2024年4月1日发(作者：司滢滢)
Chapter 1
1.1 Define in
yo盯own
word: (a) intelligence
,
(b)
artificial intelligence
,
(c) agen
t.
•
lntelligence在fíit:
Dictionary defmitions of intelligence talk about "the capacity to acquire and apply
knowledge" or "the faculty
of thought and reason" or "the
ab诅tyto∞mprehend
and
profit仕om
experience." These are all reasonable answers
,
but
if
we want
some也ing
quantifiable we would use
something like
"the ability to apply knowledge in order to perform better in an environmen
t."
智能的字典定义有一种学习或应用知识的能力，一种思考和推理的本领，领会并且得益于经验的能
力，这些都是有道理的答案，但如果我们想量化一些东西，我们将用到一些东西像为了在环境中更
好的完成任务使能力适应知识
•
Artific切1intelligenceλ工在f
íili:
We define
art出cial
intelligence as the study and construction of agent
programs阳t
perform well in a given environment
,
for a given agent
archit民ture.
作为一学习和构造智能体程序，为了一个智能体结构，在被给的环境中可以很好的完成任务。
• Agen
1Jf
íiIi体t:
We define an agent as an
entity实体that
takes action in response to percepts from an
environment.在一个环境中对一个对象做出反应的实体
1.4
Th
ere are well-known classes of problem that are intractably difficnlt for computers
,
and other
classes
that are provably undecidable. Does this mean that AI is impossible?
No. It
means也at
AI systems should
avoid位ying
to solve intractable problems.
Usually，由ism巳ans
they can
only approximate optimal behavi
or. Notice that humans don't solve NP complete problems either. Sometimes
由ey
are good at solving
spec诅c
instances
wi也a
lot of structure
,
perhaps with the aid of background
knowledge. AI systems should attempt to do
the same.
1.11
"surely computers cannot be intelligent-they can do only what their programmers tell them." Is
the latter statement true
,
and does it imply the former?
This depends
on your definition of "intelligen
t"
and "tell."
In
one sense computers
0世y
do
what自己
programmers command them to do
,
but in another sense
what也e
programmers consciously tells the
computer to
do often has very little to do with what the computer actually
do巳s.
An
yone who has written a
program wi
th an
orneηbug
knows this
,
as does anyone who has written a successful machine learning
program.
So in one sense Samuel
"told"也e
computer "learn to play checkers better than 1 do
,
and then play
that way
,"
but in another sense he told the computer "follow this learning algorithm" and it learned to play.
So
we're left in the situation where you may or may not consider learning to play checkers to be s sign of
intelligence (or you
may由ink白at
learning to play
in也e
right way
requ让es
intelligence
,
but not
in也is
way)
,
andyou
may吐血汰出e
intelligence resides in the programmer or in the computer
Chapter 2
2.1 Defme in
yo町own
words the following terms: agent
,
agent function
,
agent program
,
rationality
,
reflex agent
,
model-based agent
,
goal-based agent
,
utility-based agent
,
learning agent.
Th
e following are just some of the many possible defmitions that can be written:
•
Agentli'舷佯:
an entity
(实体)
tbat perceives
(感知)
and
acω行为;
or
,
one that can be viewed
as
perceiving and acting.
Essentially本质上any
object
qualifies限定;
the key point is the way the object
implements an agent function.
(N
ote: some autbors restrict the term to programs that operate on behaif of a
buman
,
or to programs that can cause some or
all
of their code to run on other machines on a networ
k,
as in
mobile agents. MOBILE AGENT)
一个具有感知和行文的实体，或者是一个可以观察到感觉的实体，本质上，任何限定对象，只要的观
点是一种对象执行智能体函数的方法。(注意，一些作者〉
可以感知环境，并在环境中行动的某种东西。
• Agent
function暂黯体函数:
a
function也at
specifies the agent' s action in
response归巳very
possible
p町ceptsequence智能体相应任何感知序列所采取的行动
• Agent program
tf:
f
fif体程序:
that program whicb
,
combined with a machine architecture
,
implements an
agent function.
In
0世simple
designs
,
tbe program takes a new percept on eacb invocation and returns an
ac挝on.实现了智能函数。有各种基本的智能体程序设计，反应出现实表现的一级用于决策过程的信息
种类。设计可能在效率、压缩性和灵活性方面有变化。适当的智能体程序设计取决于环境的本性
•
Rationali句;王军放:
a property of agents that choose actions that maximize tbeir expected
u创坷，
given the
percepts to date.
