Chapter 2: Intelligent Agents Flashcards
(159 cards)
1
Q
What is an agent?
A
An agent is anything that can be viewed as perceiving its environment through sensors and acting upon that environment through actuators.
Examples include humans, robotic agents, and software agents.
2
Q
What do sensors and actuators refer to in the context of an agent?
A
Sensors are organs or devices that perceive the environment, while actuators are organs or devices that act upon the environment.
Examples of sensors include eyes and cameras; examples of actuators include hands and motors.
3
Q
What is a percept?
A
A percept refers to the content an agent’s sensors are perceiving.
4
Q
Define percept sequence.
A
An agent’s percept sequence is the complete history of everything the agent has ever perceived.
5
Q
What influences an agent’s choice of action?
A
An agent’s choice of action can depend on its built-in knowledge and the entire percept sequence observed to date.
6
Q
What is an agent function?
A
An agent function maps any given percept sequence to an action.
7
Q
What is the significance of tabulating the agent function?
A
Tabulating the agent function provides an external characterization of the agent’s behavior.
8
Q
What is the difference between an agent function and an agent program?
A
The agent function is an abstract mathematical description, while the agent program is a concrete implementation running within a physical system.
9
Q
What is the purpose of the notion of an agent?
A
The notion of an agent is meant to be a tool for analyzing systems.
10
Q
Can a hand-held calculator be considered an agent?
A
Yes, it can be viewed as an agent that chooses actions based on its percept sequence.
11
Q
How do the authors categorize AI in relation to engineering?
A
AI operates at the most interesting end of the spectrum, where artifacts have significant computational resources and require nontrivial decision making.
12
Q
What is a rational agent?
A
An agent that does the right thing
13
Q
What notion of the ‘right thing’ does AI generally follow?
A
Consequentialism
14
Q
How do we evaluate an agent’s behavior in AI?
A
By its consequences
15
Q
What does a sequence of actions generated by an agent cause?
A
A sequence of states in the environment
16
Q
What indicates that an agent has performed well?
A
If the sequence of states is desirable
17
Q
How does the notion of rationality apply to humans?
A
It relates to their success in choosing desirable actions
18
Q
What is the source of the performance measure for machines?
A
The mind of the designer or users
19
Q
What is a characteristic of some agent designs regarding performance measures?
A
They have an explicit representation of the performance measure
20
Q
How might some agents perform well without understanding why?
A
The performance measure is implicit
21
Q
What warning did Norbert Wiener give regarding machines?
A
Ensure the purpose put into the machine is the purpose we really desire
22
Q
What is a potential issue with formulating a performance measure?
A
It can be hard to formulate correctly
23
Q
What is a flawed performance measure example for a vacuum-cleaner agent?
A
Measuring by the amount of dirt cleaned in a shift
24
Q
What is a better performance measure for a vacuum-cleaner agent?
A
Rewarding for having a clean floor
25
What could be a point system for a vacuum-cleaner agent?
One point for each clean square, penalties for electricity and noise
26
What is the general rule for designing performance measures?
Design according to what is actually desired in the environment
27
What philosophical question arises from the concept of 'clean floor'?
Which cleaning method is preferable—mediocre or energetic with breaks?
28
What is the King Midas problem in machine design?
Putting the wrong purpose into the machine
29
What challenge arises when designing software for different users?
Anticipating the exact preferences of each individual user
30
What type of agents may need to learn about the true performance measure over time?
Agents that reflect initial uncertainty about the performance measure
31
What are the four factors that determine what is rational at any given time?
* The performance measure that defines the criterion of success
* The agent’s prior knowledge of the environment
* The actions that the agent can perform
* The agent’s percept sequence to date
## Footnote
These factors help define the rationality of an agent's actions in a given context.
32
Define a rational agent.
A rational agent should select an action that is expected to maximize its performance measure for each possible percept sequence, given its percept sequence and built-in knowledge.
## Footnote
This definition emphasizes the importance of decision-making based on available information.
33
Is the vacuum-cleaner agent rational under certain assumptions?
Yes, if it correctly perceives its location and whether it contains dirt, it is rational.
