Agent Program Designs

Question 1

Simple reflex

Answer 1

• Doesn’t take history into account
• Condition-action rules
• Simple to implement but limited intelligence
• Success depend on observability

Question 2

Model reflex

Answer 2

• History taken into account -> previous internal state is combined with current state to update
• Helps deal with partial observability
• Requires knowledge to encode in agent to create model of world

Question 3

Goal based

Answer 3

• Selects appropriate action to achieve particular desired state
• Decision making can be complicated (when dealing w/long sequences) -> search & planning might be required
• Goal-based agents more flexible to modification than reflex-based agents

Question 4

Utility based

Answer 4

• Make use of utility-function to compare ‘desirability’ -> many actions might satisfy a goal, but which is more desirable ?
• Where outcome of goals is uncertain, utility enables agent to evaluate goal importance against likelihood of success.