three-classic-action-representation-problems

Summary

Any system that tries to reason about what happens when an agent does something faces three stubborn gaps: it can't easily say what stays the same after an action, what had to be true for the action to succeed in the first place, or what indirect effects ripple outward. These aren't minor edge cases; they are structural limits that any action-planning or world-modeling module must explicitly handle, or its predictions about the state of the world will be incomplete or wrong.