Gurkenglas comments on Approval-directed agents

[Gurkenglas]

We then loop over each action a and take the action with the highest expected answer.

Wasn’t the whole point that we want to avoid such goal-direction?