a model to the voxel patterns in visual cortex, - can extract the probability of different orientations being encoded by the subjects’ brain. - The width of this distribution reflects a (neural) measure of sensory uncertainty Type 1 and Type 2 Decisions Characteristics ○ Type 1 Decisions: Fast, intuitive, often based on heuristics. ○ Type 2 Decisions: Slower, more analytical, involving conscious reasoning. Confidence Formation ○ In Type 1: Often based on past experiences and gut feelings. ○ In Type 2: More likely to involve deliberation and weighing of evidence. Notes Notes PSYC0032 The Brain in Action Page 2 Notes PSYC0032 The Brain in Action Page 3 Notes Notes PSYC0032 The Brain in Action Page 4 PSYC0032 The Brain in Action Page 5