A Comprehensive Guide to the Foundations of
Artificial Intelligence
Based on COSC 3P71 Course Materials
Part I
Defining the Landscape of Artificial
Intelligence
1 Introduction: What is Artificial Intelligence?
Artificial Intelligence (AI) is a vast and rapidly evolving field dedicated to the creation of intelligent
entities. For thousands of years, we have sought to understand how we think and act; AI builds on
this quest by not only seeking to understand intelligence but also to engineer it. From general-purpose
capabilities like learning and reasoning to specific tasks like diagnosing diseases or driving cars, AI has
become a universal field relevant to nearly every intellectual task.
At its core, the field of AI is focused on building intelligent entities. This broad objective is often
conflated in the public eye with Machine Learning (ML), which is more accurately described as a subfield
of AI. Machine learning studies the ability of a system to improve its performance based on experience.
While many modern AI systems use machine learning methods to achieve their intelligence, it is not a
prerequisite; some AI systems, particularly classical ones, do not use ML at all. AI is the overarching
goal, and ML is one of several powerful methods for achieving it.
1.1 The Four Historical Views of AI
Historically, the pursuit of artificial intelligence has been approached from four different perspectives.
These viewpoints can be organized along two key dimensions: whether the focus is on thought processes
or on outward behavior, and whether the standard for success is human performance or an abstract ideal
of rationality.
Fidelity to Human Performance Fidelity to Ideal Rationality
Thinking Systems that think like humans Systems that think rationally
Acting Systems that act like humans Systems that act rationally
Table 1: The Four Historical Views of AI
The human-centered approaches are partly an empirical science, connected to psychology and involv-
ing observations of human behavior. In contrast, the rationalist approaches involve a combination of
mathematics and engineering, connecting to fields like statistics, control theory, and economics.
1.1.1 1. Acting Humanly: The Turing Test Approach
In his seminal 1950 paper, “Computing Machinery and Intelligence,” A.M. Turing proposed an opera-
tional test for intelligence known as the “imitation game,” now called the Turing Test. Instead of asking
the ambiguous question “Can machines think?”, Turing restated it: Can a computer system fool a human
into believing it is also human?
To pass the Turing Test, a machine would require:
• Natural Language Processing: To communicate successfully in a human language.
1
, • Knowledge Representation: To store what it knows and hears.
• Automated Reasoning: To use stored information to answer questions and draw new conclu-
sions.
• Machine Learning: To adapt to new circumstances and extrapolate patterns.
1.1.2 2. Thinking Humanly: The Cognitive Modeling Approach
This approach aims to build programs that model and simulate human thought processes directly. To
achieve this, we must first understand how humans think through:
1. Introspection: Attempting to catch our own thoughts as they occur.
2. Psychological Experiments: Observing a person in action to infer their thought processes.
3. Brain Imaging: Observing the brain in action through techniques like fMRI.
1.1.3 3. Thinking Rationally: The “Laws of Thought” Approach
This perspective traces its roots to the Greek philosopher Aristotle, who first attempted to codify “right
thinking” through the study of logic. However, this approach faces two significant challenges:
1. Logical Deliberation is Not All of Intelligence: Not all intelligent behavior is mediated by
formal logic.
2. The Problem of Uncertainty: Formal logic requires knowledge of the world that is certain,
which is rarely available in reality.
1.1.4 4. Acting Rationally: The Rational Agent Approach
A rational agent is an entity that acts to achieve the best expected outcome, given the available infor-
mation. This is the approach advocated by Russell and Norvig, and the one we primarily adopt in this
guide. This viewpoint defines intelligence by the actions an entity takes, “doing the right thing,” rather
than the thought processes behind them.
1.2 The Dual Goals and Modern Challenges of AI
The field of AI has two primary goals: an engineering goal to solve real-world problems using knowledge
and reasoning, and a scientific goal to use computers as a platform for studying intelligence itself.
However, as AI systems become more powerful, the “standard model” of AI faces a critical limitation:
the value alignment problem.
The standard model assumes we can provide a fully specified and correct objective to the machine.
For complex, real-world tasks, this becomes nearly impossible. A system deployed with an incorrect
objective will have negative consequences, and the more intelligent the system, the more severe those
consequences could be.
This reveals that for advanced AI, we don’t want machines that are intelligent in the pursuit of their
objectives. We want machines that pursue our objectives. This requires a new model of beneficial
machines—agents that are provably beneficial to humans.
2 The Intelligent Agent Paradigm
The concept of the intelligent agent serves as a powerful, unifying framework for the study and practice
of Artificial Intelligence. An agent is anything that can be viewed as perceiving its environment through
sensors and acting upon that environment through actuators.
