More results can be found in our preprint (congruency effects, DDM), feel free to check it out 😊. We're currently working on the fMRI version and preparing an EEG study. Feedback and discussions are very welcomed! (13/13)

psyarxiv.com/7cweu/

In summary, our study highlighted humans’ capacity to dynamically modulate task preparation based on their active inference of the environment. Our results also provide evidence for theories of hierarchical control and the trade-off between cognitive flexibility and stability. (12/13)

A follow-up experiment further indicated that this changes in behavior was control dependent. In a 3rd and 4th experiment, we showed that these dynamic modulations can be applied compositionally, depending on which task dimension was expected to transform. (11/13)

In contrast, in transform blocks, the RT pattern reverted by showing an increasing trend, reflecting a RT cost when the original task representation was potentially destabilized due to a prediction of task transformation. (10/13)

However, when the CTI was getting even longer, the RT pattern diverged between block types. In regular blocks, when participants were always certain about the to-be-performed task, the RT further decreased with longer CTI, suggesting continual task preparation. (9/13)

Here it shows the RT result. The x axis denotes the length of CTI. As we can see, when CTI was relatively short, RT of both block types decreased with longer CTIs, which was not surprising since longer CTIs would give participants more time to prepare for the task. (8/13)

2 block types were created. In regular blocks, tasks never transformed. In transform blocks, 9% of the trials resulted in a transformation. We ONLY analyzed the regular trials in both block types. (7/13)

Moreover, the length of CTI was predictive of the likelihood of a task transformation, with longer CTI indicating a higher likelihood. Thus, when participants were preparing for the task , the longer they waited, the more likely the task would transform. (6/13)

The main manipulation is was that on 9% of the trials a task transformation could appear, replacing one of the two task components, in which case participants had to perform the second cue instead of the first. (5/13)

As you can see, the task cue was defined by two dimensions, one specifying the stimulus category (either animal, place, or vehicle), and the other one signalling the categorization rule (either, age, size, or habitat). By permutating these two dimensions, 9 tasks were created. (4/13)

In order to investigate such flexible behavior in the lab, we created a paradigm, namely the task transformation paradigm, in which participants were asked to perform an image categorization task. A regular trial looks like this: (3/13)

In daily life, we often need to dynamically adjust our goals and intentions. E.g., in basketball, we have to monitor and predict what the opposite team will do in order to constantly adjust our strategies. Such control dynamics are a hallmark of human cognitive flexibility. (2/13)