Types of Agents — Simple Reflex Agents — Beginner-Friendly Notes

1. What It Is (simplest agent)

Simple reflex = act on the current percept only, using condition → action rules. No memory, no model.

if percept == X then do Y

Think: thermostat, automatic door.

2. How It Works (inside)

Percept → Condition Matcher (rules) → Action

• Rules are hand-written or learned for one situation.
• Example vacuum rules:
  • if status=Dirty → Suck
  • else if location=A → MoveRight
  • else → MoveLeft

3. When It Succeeds

• Fully observable environments — agent sees all it needs right now.
• Episodic tasks — no need to remember past.
• Example: industrial part sorter where each part is judged alone, spam filter per email.

4. Why It Fails (and what exam asks)

• Partial observability: if dirt not seen → can't decide — rule has no memory of other room.
• Infinite loops: vacuum with rule "if dirt → suck else move" can loop between A↔B forever if no dirt sensor history.
• Brittle: every situation needs a rule — impossible for complex worlds.

Limitation in one line: no memory → fails when past matters.

5. Visual Contrast

• Reflex: percept → action
• Next type (model-based): percept + internal state → action

6. One-Line Summary to Write in Exam

A simple reflex agent selects actions solely from current percepts via condition-action rules and works only in fully observable, episodic environments without needing memory.

7. Quick Self-Check

1. Write two rules for a traffic-light controller as a reflex agent.
2. Why would a self-driving car be a bad reflex agent?