Answer: A model-based agent keeps an internal state of the world and acts by condition-action rules; a goal-based agent adds explicit goals and plans/searches for actions that reach them; a utility-based agent adds a utility function and picks the action that gives the highest expected utility (best way to the goal).
1. Model-based reflex agent
- Works in a partially observable environment by keeping an internal state that is updated from the percept history.
- Uses a model of the world: (a) how the world changes on its own and (b) what its own actions do.
- Chooses an action with condition-action (if-then) rules applied to the current state.
- Example: a self-driving car that remembers a car in its blind spot even when the sensor cannot see it now.
Diagram (in words): Environment → Sensors → "What the world is like now" (computed from percept + internal State + "How the world evolves" + "What my actions do") → Condition-action rules → "What action I should do now" → Actuators → Environment.
2. Goal-based agent
- Has everything a model-based agent has, plus goal information describing desirable situations.
- Asks "What will happen if I do action A?" and chooses actions that lead to the goal, using search and planning.
- More flexible: changing the goal changes the behaviour without rewriting rules.
- Only distinguishes goal reached / not reached (binary).
- Example: a route-finding agent that must reach a given destination.
Diagram (in words): Environment → Sensors → "What the world is like now" (State + world model) → "What it will be like if I do action A" → compared with Goals → "What action I should do now" → Actuators → Environment.
3. Utility-based agent
- Adds a utility function that maps each state to a number showing "how happy" the agent will be in that state.
- When several action sequences reach the goal, or goals conflict, or outcomes are uncertain, it chooses the action that maximises expected utility.
- Handles trade-offs such as speed vs safety vs cost.
- Example: a navigation agent that picks the route that is fastest, cheapest in fuel and safest, not just any route.
Diagram (in words): Environment → Sensors → "What the world is like now" (State + world model) → "What it will be like if I do action A" → Utility: "How happy I will be in such a state" → "What action I should do now" → Actuators → Environment.
Comparison:
| Basis | Model-based agent | Goal-based agent | Utility-based agent |
|---|---|---|---|
| Extra knowledge used | Internal state + world model | World model + goals | World model + utility function |
| Decision method | Condition-action rules on current state | Search/planning to reach a goal | Choose action with maximum expected utility |
| Looks ahead to future? | No, reacts to the current state | Yes, considers results of actions | Yes, and also rates how good each result is |
| Measure of success | Rule fired correctly | Goal achieved or not (binary) | Degree of satisfaction (real number) |
| Flexibility | Low; rules must be rewritten for new tasks | Medium; change the goal to change behaviour | High; handles conflicting goals and uncertainty |
| Complexity | Lowest | Higher (search cost) | Highest (utility computation) |
| Example | Car tracking unseen vehicles | Agent reaching a destination | Agent choosing the best route among many |