## Footnote
Its performance is at least as good as any other agent’s under these conditions.
34
Under what circumstances could the vacuum-cleaner agent become irrational?
If the performance measure includes a penalty for each movement, the agent may oscillate back and forth unnecessarily once all dirt is cleaned.
## Footnote
This highlights how the definition of rationality can change based on the performance measure.
35
True or False: Rationality is the same as omniscience.
False
## Footnote
Rationality does not require omniscience; it is based on the current percept sequence.
36
What is the difference between rationality and perfection?
Rationality maximizes expected performance, while perfection maximizes actual performance.
## Footnote
This distinction is crucial for understanding agent design.
37
What is an example of an unintelligent activity in agent behavior?
Crossing a busy road without looking both ways.
## Footnote
This behavior would not maximize expected performance due to the risk involved.
38
What is the importance of information gathering for a rational agent?
It helps maximize expected performance by modifying future percepts.
## Footnote
Information gathering is essential for adapting to changing environments.
39
What happens to an agent if it relies too much on prior knowledge rather than its own learning?
It lacks autonomy.
## Footnote
A rational agent should learn from its experiences to compensate for any limitations in prior knowledge.
40
How can initial knowledge assist an artificial intelligent agent?
It helps the agent act effectively in environments where it has little or no experience.
## Footnote
This approach is similar to how evolution provides animals with reflexes to survive.
41
What is the outcome of incorporating learning into a rational agent's design?
The agent can become effectively independent of its prior knowledge after sufficient experience.
## Footnote
This allows the agent to adapt to a variety of environments successfully.
42
Fill in the blank: A rational agent should be _______.
[autonomous]
## Footnote
Autonomy in agents allows them to learn and adapt beyond their initial programming.
43
What does PEAS stand for in the context of task environments?
Performance, Environment, Actuators, Sensors
44
What are task environments?
Are essentially the “problems” to which rational agents are the “solutions.”
45
What is the first step in designing an agent?
Specify the task environment as fully as possible
46
Name one desirable quality for an automated taxi driver’s performance measure.
Getting to the correct destination
## Footnote
Other qualities include minimizing fuel consumption, minimizing trip time, maximizing safety, and maximizing profits.
47
What types of roads might an automated taxi driver encounter?
Rural lanes, urban alleys, 12-lane freeways
48
What are some factors that complicate the driving environment for an automated taxi?
Other traffic, pedestrians, stray animals, road works, police cars, puddles, potholes
49
What are the actuators for an automated taxi?
Control over the engine, steering, braking, display screen output, voice synthesizer
50
List three basic sensors that an automated taxi should have.
* Video cameras
* Lidar sensors
* Ultrasound sensors
51
What additional sensor is needed for speed control in an automated taxi?
Speedometer
52
What sensor helps determine the mechanical state of the vehicle?
Engine, fuel, and electrical system sensors
53
What type of input might a passenger use to request a destination in an automated taxi?
Touchscreen or voice input
54
True or False: Virtual task environments can be less complex than real-world environments.
False
55
What is an example of a virtual agent mentioned in the text?
Software agent that trades on auction and reselling websites
56
What is the purpose of specifying the performance measure in a task environment?
To define the goals and success criteria for the agent
57
Fill in the blank: The nature of the task environment directly affects the appropriate design for the _______.
agent program
58
What are the dimensions used to categorize task environments in AI?
The dimensions include:
* Fully observable vs. partially observable
* Single-agent vs. multiagent
* Deterministic vs. nondeterministic
* Episodic vs. sequential
* Static vs. dynamic
* Discrete vs. continuous
* Known vs. unknown
## Footnote
These dimensions determine agent design and the applicability of techniques for agent implementation.
59
What defines a fully observable environment?
An environment is fully observable if an agent’s sensors provide access to the complete state of the environment at each point in time.
## Footnote
This means that the sensors detect all aspects relevant to the choice of action.
60
What is a partially observable environment?
A partially observable environment occurs when sensors are noisy, inaccurate, or missing data, preventing the agent from accessing the complete state.
## Footnote
An example is a vacuum agent with a local dirt sensor that cannot detect dirt in other squares.
