Solution Manual ArtificialIntelligence AModernApproach
computationalintelligence(UniversityofKarachi)
, Instructor’sManual:
ExerciseSolutions
for
ArtificialIntelligence
AModernApproach
SecondEdition
StuartJ.RussellandPeterNorvig
, SolutionsforChapter1
Introduction
1.1 a. Dictionary definitions of intelligence often refer to "the capacity to acquire and apply knowledge," "the faculty of
thought and reason," or "the ability to comprehend and profit from experience." While these definitions are reasonable, fo
quantifiable purposes, we might define intelligence as "the ability to apply knowledge in order to perform better in a give
environment."
b. Artificial intelligence is defined as the study and construction of agent programs designed to perform well within a specif
environment, given a particular agent architecture.
c. An agent is defined as an entity that takes action in response to percepts from its environment. See the solution for Exerc
26.1 for a discussion of potential objections.
The probability of fooling an interrogator depends on the skill level of the interrogator. While one entrant in the 2002 Loeb
prize competition did manage to fool one judge, it's unclear what that judge was thinking based on the transcript.
Nonetheless, there have been instances of chatbots or online agents fooling humans, such as the Julia chatbot. Today, we
estimate the chance of fooling an interrogator at around 10%, with variation primarily depending on the skill of the
interrogator rather than the program. In 50 years, with the significant investments expected in artificial actors by the
entertainment industry, it's likely that highly credible impersonators will emerge.
The 2002 Loebner prize was awarded to Kevin Copple's program ELLA. It consists of a prioritized set of pattern/action rule
utilizing a large database of text and the Wordnet online dictionary. While ELLA is effective within its domain, it does not
contribute significantly to advancing the theory of AI; it mainly provides evidence on the number and type of rules sufficie
for producing one type of conversation.
No, it means that AI systems should avoid attempting to solve intractable problems, as they can usually only approximate
optimal behavior. Humans also don't solve NP-complete problems; instead, they excel at solving specific instances with a l
of structure, often with the aid of background knowledge. AI systems should strive for similar approaches.
, No, while IQ test scores correlate well with certain measures like success in college, they don't measure everything. A progr
specialized only for IQ tests, particularly the analogy part, would likely perform poorly on other measures of intelligence. I
tests have limitations, and there are other aspects to "intelligence" beyond what they measure.
Just as you are unaware of all the steps involved in making your heart beat, you are also unaware of most of your thought
While you may have conscious awareness of some thought processes, the majority remain opaque to your consciousness
Psychoanalysis suggests that trained professional help is often necessary to analyze one's own thoughts effectively.
a. (ping-pong) A reasonable level of proficiency was achieved by Andersson's robot (Andersson, 1988).
b. (driving in Cairo) No. Although there has been significant progress in automated driving, current systems rely on relative
constant clues. They are not yet capable of avoiding obstacles or other cars or changing lanes appropriately, especially in
unpredictable environments like downtown Cairo.
c. (shopping at the market) No. No robot can currently perform the complex tasks involved in moving in a crowded
environment, identifying a wide variety of objects using vision, and grasping objects (including squishable vegetables) witho
damaging them. While component pieces can handle individual tasks, integrating them into a cohesive system poses a
significant challenge.
d. (shopping on the web) Yes. Software robots are capable of handling tasks such as web grocery shopping, particularly if t
design of the website remains relatively stable over time.
e. (bridge) Yes. Programs like GIB can now play at a solid level.
f. (theorem proving) Yes. For example, the proof of Robbins' algebra described on page 309.
g. (funny story) No. While some computer-generated prose and poetry may be unintentionally funny, intentional humor
remains a challenge for AI.
h. (legal advice) Yes, in some cases. AI has a history of research into automated legal reasoning, with notable applications
guiding members of the public in dealing with complex legal matters. However, further progress in more complex areas awa
satisfactory encoding of common-sense knowledge pertaining to commercial transactions and agreements.
i. (translation) Yes, to a limited extent. Speech translation is already being done, albeit with certain limitations on the curren
state of the art.
j. (surgery) Yes. Robots are increasingly being used for surgery, albeit always under the supervision of a doctor.
