Early on, it became clear that agents needed to learn:
- Describing all possible situations is not feasible
- Environments might change or be unknown
Turing was already describing learning agents in 1950
#Conceptual Components
#Learning Element
- responsible for making improvements
- Improvements can be made in all parts of the model
#Performance Element
- responsible for selecting actions
- Usually designed first, affects the design of the learning element
#Critic
- provides feedback to the learning element
- Uses an external performance standard to measure current performance
#Problem Generator
- suggesting new actions
- Sometimes actions will be taken to gather information, rather than do the known good thing
Goal-based Agents