61
What distinguishes single-agent environments from multiagent environments?
Single-agent environments involve one agent, while multiagent environments involve multiple agents that may compete or cooperate.
## Footnote
For instance, solving a crossword puzzle is single-agent, while chess is multiagent.
62
What is the key distinction between competitive and cooperative multiagent environments?
In competitive environments, agents aim to maximize their performance at the expense of others, while in cooperative environments, agents work towards common goals.
## Footnote
Chess is competitive, and taxi driving can be seen as partially cooperative.
63
What defines a deterministic environment?
A deterministic environment is one where the next state is completely determined by the current state and the action taken.
## Footnote
In contrast, a nondeterministic environment has outcomes that are not strictly predictable.
64
What is the difference between stochastic and nondeterministic environments?
Stochastic environments explicitly deal with probabilities, while nondeterministic environments list possibilities without quantifying them.
## Footnote
For example, 'there's a 25% chance of rain tomorrow' is stochastic.
65
Define episodic task environments.
Episodic environments divide the agent’s experience into atomic episodes, where each episode's outcome does not depend on previous actions.
## Footnote
An example is a quality control agent checking parts on an assembly line.
66
What characterizes sequential task environments?
In sequential environments, the current decision can affect future decisions, requiring the agent to think ahead.
## Footnote
Examples include chess and taxi driving.
67
What defines a dynamic environment?
A dynamic environment changes while the agent is deliberating, requiring ongoing attention.
## Footnote
Taxi driving is dynamic due to the movement of other vehicles.
68
What is the distinction between discrete and continuous environments?
Discrete environments have a finite number of distinct states, while continuous environments allow states to vary smoothly over time.
## Footnote
Chess is discrete, whereas taxi driving is continuous.
69
What does it mean for an environment to be known?
A known environment provides the agent with complete information about the outcomes of all actions.
## Footnote
In contrast, an unknown environment requires the agent to learn how it operates.
70
How can an environment be partially observable yet known?
An environment can be partially observable if certain information is hidden, even if the rules are understood.
## Footnote
An example is solitaire, where the player knows the rules but not all card positions.
71
What is a semidynamic environment?
A semidynamic environment does not change with time, but the agent's performance score can vary.
## Footnote
Chess with a clock is an example, as time affects the player's performance.
72
What are some characteristics of the hardest task environments?
The hardest environments are partially observable, multiagent, nondeterministic, sequential, dynamic, continuous, and unknown.
## Footnote
Taxi driving is often cited as a challenging environment in these respects.
73
Why is the medical-diagnosis task considered single-agent?
The disease process in a patient is not modeled as an agent, making it a single-agent task.
## Footnote
However, it may have multiagent aspects due to interactions with patients and staff.
74
What is the formula that represents an agent?
agent = architecture + program
75
What must the program chosen for an agent be appropriate for?
the architecture
76
What does the architecture of an agent include?
physical sensors and actuators
77
What is the primary function of the agent architecture?
makes the percepts from the sensors available to the program
78
What is the job of AI in relation to agents?
design an agent program that implements the agent function
79
How does the agent program differ from the agent function?
The agent program takes the current percept as input; the agent function may depend on the entire percept history.
80
What is a significant issue with the table-driven approach to agent construction?
The size of the lookup table becomes impractical.
81
What is the estimated number of entries in a lookup table for an automated taxi after one hour of driving?
over 10⁶⁰⁰,⁰⁰⁰,⁰⁰⁰,⁰⁰⁰,⁰⁰⁰ entries
82
How many entries does the lookup table for chess have?
at least 10¹⁵⁰ entries
83
What is the function of the REFLEX-VACUUM-AGENT?
returns an action based on location and status
84
What is the key challenge for AI in agent program design?
to write programs that produce rational behavior from a small program rather than a vast table
85
What is an example of a successful simplification in another field mentioned in the text?
replacement of huge tables of square roots with a five-line program for Newton's method
86
What are the four basic kinds of agent programs outlined?
* Simple reflex agents
* Model-based reflex agents
* Goal-based agents
* Utility-based agents
87
What is a simple reflex agent?
