Week 1 (23) Flashcards
agent
-Perceives environment through sensors
-acts on environment through actuators
Percept
Agents perceptual input at any given instant
Percept sequence
Complete history of everything the agent has ever perceived
Agent function
Describes an agents behaviour
Maps any given Percept sequence to an action
Agent program
Concrete implementation of an agent function running within some physical system
(Check if this is exact as textbook says ‘it is important to keep these two ideas distinct’)
Rational agent
For each possible Percept sequence,
a rational agent should select an action that is expected to maximise its performance measure,
given the evidence provided by the Percept sequence and whatever built in knowledge the agent has
Performance measure
Something that evaluate any given sequence of environment states (not agent states)
4 factors that determine rationality
Performance measure defining criterion of success
Agents prior knowledge of env
Actions that agent can perform
Agents Percept sequence to date
Omniscience
Agent knows actual outcome of actions and can act accordingly
Information gathering
Doing actions in order to modify future Percepts
Exploration
A type of information gathering in which the search space is inspected
Autonomy
Relying on own precepts over prior knowledge
Task environment
PEAS
Performance
Environment
Actuators
Sensors
Softbots
Software agents
Software robots
Fully observable v partial observable
If an agents sensors give it access to the complete state of the env at each point in time
Example of emergent behaviour in competitive environment
Randomised behaviour (to avoid being predictable)
Example of emergent behaviour in cooperative environment
Communication
Deterministic env
If next state of env is completely determined by the current state and the action executed by the agent
When can an environment appear to be stochastic
When it is partially observable
Uncertain environment
Not fully observable
Not deterministic
Distinction between non deterministic environment and stochastic environment
Non determinism: actions are characterised by possible outcomes
Stochastic: same as above + probabilities associated to possible outcomes
Episodic env v sequential env
Episodic: the actions taken in a previous episode do not effect subsequent episodes
Episode: agent receives a Percept then performs a single action
Discrete env v Continuous env
Refers to handling of time
Discrete set of percepts and actions
Known env v unknown env
Known: outcomes for all actions are given
Unknown: agent must learn how env works to make good decisions
Example of distinction between (un)known and partially/fully observable
Solitaire:
Rules known
Exact cards not yet turned over
Environment generator
Selects particular environments from environment class with given likelihoods to evaluate agent
Agent =
Architecture + program
Difference between agent function and agent program inputs
Program takes just current Percept as input
Function taken entire Percept history
Simple reflex agent
Selection actions based on current Percept ignoring history
How can a simple reflex agent escape a loop
If the agent can randomise actions
Internal state
For model based reflex agent
Depends on Percept history and reflects the data that an agent maintains internally of environment
2 pieces of info needed by model based agent
Information about how world evolves independent of agent
Information about how agents actions effect world
Utility function
An agents internalisation of the performance measure
Difference between performance measure and utility function
Performance measure is objective measurement of environment not done by agent
Rational utility based agent chooses the action that maximises
EXPECTED utility
