Project Details
Projekt Print View

The influence of the outcome on predictive learning

Applicant Dr. Anna Thorwart
Subject Area General, Cognitive and Mathematical Psychology
Term from 2014 to 2021
Project identifier Deutsche Forschungsgemeinschaft (DFG) - Project number 256632786
 
Final Report Year 2022

Final Report Abstract

Humans (and non-human animals) learn automatically and with little effort about relationships between events. They have evolved this capacity because knowledge about such relationships allows them to predict (and thereby control) both appetitive and aversive outcomes of other events or their own behaviour. The current project focused on an important but often overlooked component: the influence of the outcome itself on such predictive learning. In particular, the project investigated how knowledge about the prior predictability of the outcome biases future learning. After having established that such a learning bias towards predictability exists in a previous project, the current project concentrated on possible mechanisms. A first series of experiments demonstrated that learning biases are related to effects of predictability on the motivational value of the outcome. When humans are given a free choice about which outcomes to learn next, they choose outcomes that they have previously been able to successfully predict in another task. However, people are not only motivated by making as many correct predictions as possible. Interestingly, as soon as they can predict these outcomes correctly in the new task as well, they change their strategy. They then choose to learn about those outcomes for which they previously made many wrong predictions and for which they will do so in the beginning of the new tasks as well. Another experiment showed that humans also learn better about outcomes that were previously associated with a higher reward for correct predictions. On the other hand, the project found no evidence that competing knowledge, which humans acquired about the poorly predictable outcomes in the first task, blocks learning in the second task. Also, reasoning processes and higher-order inferences are not necessarily involved in the emergence of the effects.

Publications

 
 

Additional Information

Textvergrößerung und Kontrastanpassung