Introduction to AI

Topics for This Week

• What is AI? History and evolution.
• Impact of AI on people's daily lives.
• Impact of AI on modern education.
• AI applications and use cases in different fields.
• Various AI tools (chatbots such as Clova X, Nate AI, wrtn, ChatGPT, Gemini,
Perplexity and others.
History of AI
History of AI
The Inception of Artificial Intelligence (1943-1956)

• The beginning of artificial intelligence (AI) began

in 1943 with the work of Warren McCulloch and
Walter Pitts, who proposed a model of artificial
neurons based on brain function, propositional
logic, and Turing’s theory of computation.
• They demonstrated that neural networks could
perform logical operations and suggested that
such networks could learn.
• Donald Hebb later introduced Hebbian learning, a
fundamental rule for updating neural connections.

Image Source: [Link]

What is a Neuron?

• A neuron is a nerve cell that

acts as the building block of
the brain and nervous system.
• Neurons send and receive
information using electrical
and chemical signals.
• Microscopic imaging of mouse
neurons is shown in the figure
(source:
[Link]
ent/article/controversial-
study-redraws-classical-
picture-neuron)
Hebbian Learning
Hebbian learning is a fundamental principle in
neuroscience and artificial intelligence. It is often
summarized as:
👉 "Neurons that fire together, wire together."
How it works:
• When two neurons activate at the same time, the
connection between them strengthens.
• If they rarely activate together, the connection
weakens.
Example in daily life:
• When learning a new language, repeating words
alongside their meanings strengthens the brain’s Image Source: [Link]
connections. Over time, the association becomes
stronger, making recall easier.
The Inception of Artificial Intelligence (1943-1956)
• In 1950, Marvin Minsky and Dean Edmonds built
the first neural network computer, the Stochastic
Neural Analog Reinforcement Calculator
(SNARC), which simulated 40 neurons.
• Around the same time, early AI applications
emerged, such as checkers-playing programs by
Christopher Strachey and Arthur Samuel.
• Alan Turing significantly influenced AI by
introducing the Turing test. He proposed that
human-level AI could be achieved through
learning algorithms rather than manual
Image Source: Wikipedia
programming. The control panel for the C-1
gyroscopic autopilot. A surplus C-1 was
used in the SNARC.
The Inception of Artificial Intelligence (1943-1956)
• In 1956, John McCarthy organized the Dartmouth
Workshop, bringing together key AI researchers,
including Minsky, Claude Shannon, Nathaniel
Rochester, Allen Newell, and Herbert Simon.
• The workshop aimed to explore how machines
could simulate intelligence, but it did not
produce immediate breakthroughs.
• However, Newell and Simon’s Logic Theorist, a
mathematical theorem-proving system, marked
a significant achievement, even impressing
Bertrand Russell.
•
Image Source: Wikipedia
Despite setbacks, this period laid the foundation Marvin Minsky

for AI as a scientific field.

Early Enthusiasm, Great Expectations (1952-1969)
• In the 1950s, many intellectuals were skeptical about artificial intelligence, often
asserting that "a machine can never do X," where X represented various cognitive
tasks that were believed to require human intelligence. Alan Turing compiled a long
list of such tasks.
• AI researchers responded by proving these assumptions wrong, systematically
demonstrating that machines could, in fact, perform many of these so-called
impossible tasks. They focused on areas traditionally associated with intelligence,
such as playing strategic games (e.g., chess and checkers), solving puzzles,
performing complex mathematical calculations, and even tackling IQ test-style
problems.
• Essentially, X represents any intellectual challenge that was thought to be uniquely
human—until AI researchers showed otherwise.
Early Enthusiasm, Great Expectations (1952-1969)
• At IBM, Nathaniel Rochester and his
colleagues created some of the first AI
programs.
• Herbert Gelernter (1959) constructed the
Geometry Theorem Prover, which could
prove theorems that many mathematics
students would find quite difficult. This work
was a precursor to modern mathematical
theorem provers.
• In 1958, John McCarthy made two major Image Source: Wikipedia
contributions to AI. In MIT AI Lab Memo No. 1,
he defined the high-level language Lisp,
which would become the dominant AI
programming language for the next 30 years.
Early Enthusiasm, Great Expectations (1952-1969)
• At MIT, Minsky supervised a number of students
who chose bounded problems that seemed to
require intelligence to solve. These restricted
domains became known as microworlds.
• The most famous microworld is the Block World,
which consists of a set of solid blocks placed on a
tabletop (or, more often, a simulation of a tabletop),
as shown on the next page.
• A typical task in this world is to rearrange the
blocks in a particular way, using a robotic hand that
can pick up one block at a time.

