What does a simple agent need?
- A body
- Effectors
- Sensors
- Nervous system
- States
- An explicit set of tasks?
- An environment
Smart agents: a repertoire of actions
- Some essential actions
- Locomotion
- Eating
- Reproducing
- Approaching, pursuing food
- Avoiding, fleeing from predators
Smart agents: sensing/perceiving
- Getting information about the environment
- What objects there are
- Where the objects are
- How the objects are related to one another
- Getting information about the agent itself
- Raw sensory input contains too much information;
perception extracts useful information from sensory input:
percepts
- Touch, smell, audition, sonar, vision, proprioception
- Multiple sensors
Smart agents: associating percepts with actions in memory
- Genetic "memory"
- Long-term memory
- Cultural memory
- Implementation
- Lookup tables: store every percept/action combination;
given a percept, just look it up in the table
- Neural networks: store associations of percept features and action features; given a percept in the form of a set of features, look up the appropriate action in the form of a set of features
Even smarter agents: adapting
- Adaptation: changing behavior in response to feedback
- Learning
- Evolution
- Learning vs. evolution
- Learning + evolution