•
Autonomy
fJ主:
a property of
agenωwhose
bebavior is determined by tbeir own experience rather than
solely by their initial programming.
.R伪'xagent反射却在FSE体:
an agent whose action depends only on the current percept.
一个智能体的行为仅仅依赖于当前的知觉。
• Model-based
agent基于茹苦型的主FifS体:
an agent wbose actioD is derived directly from an internal model
ofthe
c田rent
world state that is updated over time.
一个智能体的行为直接得自于内在模型的状态，这个状态是当前世界通用的不断更新。
• Goal-based
agen基'FfU萃的苟能俐:
an agent that selects actions that it believes will acbieve explicitly
represented
goals.智能体选择它相信能明确达到目标的行动。
•
Utility-bωed agen基于效用的主F磁仰:
an agent that
sel饵ts
actions that it
bel即es
will
maximize也e
expected
util即of
the outcome
state.试图最大化他们自己期望的快乐
• Learning
agent学习智能做:
an agent whose behavior improves over time based on its experience.
2.2 Both the performance
m阅sure
and the utility function
me描盯e
how well an agent
is
doing.
Explain the difference between the two.
A performance
measure
(性能度量)
is used by an outside observer to evaluate
(评估)
bow successful an
agent is. It is a function from bistories to a real numbe
r.
A
ut诅ity
function
(效用函数)
is used by an agent
its
elf
ωevaluate
how desirable
(令人想要)
states or bistories are.
In
our framewor
k,
the
ut让ity
function
m
ay not be the same as the performance measure;
fl町由ermore，
an agent may have no
expliαt u创ity
function at all
,
whereas there is always a
perfo口nance
measure.
following agents
,
develop a PEAS description of the task environment:
2.5
For each of
a. Robot soccer player;
b.
In
ternet book-shopping agent;
c. Autonomous Mars rover;
d.
Mathema创cian's
theorem-proving assistant.
Some representative
,
but not
e对laustive，
answers are given in Figure S2.
1.
2024年4月1日发(作者：司滢滢)
Chapter 1
1.1 Define in
yo盯own
word: (a) intelligence
,
(b)
artificial intelligence
,
(c) agen
t.
•
lntelligence在fíit:
Dictionary defmitions of intelligence talk about "the capacity to acquire and apply
knowledge" or "the faculty
of thought and reason" or "the
ab诅tyto∞mprehend
and
profit仕om
experience." These are all reasonable answers
,
but
if
we want
some也ing
quantifiable we would use
something like
"the ability to apply knowledge in order to perform better in an environmen
t."
智能的字典定义有一种学习或应用知识的能力，一种思考和推理的本领，领会并且得益于经验的能
力，这些都是有道理的答案，但如果我们想量化一些东西，我们将用到一些东西像为了在环境中更
好的完成任务使能力适应知识
•
Artific切1intelligenceλ工在f
íili:
We define
art出cial
intelligence as the study and construction of agent
programs阳t
perform well in a given environment
,
for a given agent
archit民ture.
作为一学习和构造智能体程序，为了一个智能体结构，在被给的环境中可以很好的完成任务。
• Agen
1Jf
íiIi体t:
We define an agent as an
entity实体that
takes action in response to percepts from an
environment.在一个环境中对一个对象做出反应的实体
1.4
Th
ere are well-known classes of problem that are intractably difficnlt for computers
,
and other
classes
that are provably undecidable. Does this mean that AI is impossible?
No. It
means也at
AI systems should
avoid位ying
to solve intractable problems.
Usually，由ism巳ans
they can
only approximate optimal behavi
or. Notice that humans don't solve NP complete problems either. Sometimes
由ey
are good at solving
spec诅c
instances
wi也a
lot of structure
,
perhaps with the aid of background
knowledge. AI systems should attempt to do
the same.
1.11
"surely computers cannot be intelligent-they can do only what their programmers tell them." Is
the latter statement true
,
and does it imply the former?
This depends
on your definition of "intelligen
t"
and "tell."
In
one sense computers
0世y
do
what自己
programmers command them to do
,
but in another sense
what也e
programmers consciously tells the
computer to
do often has very little to do with what the computer actually
do巳s.