Image Source: [Link]

Early Enthusiasm, Great Expectations (1952-1969)

Image Source: Russell, S. J., &

Norvig, P. (2022). Artificial
intelligence: a modern approach.
4th Edition. Pearson.
A Dose of Reality (1966-1973)

• From the beginning, AI researchers have not been shy about predicting their
future success. The following statement by Herbert Simon in 1957 is often
quoted:
• “It is not my aim to surprise or shock you—but the simplest way I can summarize is to say
that there are now in the world machines that think, that learn and that create. Moreover,
their ability to do these things is going to increase rapidly until—in a visible future—the
range of problems they can handle will be coextensive with the range to which the
human mind has been applied.”
• The term "visible future" is unclear, but Simon also made more concrete
predictions: that within 10 years, a computer would be a chess champion and
that an important mathematical theorem would be proved by machine. These
predictions came true (or nearly so) within 40 years rather than 10.
Deep Blue versus Garry Kasparov (1997)

• In 1997, IBM’s supercomputer Deep Blue

defeated Garry Kasparov, the reigning
world chess champion, in a six-game
match.
• Deep Blue was the first computer to defeat
a world champion under standard
tournament controls.
• Documentary (2003):
Game Over: Kasparov and the Machine
(trailer)

Image Source: Wikipedia

AlphaGo versus Lee Sedol (2016)

• The DeepMind Challenge Match was a five-

game Go competition between renowned
Go player Lee Sedol and AlphaGo, an AI
program developed by DeepMind.
• AlphaGo emerged victorious in four out of
the five games, with Lee Sedol winning only
the fourth match. Each game concluded
with a resignation.
• This landmark event has been likened to
the 1997 chess showdown between Deep
Blue and Garry Kasparov.
Image Source: Wikipedia
A Dose of Reality (1966-1973)

• Early AI systems failed for several key reasons. First, they relied too much on
human intuition rather than rigorous task analysis, leading to unreliable
solutions.
• Many early AI programs attempted to replicate human problem-solving without
fully understanding the complexities of the tasks they were tackling. Second,
researchers underestimated the computational difficulty of AI problems.
• Early problem-solving approaches relied on trial-and-error methods that
worked in simple environments but became unworkable as problems scaled up.
• Before computational complexity theory was established, it was assumed that
increasing processing power and memory would naturally lead to better AI
performance, an assumption that proved incorrect.
Possible Move Variations in Chess Games

• There are between 1078 to 1082 atoms in the observable universe. That’s
between ten quadrillion vigintillion and one-hundred-thousand quadrillion
vigintillion atoms, which is a lot.
• But amazingly, there are even more possible variations of chess games than
there are atoms in the observable universe.
• This is the Shannon Number and represents all of the possible move
variations in the game of chess.
• It is estimated there are between 10111 and 10123 positions (including illegal
moves) in chess. (If you rule out illegal moves, that number drops dramatically
to 1040 moves. Which is still a lot!).
Source: [Link]
universe-or-number-of-chess-moves
Expert Systems (1969-1986)

