- Agent = Architecture + Program
- Agent architecture: computing device with physical/software sensors and actuators
- Job of AI: design an agent program
Basic agent Program: take an input from the sensors and return an action to the actuators.
Agent function can be explicitly represented as a table-driven agent
function TABLE-DRIVEN-AGENT(percept) returns an action
persistent: percepts, a sequence, initially empty
table, a table of actions, indexed by percept sequences, initially fully specified
append percept to the end of percepts
action ← LOOKUP(percepts, table)
return action
Table-driven approach does not work in practice:
- The table will be potentially infinite, but even in a limited lifetime agent, will likely be massive
- Filling in the table is by no means trivial, taking infinite time potentially
- Does not allow for learning, as the table is set from the start
E.g. for chess, there would be $10^{123}$ entries. For go, that number is $10^{360}$
Key challenge: Write minimal programs that, to the extent possible, produce rational behavior
#Vacuum Agent Example
sensors: current position, dirtiness actions: suck, right, left
function REFLEX-VACUUM-AGENT([location,status]) returns an action
if status = Dirty then return Suck
else if location = A then return Right
else if location = B then return Left
Environments