Types of Environments — Single-Agent vs Multi-Agent — Beginner-Friendly Notes
1. Simple Definition
- Single-agent: only one agent acts; other things are just objects/nature — no other decision-makers to model.
- Multi-agent: multiple agents act, each with its own goals — what you should do depends on what others do.
One agent = puzzle. Many agents = game/society.
2. Two Sub-Cases of Multi-Agent
- Competitive: agents have conflicting goals (chess, poker, market bidding) — adversarial search (Module 2: Minimax, Alpha-Beta).
- Cooperative: agents share or align goals (warehouse robots, drone swarm, self-driving cars avoiding collisions) — needs coordination/communication.
Many real tasks are mixed — e.g., self-driving cooperates to avoid crash but competes for road space.
3. Examples
| Environment | Single or Multi? | Reason |
|---|---|---|
| Crossword, Vacuum (alone) | Single-Agent | Only you act |
| Chess, Tic-Tac-Toe | Multi-Agent (competitive) | Opponent responds |
| Poker | Multi-Agent (competitive + chance) | Hidden info + opponents |
| Factory floor with 5 robots | Multi-Agent (cooperative) | Must avoid collision, share tasks |
| Taxi driving | Multi-Agent (mostly cooperative, partly competitive) | Other drivers/pedestrians are agents |
4. Why It Matters for Design
- Single-agent: use search / planning — environment is just physics.
- Multi-agent: must model other agents — game theory, minimax, or communication/negotiation. Complexity explodes (branching includes others' moves).
5. Two Traps for Exams
- Is a human in the loop an agent? Yes — if their behaviour affects your success and they make decisions (e.g., pedestrian, opponent, user). A wall is not an agent.
- Does "single-agent" mean one robot in the world? No — if there are other decision-makers (people, other cars), it's multi-agent even if your system is one robot.
6. One-Line Summary to Write in Exam
A single-agent environment contains one decision-maker acting against nature, while a multi-agent environment contains multiple decision-makers requiring adversarial or cooperative reasoning.
7. Quick Self-Check
- Is spam filtering single or multi-agent?
- Why is chess called competitive multi-agent rather than just stochastic?
- Give a cooperative multi-agent AI example and its coordination need.