• The early approach to AI problem-solving relied on general-purpose search

mechanisms, but these methods struggled with complex problems. The
alternative was to use expert knowledge specific to a domain, leading to the
development of expert systems.
• One of the first successful examples was DENDRAL, a system that helped
chemists figure out molecular structures from mass spectrometry data.
Instead of using general rules, it relied on specific expert knowledge, which
made it very effective.
• Following DENDRAL, the MYCIN system was developed for diagnosing blood
infections. Unlike DENDRAL, MYCIN had no theoretical model to derive rules
from, so experts were interviewed to provide knowledge. Additionally, it
introduced certainty factors to handle uncertainty in medical diagnosis.
Expert Systems (1969-1986)
• The impact of domain knowledge extended to natural language processing,
where researchers like Roger Schank emphasized that understanding language
required general world knowledge. AI research during this period also saw the
development of various reasoning tools, including the Prolog programming
language in Europe and Japan and Minsky’s frames for structured knowledge
representation.
• Minsky's frames are a structured knowledge representation model that organizes
information into hierarchical data structures, where each "frame" represents a
stereotypical situation with slots for attributes and values, enabling AI systems to
reason about and understand complex concepts efficiently.
• Government-backed initiatives, such as Japan’s Fifth Generation project and the
U.S. MCC consortium, aimed to advance AI capabilities, but they failed to meet
their ambitious goals.
Prolog (1972)
• Prolog is a logic programming language
with origins in artificial intelligence,
automated theorem proving, and
computational linguistics.
• Prolog is like a detective—it solves
problems by searching for facts and rules
that match a given question. You just state
what you want to find, and Prolog figures
out the answer! Image Source: Wikipedia

• It’s widely used in AI for things like expert systems, natural language processing, and
even chatbots because it can reason logically and find solutions efficiently.
• Unlike most programming languages, Prolog doesn’t follow step-by-step instructions.
Instead, it works like a puzzle solver, matching pieces of information to reach
conclusions.
 Minsky’s Frames (1974)

Source: [Link]
Minsky’s Frames (explanation by ChatGPT)

● Minsky's frames are a way to organize knowledge in AI.

● Think of a "frame" like a template or a folder that holds information
about a common situation.
● Each frame has "slots" where specific details can be filled in.
● For example, a "restaurant" frame might have slots for "menu," "waiter,"
and "bill.“
● This helps AI understand and reason about different situations more
efficiently.
Expert Systems (1969-1986)
• By the late 1980s, the AI industry had
grown significantly, with billions invested
in expert systems, robotics, and AI
hardware.
• However, maintaining expert systems
proved challenging, as they struggled
with uncertainty and lacked the ability to
learn from experience.
• As a result, the industry entered an “AI
winter,” with many companies failing to
meet expectations, leading to a decline in
funding and interest. Image Source:
[Link]
nce-between-ai-and-expert-system/
 Discussion:
Where are expert systems used?
Expert Systems in Medicine (Perplexity)
Expert Systems in Medicine (ChatGPT)
Expert Systems in Medicine (Llama & Copilot)
Difference Between AI and Expert System
• Expert systems are an important area • Artificial Intelligence (AI) refers to the
of AI research and represent one of creation of computer systems that
its biggest successes. They were the can perform tasks requiring human-
first practical and commercial like intelligence, such as reasoning,
application of AI, demonstrating how learning, and problem-solving.
computers can assist in decision- Essentially, AI enables machines to
making. analyze situations, make decisions,
• An expert system is a computer and improve over time based on
program designed to mimic the experience.
decision-making process of a human • While the idea of intelligent machines
expert in a specific field, helping solve has existed for centuries, AI became
complex problems by applying a scientific discipline with the
specialized knowledge. development of digital computers.
The Rise of Machine Learning and Neural Networks
• From the mid-1980s onward, artificial intelligence experienced a major shift with the
resurgence of neural networks and the rise of probabilistic reasoning. The
backpropagation algorithm, initially developed in the 1960s, was rediscovered by
multiple research groups, leading to the widespread adoption of connectionist
models.
• These models use artificial neurons, which are inspired by how our brains work, to
learn from data and improve their predictions. Unlike early AI methods that used strict,
predefined rules, neural networks learn patterns directly from data in a more flexible
way.
• This makes them better at handling complex, real-world problems. Meanwhile, expert
systems, which relied on rigid rules to make decisions, started facing difficulties
because they couldn't handle uncertainty well. This led researchers to explore new
models, like probabilistic models, that could deal with uncertainty more effectively.
Probabilistic Reasoning (ChatGPT)
Probabilistic reasoning is a method used to make decisions or predictions when the information
available is uncertain or incomplete. It involves using probabilities to quantify uncertainty and
then reasoning based on those probabilities.
Key concept:
• Probability represents the likelihood of an event happening. It is a value between 0
(impossible) and 1 (certain).
• Probabilistic reasoning helps you make educated guesses or decisions even when the exact
answer is not known.
Probabilistic Reasoning Examples (ChatGPT)
Weather Forecasting Traffic
Example: "There is a 70% chance of rain Example: "There’s an 80% chance that traffic will
tomorrow." be heavy at 5 p.m. today."
Reasoning: This means that, based on past data Reasoning: Based on past traffic patterns and the
and current conditions, it is more likely that it will time of day, there’s a high probability of traffic, but
rain than not, but it’s still uncertain. You might carry it's still not a guarantee. You can plan accordingly.
an umbrella, but you know it’s not guaranteed to
rain. Why It’s Useful
Probabilistic reasoning is essential in situations
Medical Diagnosis where you don't have all the answers but need to
Example: A doctor says, "There is a 90% chance make the best decision possible, like in medicine,
that you have the flu, based on your symptoms." weather forecasting, artificial intelligence, and
Reasoning: The doctor uses your symptoms and even everyday choices.
other factors to estimate the likelihood that you
have the flu. It’s not a certainty, but it helps guide
the doctor in making decisions about treatment.
Backpropagation Algorithm
The Rise of Machine Learning and Neural Networks

