Help Privacy Policy Disclaimer
  Advanced SearchBrowse





Learning about Self-Performance in the Absence of Feedback

There are no MPG-Authors in the publication available
Fulltext (restricted access)
There are currently no full texts shared for your IP range.
Fulltext (public)
There are no public fulltexts stored in PuRe
Supplementary Material (public)
There is no public supplementary material available

Rouault, M., Dayan, P., & Fleming, S. (2018). Learning about Self-Performance in the Absence of Feedback. Poster presented at Eighth International Symposium on Biology of Decision Making (SBDM 2018), Paris, France.

Cite as: https://hdl.handle.net/21.11116/0000-0004-DA72-6
Accurately evaluating the outcome of our decisions is essential to adaptive behavior. Thus, learning from external feedback over time has been extensively studied. However, in real life, immediate feedback is often lacking, and much less is known about the computational mechanisms underlying learning in this case. Previous work on metacognition has focused on mechanisms supporting “instantaneous” metacognitive evaluations, elicited at or around the time of a particular decision, showing for instance that subjects can detect performance errors, and be appropriately confident in their decisions. In contrast, it remains poorly understood how people aggregate confidence over time to build “global” beliefs about performance. For instance, when estimating our skill level at a sport, we may reflect on our performance over multiple games, gradually forming a belief about our ability.
Here we developed a novel behavioral paradigm to investigate how subjects learn about self-performance over time, and specifically to assess whether internal confidence may serve as a learning signal in the absence of feedback. In short learning blocks, human subjects (N=29) performed two perceptual tasks (interleaved trials). Each pair of tasks was chosen according to a 2 by 2 factorial design crossing task difficulty (easy, difficult) and feedback (present, absent). At the end of a block, we measured subjects’ beliefs about their performance in each task (self-performance estimates, SPEs), either indirectly (by asking them to choose which task they think they are better at) or directly (via confidence ratings). We found that objective task performance and reaction times were similar in the presence and absence of feedback. Strikingly however, participants showed substantially lower SPEs in the absence of feedback. We replicated these findings (N=29 new subjects), whilst also varying the length of each block to assess effects of learning duration on SPEs. Notably we found that performance on a given block (both in the presence and absence of feedback) influenced end-of-block SPE’s, indicating subjects were sensitive to local fluctuations in performance over and above objective difficulty level.
To explain the formation of subjects’ SPEs, we developed a hierarchical learning model which updates global beliefs about self-performance based on locally computed confidence estimates. Initial model simulations and model comparison suggest that local confidence may act as a learning signal in the absence of feedback, allowing subjects to develop insight into their abilities. In a final experiment (N=34), we sought to assess directly whether local confidence affects global SPEs over the course of learning. On trials in which feedback was not provided subjects were instead asked to provide a local confidence estimate. We found that the greater the difference in local confidence between a pair of tasks, the greater the difference in end-of-block SPEs, supporting this link. Our findings build a bridge between literatures on metacognition and learning, and support a functional role for confidence in higher-order behavioral control.