An agent that selects actions based on the current percept, ignoring percept history.
88
What is an example of a simple reflex agent?
The vacuum agent.
89
What is a condition–action rule?
A rule that connects a condition to an action, e.g., if car-in-front-is-braking then initiate-braking.
90
True or False: Simple reflex agents are capable of handling partially observable environments effectively.
False.
91
What happens if a simple reflex agent operates in a partially observable environment?
It may enter infinite loops or make incorrect decisions.
92
What is one way to escape from infinite loops in simple reflex agents?
By randomizing actions.
93
What is a potential problem for a simple reflex vacuum agent with only a dirt sensor?
It may fail to move correctly if it starts in a specific square.
94
What is the average number of steps for a randomised simple reflex agent to reach the other square for the vacuum agent?
Two steps.
95
What is a limitation of simple reflex agents?
They have limited intelligence and require fully observable environments.
96
What does the INTERPRET-INPUT function do in the general program for condition–action rules?
Generates an abstracted description of the current state from the percept.
97
What is the purpose of the RULE-MATCH function?
Returns the first rule in the set that matches the given state description.
98
True or False: The implementation of condition–action rules can only be done using if-then-else statements.
False.
99
What can replace logic gates in a neural circuit?
Nonlinear units of artificial neural networks.
100
List the two types of connections humans have that are similar to condition–action rules.
* Learned responses
* Innate reflexes
101
What is a more general approach than a specific agent program for a vacuum environment?
Building a general-purpose interpreter for condition–action rules.
102
What is the most effective way to handle partial observability?
The agent should keep track of the part of the world it can’t see by maintaining an internal state that reflects unobserved aspects of the current state.
## Footnote
This involves using percept history to inform the agent's understanding of its environment.
103
What does the internal state for the braking problem consist of?
The previous frame from the camera.
## Footnote
This allows the agent to detect simultaneous changes, such as red lights going on or off.
104
What additional information does an agent need for tasks like changing lanes?
The location of other cars that may not be visible all at once.
## Footnote
This is crucial for safe navigation and decision-making.
105
What must an agent keep track of for driving to be possible?
The location of its keys.
## Footnote
This is a basic requirement for operating the vehicle.
106
What are the two kinds of knowledge required to update the internal state?
* Effects of the agent’s actions
* How the world evolves independently of the agent
107
What is the transition model of the world?
Knowledge about how the world changes over time, including how actions affect the environment.
## Footnote
This can be implemented in various forms, from Boolean circuits to complex scientific theories.
108
What is the sensor model?
Knowledge about how the state of the world is reflected in the agent’s percepts.
## Footnote
For example, changes in the camera image when objects in front of the vehicle change.
109
What is a model-based agent?
An agent that uses transition and sensor models to keep track of the state of the world.
## Footnote
This enhances its ability to make informed decisions despite partial observability.
110
What does the function UPDATE-STATE do in a model-based reflex agent?
It creates a new internal state description by combining the current percept with the old internal state.
## Footnote
This is crucial for maintaining an accurate representation of the environment.
111
What is often the limitation faced by agents in partially observable environments?
It is seldom possible for the agent to determine the current state exactly.
## Footnote
Agents must often work with their best guesses about the state of the environment.
112
True or False: An automated taxi can always see around obstacles like large trucks.
False.
## Footnote
It may have to guess about the situation beyond its immediate view.
113
What does the box labeled 'what the world is like now' represent?
The agent’s best guess about the current state of the world.
## Footnote
This reflects the uncertainty inherent in partially observable environments.
114
What is a goal-based agent?
An agent that requires goal information to decide actions that achieve desired situations.
115
What are the two subfields of AI devoted to finding action sequences that achieve an agent's goals?
* Search
* Planning
116
How does decision making in goal-based agents differ from reflex agents?
Goal-based agents consider the future and the consequences of actions, while reflex agents operate on predefined rules without understanding the reasons.
117
What makes goal-based agents more flexible than reflex agents?
Knowledge supporting decisions is explicitly represented and can be modified easily.
118
Fill in the blank: Goals alone are not enough to generate high-quality behavior in most environments; a more general performance measure is called _______.
