Please select your name from the dropdown list after filling out the registration form.

The NUC model assumes that rational agents undergo a hierarchical planning process: First select an intention (a sequence/plan of goals), then take actions to achieve a goal.

Would it be possible to model agents using a "flat" planning process, where agents directly optimize for subjective utility (rewards minus costs), without going through the process of selecting goals?

If so, how would that generative model differ in its predictions?

In Experiment 1, when computing the correlations between human responses and model predictions, the authors ensured that "cost and reward judgments were z-scored separately for each subject [...] then averaged across participants".

Why was it important to perform z-scoring before averaging?

How might the analysis have been affected if z-scoring was not performed?

Experiment 5 tests whether the NUC model can infer whether agents are initially ignorant or knowledgeable about their costs or rewards (i.e. their desires).

However, the paper itself doesn't describe how knowledge/ignorance about one's own desires is accounted for in the generative model.

How do you think the generative model presented in Section 2 could be modified to account for knowledge/ignorance of one's own rewards or costs?