2
Artificial Intelligence
Based on COSC 3P71 Course Materials
Part I
Defining the Landscape of Artificial
Intelligence
1 Introduction: What is Artificial Intelligence?
Artificial Intelligence (AI) is a vast and rapidly evolving field dedicated to the creation of intelligent
entities. For thousands of years, we have sought to understand how we think and act; AI builds on
this quest by not only seeking to understand intelligence but also to engineer it. From general-purpose
capabilities like learning and reasoning to specific tasks like diagnosing diseases or driving cars, AI has
become a universal field relevant to nearly every intellectual task.
At its core, the field of AI is focused on building intelligent entities. This broad objective is often
conflated in the public eye with Machine Learning (ML), which is more accurately described as a subfield
of AI. Machine learning studies the ability of a system to improve its performance based on experience.
While many modern AI systems use machine learning methods to achieve their intelligence, it is not a
prerequisite; some AI systems, particularly classical ones, do not use ML at all. AI is the overarching
goal, and ML is one of several powerful methods for achieving it.
1.1 The Four Historical Views of AI
Historically, the pursuit of artificial intelligence has been approached from four different perspectives.
These viewpoints can be organized along two key dimensions: whether the focus is on thought processes
or on outward behavior, and whether the standard for success is human performance or an abstract ideal
of rationality.
Fidelity to Human Performance Fidelity to Ideal Rationality
Thinking Systems that think like humans Systems that think rationally
Acting Systems that act like humans Systems that act rationally
Table 1: The Four Historical Views of AI
The human-centered approaches are partly an empirical science, connected to psychology and involv-
ing observations of human behavior. In contrast, the rationalist approaches involve a combination of
mathematics and engineering, connecting to fields like statistics, control theory, and economics.
1.1.1 1. Acting Humanly: The Turing Test Approach
In his seminal 1950 paper, “Computing Machinery and Intelligence,” A.M. Turing proposed an opera-
tional test for intelligence known as the “imitation game,” now called the Turing Test. Instead of asking
the ambiguous question “Can machines think?”, Turing restated it: Can a computer system fool a human
into believing it is also human?
To pass the Turing Test, a machine would require:
• Natural Language Processing: To communicate successfully in a human language.
1
, • Knowledge Representation: To store what it knows and hears.
• Automated Reasoning: To use stored information to answer questions and draw new conclu-
sions.
• Machine Learning: To adapt to new circumstances and extrapolate patterns.
1.1.2 2. Thinking Humanly: The Cognitive Modeling Approach
This approach aims to build programs that model and simulate human thought processes directly. To
achieve this, we must first understand how humans think through:
1. Introspection: Attempting to catch our own thoughts as they occur.
2. Psychological Experiments: Observing a person in action to infer their thought processes.
3. Brain Imaging: Observing the brain in action through techniques like fMRI.
1.1.3 3. Thinking Rationally: The “Laws of Thought” Approach
This perspective traces its roots to the Greek philosopher Aristotle, who first attempted to codify “right
thinking” through the study of logic. However, this approach faces two significant challenges:
1. Logical Deliberation is Not All of Intelligence: Not all intelligent behavior is mediated by
formal logic.
2. The Problem of Uncertainty: Formal logic requires knowledge of the world that is certain,
which is rarely available in reality.
1.1.4 4. Acting Rationally: The Rational Agent Approach
A rational agent is an entity that acts to achieve the best expected outcome, given the available infor-
mation. This is the approach advocated by Russell and Norvig, and the one we primarily adopt in this
guide. This viewpoint defines intelligence by the actions an entity takes, “doing the right thing,” rather
than the thought processes behind them.
1.2 The Dual Goals and Modern Challenges of AI
The field of AI has two primary goals: an engineering goal to solve real-world problems using knowledge
and reasoning, and a scientific goal to use computers as a platform for studying intelligence itself.
However, as AI systems become more powerful, the “standard model” of AI faces a critical limitation:
the value alignment problem.
The standard model assumes we can provide a fully specified and correct objective to the machine.
For complex, real-world tasks, this becomes nearly impossible. A system deployed with an incorrect
objective will have negative consequences, and the more intelligent the system, the more severe those
consequences could be.
This reveals that for advanced AI, we don’t want machines that are intelligent in the pursuit of their
objectives. We want machines that pursue our objectives. This requires a new model of beneficial
machines—agents that are provably beneficial to humans.
2 The Intelligent Agent Paradigm
The concept of the intelligent agent serves as a powerful, unifying framework for the study and practice
of Artificial Intelligence. An agent is anything that can be viewed as perceiving its environment through
sensors and acting upon that environment through actuators.
2
