I think this is a real and useful distinction. In a recent post, I sketched a system which I’d describe in your terms as actualizing by default, but would be more like a possibilizer if the observation = execute(action)line were commented out. The possibilizing system could be used to cache sequences of actions for reaching a particular world state, stored for potential later execution.
In the limit of perfect world modelling, planning heuristics, and compute resources, the set of reachable nodes in the tree might be the same as those in the actualizing version. But for more practical instantiations, the possibilizing system likely becomes much less accurate when searching deep in the tree.
Instead of preventing action execution entirely, you can imagine restricting only certain kinds of actions or action sequences whose effects are irreversible or otherwise high-impact. As you lift more and more restrictions, the systems transitions back towards actualizing.
I think this is a real and useful distinction. In a recent post, I sketched a system which I’d describe in your terms as actualizing by default, but would be more like a possibilizer if the
observation = execute(action)line were commented out. The possibilizing system could be used to cache sequences of actions for reaching a particular world state, stored for potential later execution.In the limit of perfect world modelling, planning heuristics, and compute resources, the set of reachable nodes in the tree might be the same as those in the actualizing version. But for more practical instantiations, the possibilizing system likely becomes much less accurate when searching deep in the tree.
Instead of preventing action execution entirely, you can imagine restricting only certain kinds of actions or action sequences whose effects are irreversible or otherwise high-impact. As you lift more and more restrictions, the systems transitions back towards actualizing.