Concept · Chapter 1: What Is Artificial Intelligence?
Planning
Planning means finding a sequence of actions that turns the current situation into a goal situation, using a model of what each action requires and changes.
The problem
A robot or program that must achieve a goal needs to decide what to do, in what order — not just react.
The solution
Describe actions by their preconditions and effects (as STRIPS did for Shakey), then search for a sequence of actions whose combined effects achieve the goal.
The consequence
Planning became a core AI discipline; today the same idea reappears when language-model agents break a task into steps and choose tools.
You should understand first
- The Turing Test
- Symbolic AI
- Search
- Planning
Tiny example
Goal: a cup of tea. Actions: boil water (needs: kettle filled; causes: hot water), fill kettle (needs: nothing; causes: kettle filled), brew (needs: hot water, teabag in cup; causes: tea). A planner searches for an order that satisfies every precondition: fill → boil → (put teabag in cup) → brew.
From Shakey to agents
SRI's Shakey robot (1966–72) was the classic early planner, using the STRIPS representation of actions as preconditions and effects. The difficulty that emerged — describing everything an action does and doesn't change in a messy world — is known as the frame problem.
Modern LLM-based agents plan in natural language rather than formal logic; they're far more flexible, but their plans aren't checked for validity the way a classical planner's are Active research — a theme of Chapter 13.
What to remember
- Plan = sequence of actions from the current state to a goal.
- Each action has preconditions (when it can be used) and effects (what it changes).
- Planning is search over states described symbolically.
- LLM agents 'plan' too, but without guarantees that the plan is valid.