An
yone who has written a
program wi
th an
orneηbug
knows this
,
as does anyone who has written a successful machine learning
program.
So in one sense Samuel
"told"也e
computer "learn to play checkers better than 1 do
,
and then play
that way
,"
but in another sense he told the computer "follow this learning algorithm" and it learned to play.
So
we're left in the situation where you may or may not consider learning to play checkers to be s sign of
intelligence (or you
may由ink白at
learning to play
in也e
right way
requ让es
intelligence
,
but not
in也is
way)
,
andyou
may吐血汰出e
intelligence resides in the programmer or in the computer
Chapter 2
2.1 Defme in
yo町own
words the following terms: agent
,
agent function
,
agent program
,
rationality
,
reflex agent
,
model-based agent
,
goal-based agent
,
utility-based agent
,
learning agent.
Th
e following are just some of the many possible defmitions that can be written:
•
Agentli'舷佯:
an entity
(实体)
tbat perceives
(感知)
and
acω行为;
or
,
one that can be viewed
as
perceiving and acting.
Essentially本质上any
object
qualifies限定;
the key point is the way the object
implements an agent function.
(N
ote: some autbors restrict the term to programs that operate on behaif of a
buman
,
or to programs that can cause some or
all
of their code to run on other machines on a networ
k,
as in
mobile agents. MOBILE AGENT)
一个具有感知和行文的实体，或者是一个可以观察到感觉的实体，本质上，任何限定对象，只要的观
点是一种对象执行智能体函数的方法。(注意，一些作者〉
可以感知环境，并在环境中行动的某种东西。
• Agent
function暂黯体函数:
a
function也at
specifies the agent' s action in
response归巳very
possible
p町ceptsequence智能体相应任何感知序列所采取的行动
• Agent program
tf:
f
fif体程序:
that program whicb
,
combined with a machine architecture
,
implements an
agent function.
In
0世simple
designs
,
tbe program takes a new percept on eacb invocation and returns an
ac挝on.实现了智能函数。有各种基本的智能体程序设计，反应出现实表现的一级用于决策过程的信息
种类。设计可能在效率、压缩性和灵活性方面有变化。适当的智能体程序设计取决于环境的本性
•
Rationali句;王军放:
a property of agents that choose actions that maximize tbeir expected
u创坷，
given the
percepts to date.
•
Autonomy
fJ主:
a property of
agenωwhose
bebavior is determined by tbeir own experience rather than
solely by their initial programming.
.R伪'xagent反射却在FSE体:
an agent whose action depends only on the current percept.
一个智能体的行为仅仅依赖于当前的知觉。
• Model-based
agent基于茹苦型的主FifS体:
an agent wbose actioD is derived directly from an internal model
ofthe
c田rent
world state that is updated over time.
一个智能体的行为直接得自于内在模型的状态，这个状态是当前世界通用的不断更新。
• Goal-based
agen基'FfU萃的苟能俐:
an agent that selects actions that it believes will acbieve explicitly
represented
goals.智能体选择它相信能明确达到目标的行动。
•
Utility-bωed agen基于效用的主F磁仰:
an agent that
sel饵ts
actions that it
bel即es
will
maximize也e
expected
util即of
the outcome
state.试图最大化他们自己期望的快乐
• Learning
agent学习智能做:
an agent whose behavior improves over time based on its experience.
2.2 Both the performance
m阅sure
and the utility function
me描盯e
how well an agent
is
doing.
Explain the difference between the two.
A performance
measure
(性能度量)
is used by an outside observer to evaluate
(评估)
bow successful an
agent is. It is a function from bistories to a real numbe
r.
A
ut诅ity
function
(效用函数)
is used by an agent
its
elf
ωevaluate
how desirable
(令人想要)
states or bistories are.
In
our framewor
k,
the
ut让ity
function
m
ay not be the same as the performance measure;
fl町由ermore，
an agent may have no
expliαt u创ity
function at all
,
whereas there is always a
perfo口nance
measure.
following agents
,
develop a PEAS description of the task environment:
2.5
For each of
a. Robot soccer player;
b.
In
ternet book-shopping agent;
c. Autonomous Mars rover;
d.
Mathema创cian's
theorem-proving assistant.
Some representative
,
but not
e对laustive，
answers are given in Figure S2.
1.