A.I. Chapter 2 rational agents
Multi-agent
B's behavior maximizes a performance measure whose value depends on agent A's behavior. Communication emerges as rational behavior.
Model
Based on information about how the world evolves independently of the agent
Goal-based agent
Can combine some sort of goal information of desirable situations and the model to choose actions to achieve the goal. Act to achieve their goals.
Sequential
Current decision could affect all future decisions
Information Gathering
Doing actions in order to modify future percepts
Partially Observable
Due to noisy and inaccurate sensors or because parts of the state are simple missing from the sensor data
Rational agent
For each possible percept sequence, it should select an action that is expected to maximize its performance measure, given the evidence provided by the percept sequence and whatever built-in knowledge it has
Unobservable
If an agent has no sensors at all
Fully Observable
If an agent's sensors give it access to the complete state of the environment at each point in time (relevant to the choice of action, depending on performance measure)
Autonomy
An agent should learn what it can to compensate for partial or incorrect prior knowledge
Non-deterministic
An environment in which actions are characterized by their possible outcomes but no probabilities are attached to them Usually associated with performance measures that require the agent to succeed for all possible outcomes of its actions.
Utility function
An internalization of the performance measure.
True
Any rational agent must behave as if it posses a utility function whose expected value it tries to maximize.
Rationality
1. The performance measure that defines the criterion of success. 2. The agent's prior knowledge of the environment 3. The actions that the agent can perform 4. The agent percept sequence to date
Uncertain
If an environment is not fully observable or deterministic
Dynamic
If the environment can change while the agent is deliberating
Static
If the environment does not change while the agent is deliberating
Discrete
If the environment has a finite set of states
Deterministic
If the next state of the environment is completely determined by the current state and action executed by the agent
Simple reflex agent property
If will work only if the correct decision can be made on the basis of only the current percept - that is, only if the environment is fully observable
Agent Program
Implements agent function
Randomization
In a single-agent environment, _________ is usually not rational.
Critic
Provides feedback to learning element on how the agent is doing and determines how the performance element should be modified to do better in the future.
Learning element
Responsible for making improvements
Performance element
Responsible for selecting external action
Task Environment
The "problems" to which rational agents are the "solutions"
Model-based reflex agent
The agent maintains internal state that depends on the percept history and thereby reflects at least some of the unobserved aspects of the current states Maintain internal state to track aspects of the worlds that are not evident in the current state.
Unknown
The agent will have to learn how it works in order to make good decisions. Could be fully observable
Episodic
The agent's experience is divided into atomic episodes, and in each, the agent receives a percept and then performs a single action.
Known
The outcome for all possible states should be given. Could be partially observable
Continuous
There is a range of states that it could go through
Simple reflex agent
These agents select the actions on the basis of the current percept, ignoring the rest of the percept history
In some competitive environments, randomized behavior is rational because it avoids the pitfalls of predictability.
True
Utility-based agent
try to maximize their own expected "happiness"