computationalintelligence(UniversityofKarachi)
, Instructor’sManual:
ExerciseSolutions
for
ArtificialIntelligence
AModernApproach
SecondEdition
StuartJ.RussellandPeterNorvig
, SolutionsforChapter1
Introduction
1.1 a. Dictionary definitions of intelligence often refer to "the capacity to acquire and apply knowledge," "the faculty of
thought and reason," or "the ability to comprehend and profit from experience." While these definitions are reasonable, fo
quantifiable purposes, we might define intelligence as "the ability to apply knowledge in order to perform better in a give
environment."
b. Artificial intelligence is defined as the study and construction of agent programs designed to perform well within a specif
environment, given a particular agent architecture.
c. An agent is defined as an entity that takes action in response to percepts from its environment. See the solution for Exerc
26.1 for a discussion of potential objections.
The probability of fooling an interrogator depends on the skill level of the interrogator. While one entrant in the 2002 Loeb
prize competition did manage to fool one judge, it's unclear what that judge was thinking based on the transcript.
Nonetheless, there have been instances of chatbots or online agents fooling humans, such as the Julia chatbot. Today, we
estimate the chance of fooling an interrogator at around 10%, with variation primarily depending on the skill of the
interrogator rather than the program. In 50 years, with the significant investments expected in artificial actors by the
entertainment industry, it's likely that highly credible impersonators will emerge.
The 2002 Loebner prize was awarded to Kevin Copple's program ELLA. It consists of a prioritized set of pattern/action rule
utilizing a large database of text and the Wordnet online dictionary. While ELLA is effective within its domain, it does not
contribute significantly to advancing the theory of AI; it mainly provides evidence on the number and type of rules sufficie
for producing one type of conversation.
No, it means that AI systems should avoid attempting to solve intractable problems, as they can usually only approximate
optimal behavior. Humans also don't solve NP-complete problems; instead, they excel at solving specific instances with a l
of structure, often with the aid of background knowledge. AI systems should strive for similar approaches.
, No, while IQ test scores correlate well with certain measures like success in college, they don't measure everything. A progr
specialized only for IQ tests, particularly the analogy part, would likely perform poorly on other measures of intelligence. I
tests have limitations, and there are other aspects to "intelligence" beyond what they measure.
Just as you are unaware of all the steps involved in making your heart beat, you are also unaware of most of your thought
While you may have conscious awareness of some thought processes, the majority remain opaque to your consciousness
Psychoanalysis suggests that trained professional help is often necessary to analyze one's own thoughts effectively.
a. (ping-pong) A reasonable level of proficiency was achieved by Andersson's robot (Andersson, 1988).
b. (driving in Cairo) No. Although there has been significant progress in automated driving, current systems rely on relative
constant clues. They are not yet capable of avoiding obstacles or other cars or changing lanes appropriately, especially in
unpredictable environments like downtown Cairo.
c. (shopping at the market) No. No robot can currently perform the complex tasks involved in moving in a crowded
environment, identifying a wide variety of objects using vision, and grasping objects (including squishable vegetables) witho
damaging them. While component pieces can handle individual tasks, integrating them into a cohesive system poses a
significant challenge.
d. (shopping on the web) Yes. Software robots are capable of handling tasks such as web grocery shopping, particularly if t
design of the website remains relatively stable over time.
e. (bridge) Yes. Programs like GIB can now play at a solid level.
f. (theorem proving) Yes. For example, the proof of Robbins' algebra described on page 309.
g. (funny story) No. While some computer-generated prose and poetry may be unintentionally funny, intentional humor
remains a challenge for AI.
h. (legal advice) Yes, in some cases. AI has a history of research into automated legal reasoning, with notable applications
guiding members of the public in dealing with complex legal matters. However, further progress in more complex areas awa
satisfactory encoding of common-sense knowledge pertaining to commercial transactions and agreements.
i. (translation) Yes, to a limited extent. Speech translation is already being done, albeit with certain limitations on the curren
state of the art.
j. (surgery) Yes. Robots are increasingly being used for surgery, albeit always under the supervision of a doctor.
