AI Exam Questions
agent function
a mathematical way of describing an agent's behavior that maps any given percept sequence to an action
goal
information that describes situations that are desirable
What is the hardest case?
partially observable, multiagent, stochastic, sequential, dynamic, continuous, and unknown
What does an agent's choice of action depend on?
the entire percept sequence observed to date, but not on anything it hasn't perceived.
atomic vs factored representation
Factored representation more easily represents how to reach from one state to another, and also represent uncertainty through the sudden absence of a variable/attribute
Is randomization better in single or multiagent environments?
Multiagent, to avoid predictability in competitive environments.
When does randomizing actions help an agent?
When the agent is stuck in an infinite loop
Is it possible for an environment to be unknown but fully observable?
Yes, as video games can give you the full state of things... but a new player might not know the controls.
Is it possible for an environment to be known but be partially observable?
Yes, such as in games of chance, the rules are known, but not the outcome of obscured cards.
simple reflex agent
agents that select actions on the basis of the current percept and nothing more and uses condition-actions rules
learning agent
an agent that can operate in initially unknown environments and become more competent than its initial knowledge alone might allow
rational agent
an agent that selects an action that is expected to maximize its performance measure, given the evidence provided by the percept sequence and whatever built-in knowledge the agent has
model-based agent
an agent that uses models
percepts
an agent's perceptual inputs at any given instant
nondeterministic environment
an environment in which actions are characterized by their possible outcomes, but the probability of them happening is not taken into consideration.
uncertain environment
an environment that is not fully observable or not deterministic
performance measure
an evaluation of any given sequence of environment states
agent
anything that can be viewed as perceiving its environment through sensors and acting upon that environment through actuators
ways to represent state
atomic, factored or structured
condition-action rule
basically an if -> then statement
single agent vs multiagent
broad and shallow description of if there are multiple agents performing in the environment or not
utility
comparison of different world states according to exactly how "happy" they would make the agent
critic
conceptual component of learning agent responsible for giving feedback to the learning element (with respect to a fixed performance standard) on how the agent is doing and determines how the performance element should be modified (by the learning element) to do better
learning element
conceptual component of learning agent responsible for making improvements
performance agent
conceptual component of learning agent responsible for selecting external actions - basically an agent without the ability to get better
problem generator
conceptual component of learning agent responsible for suggesting actions that will lead to now and informative experiences i.e. what causes the agent to explore and take risks
competitive environment
each agent in the environment is trying to maximize their performance measure which means minimizing other agents' performance measure if possible
goal vs utility
goals tend to be binary, which utility (happiness) can be based on a scale
actuator
how an agent interacts with its environment
sensor
how an agent perceives its environment
internal state
how an agent would depend on its percept history to reflect on at least some of the unobserved aspects of the current state
agent program
how the agent goes through mapping out and choosing an action
semidynamic environment
is the environment does not chance with the passage of time, but an agent's performance measure score does
atomic representation of state
it's a black box - each state of the world is indivisible with no internal structure
fully observable vs partial observable environment
knowing the complete state of the environment (as relevant to actions satisfying the performance measure) via an agent's sensors or not
model (of the world)
knowledge of how the environment works (What rules the agent knows about the environment)
conceptual components of a learning agent
learning element, performance element, critic, problem generator
autonomy
not relying on the prior knowledge of one designer and rather on ones own percepts
PEAS
performance, environment, actuators, sensors
discrete vs. continuous environment
referring to how these words are applied to describe the state of the environment, how time is handled and to the precepts and actions of the agent All three aspects can be one or the other.
example of atomic representation of state
search and game playing algorithms
factored representation of state
splits up each state into a fixed set of variables or attributes which have some value - these variables can be shared between states
what does a nondeterministic environment normally have the agent do?
succeed for all possible outcomes of its actions
task environment
the PEAS
agent program vs agent function
the agent program is a concrete implementation (something actually running in the agent) while the agent function is an abstract mathematical description
percept sequence
the complete history of everything the agent has ever perceived
architecture
the computing device (which physical sensors and actuators) that the agent program runs on
evironment
the situation a agent is in
cooperative environment
where all agents take actions to improve everyone's performance measures
episodic vs sequential environments
whether an agent's experience is divided into atomic episodes in which the agent receives a percept and performs a single action and each episode does not depend on the actions taken in previous episodes i.e. decisions/episodes do not effect anything/have long-standing consequences
known vs unknown environment
whether or not the agent is aware of the rules that define the environment (i.e. the laws of physics governing said environment).
deterministic vs stochastic
whether the entire state of the environment is the result of an agent's actions on the environment or not
dynamic or static environments
whether the environment can change while an agent is deliberating or not