[utility]
119
What does a utility function represent in the context of an agent?
An internalization of the performance measure that helps maximize the expected utility.
120
In what two cases are goals inadequate for rational decision making?
* Conflicting goals
* Uncertain goals
121
What does a rational utility-based agent choose to maximize?
The expected utility of the action outcomes.
122
What is the significance of the expected utility in decision making?
It allows the agent to weigh the likelihood of success against the importance of goals.
123
True or False: A utility-based agent can operate without a specific utility function.
True
124
What challenges do utility-based agents face in real-world applications?
Modeling and keeping track of the environment, perception, representation, reasoning, and learning.
125
What is often unachievable in practice for utility-based agents due to computational complexity?
Perfect rationality
126
What must a utility-based agent do to make rational decisions?
Model its environment and choose the utility-maximizing course of action.
127
How is decision making in multiagent environments studied?
Within the framework of utility theory.
128
What does the structure of a utility-based agent include?
An internal utility function that guides decision making.
129
What is a learning agent?
An agent that can improve its performance based on feedback and experiences
## Footnote
Learning agents can be model-based, goal-based, utility-based, etc.
130
Who proposed the idea of building learning machines?
Turing in 1950
## Footnote
Turing suggested that programming intelligent machines by hand was labor-intensive and a more efficient method was needed.
131
What are the four conceptual components of a learning agent?
* Learning element
* Performance element
* Critic
* Problem generator
132
What is the role of the learning element in a learning agent?
Responsible for making improvements to the agent's performance
133
What does the performance element do?
Selects external actions based on percepts
134
What is the function of the critic in a learning agent?
Tells the learning element how well the agent is performing against a fixed performance standard
135
True or False: The performance standard can be modified by the agent.
False
## Footnote
The performance standard must remain fixed and outside the agent's control.
136
What is the purpose of the problem generator?
Suggests exploratory actions that may lead to new and informative experiences
137
Fill in the blank: The learning element can modify any of the _______ components of the agent.
knowledge
138
How can an agent learn about its actions and the world's response?
By observing pairs of successive states of the environment
139
Why is it sometimes better to have a simple model rather than a complex one?
A simple model can be more computationally efficient
140
What does the external performance standard help the agent learn in terms of utility?
It informs the agent about negative contributions to its performance
141
How can human behavior provide information to a learning agent?
It can indicate human preferences based on their reactions to the agent's actions
142
In summary, what does learning in intelligent agents involve?
Modification of each component to align with available feedback information
143
What is the main function of agent programs?
To answer questions about the environment and actions.
144
What are the three types of representations based on complexity and expressive power?
* Atomic
* Factored
* Structured
145
Define atomic representation.
Each state of the world is indivisible and has no internal structure.
146
What is an example of a task that can use atomic representation?
Finding a driving route from one end of a country to the other.
147
What do standard algorithms like search, game-playing, hidden Markov models, and Markov decision processes work with?
Atomic representations.
148
Define factored representation.
Splits each state into a fixed set of variables or attributes, each with a value.
149
What additional attributes might be included in a factored representation for driving?
* Amount of gas in the tank
* Current GPS coordinates
* Oil warning light status
* Money for tolls
* Radio station
150
How do factored states differ from atomic states?
Factored states can share attributes, while atomic states cannot.
151
What important AI areas use factored representations?
* Constraint satisfaction algorithms
* Propositional logic
* Planning
* Bayesian networks
* Machine learning algorithms
152
What is structured representation?
Describes objects and their relationships explicitly.
153
What is an example of a situation that requires a structured representation?
A truck reversing while a loose cow blocks its path.
154
What underlies relational databases and first-order logic?
Structured representations.
155
What is the axis of increasing expressiveness?
A more expressive representation can capture everything a less expressive one can, plus more.
156
What is localist representation?
A one-to-one mapping between concepts and memory locations.
157
What is distributed representation?
Concept representation spread over multiple memory locations.
158
What is a key advantage of distributed representations?
More robust against noise and information loss.
159
What happens to a concept in a distributed representation if a few bits are garbled?
It moves to a nearby point in multidimensional space with similar meaning.
