Reasoning is the process of:
At the simplest level:
Input → Thinking → Conclusion
Example:
It is raining.
I will get wet outside.
Therefore:
Take an umbrella.Rule → specific conclusion
Example:
All birds have wings.
A sparrow is a bird.
Therefore:
A sparrow has wings.If rules are true:
Very logical and strict.
Used in:
Examples → general rule
Example:
The sun rose yesterday.
The sun rose today.
Probably:
The sun rises every day.Conclusion is probable, not guaranteed.
Used in:
Best explanation from incomplete evidence.
Example:
The ground is wet.
Best explanation:
It probably rained.Could be wrong:
But it is the most likely explanation.
Used in:
Planning is reasoning about the future.
You start with:
Current state → Goal → StepsExample:
Goal: “Make tea”
Plan:
1. Boil water
2. Put tea bag in cup
3. Pour water
4. Wait
5. DrinkPlanning = breaking a goal into ordered actions.
Very simple planning system:
Goal
↓
Requirements
↓
Actions
↓
Order
↓
ResultGoal:
Reach schoolRequirements:
- clothes
- backpack
- transportationActions:
1. Wake up
2. Get dressed
3. Eat breakfast
4. Leave house
5. TravelOrder matters.
You cannot:
This introduces dependency reasoning.
Planning is often based on states.
Example:
Door = closedAction:
open doorNew state:
Door = openAI planners often think like this:
Current state
→ apply action
→ new state
→ repeatEvery action has:
Example:
Action: “Drive car”
Preconditions:
- have key
- have fuel
- inside carEffects:
- reach destinationThis is the foundation of classical AI planning.
Very basic reasoning loop:
1. Define goal
2. Check current state
3. Find missing requirements
4. Choose action
5. Apply action
6. Repeat until goal reachedGoal:
Eat foodCurrent state:
No food availableReasoning:
Need food
→ go to store
→ buy food
→ cook food
→ eatThis is chain reasoning.
Large problems become smaller problems.
Example:
Goal:
Build a websiteSubgoals:
- design UI
- create backend
- setup database
- deploy serverEach subgoal has its own plan.
This is hierarchical planning.
Compare:
Current state
vs
Goal stateThen reduce the difference.
Example:
Current:
I am in MissouriGoal:
Be in New YorkDifference:
location mismatchAction:
travelRepeat until difference disappears.
Planning is often searching through possibilities.
Example maze:
Start
├── Left
├── Right
└── ForwardAI evaluates paths until goal found.
Common strategies:
Heuristic = useful shortcut.
Example:
When finding a route:
Move toward destinationNot always perfect. But faster.
Humans use heuristics constantly.
Immediate response.
Example:
Touch fire
→ pull hand awayFast. No deep planning.
Thinking ahead.
Example:
If I leave now,
I can avoid traffic.Slower but smarter.
Uses explicit symbols and rules.
Example:
IF hungry
AND food exists
THEN eatClassic AI.
Very interpretable.
Handles uncertainty.
Example:
Clouds are dark
→ 80% chance of rainUsed in:
Uses everyday knowledge.
Example:
Ice melts in heatHumans do this naturally.
AI struggles with this more than expected.
Understands cause and effect.
Example:
Push glass
→ glass falls
→ glass breaksNot just correlation.
Very important for intelligent systems.
Reasoning about time.
Example:
Breakfast happens before lunch.Needed for:
Robot task:
Goal:
Cup on tableCurrent state:
Cup in kitchen
Robot in living roomPlan:
1. Move to kitchen
2. Pick up cup
3. Move to table
4. Put cup downEach action changes world state.
Simple architecture:
Input
↓
Perception
↓
Knowledge
↓
Reasoning
↓
Planning
↓
ActionExample:
Camera sees obstacle
↓
Reasoning detects danger
↓
Planning computes new path
↓
Robot turns leftHumans usually:
1. Define goal
2. Estimate difficulty
3. Break into tasks
4. Prioritize
5. Execute
6. Monitor progress
7. AdjustGood planning is iterative.
You can think of reasoning as:
Facts + Rules + Goals
→ DecisionsAnd planning as:
Current State + Goal
→ Sequence of Actions