• By the late 1980s, AI research embraced statistical methods, optimization methods,

and machine learning, integrating knowledge from fields like control theory and
decision science.
• Judea Pearl’s development of Bayesian networks introduced a structured way to
represent uncertainty, while Rich Sutton’s reinforcement learning research connected
AI decision-making to Markov decision processes (MDPs).
• These advances paved the way for significant progress in speech recognition,
robotics, and natural language processing.
• Machine learning benchmarks and competitions, such as ImageNet and SQuAD,
further drove innovation by standardizing evaluation methods. This shift towards
data-driven AI not only improved real-world applications but also helped unify
previously separate subfields, setting the stage for modern AI breakthroughs.
What is a Neural Network?

YouTube
video
6 min
What is a Neural Network?

YouTube
video
5 min
 Discussion:
Can machines think?
Can Machines Think?
• Alan Turing, in 1950, posed the question: "Can
machines think?" He suggested reframing the
discussion from whether a machine "thinks" to
whether it can exhibit intelligent behavior.
• To assess this, he developed the Turing test, which
evaluates a machine's ability to mimic human
conversation. Since we can only observe a
machine’s behavior, the question of whether it truly
"thinks" or has a "mind" is irrelevant.
• Turing pointed out that we cannot determine this
about other people either, yet we assume they
think as a social convention.
Image Source: Wikipedia
Can Machines Think?
• The Turing test provides some indication of
intelligence but unfairly penalizes forms of intelligence
that do not resemble human behavior.
• Some researchers criticize the Turing test for
requiring machines to imitate humans, noting that
fields like aeronautics do not aim to build aircraft that
mimic birds precisely.
• AI pioneer John McCarthy also rejected the idea that
artificial intelligence must simulate human intelligence,
defining it instead as "the computational part of the
ability to achieve goals in the world.“
• Similarly, Marvin Minsky described intelligence as "the
ability to solve hard problems."
Image Source: [Link]
Turing Test

YouTube
video
4 min
Can you think? (ChatGPT)
Can you think? (Google Gemini)
Can you think? (Google Gemini)
 Discussion:
Can animals think?
Deciphering of Animal Languages
• “Similar tools could also be employed
for the training of blind animals,
although the precise number of such
animals is unknown. It’s known that
animals communicate through a variety
of signs, such as sounds and
movements.”
• “In the future, AI could assist in the
deciphering of animal language,
thereby enabling us to comprehend the
meaning of animal communication.”
• “Various AI tools could be utilized to
facilitate communication with and the
training of animals to live in the world
of darkness.”
Deciphering of Animal Languages
• "The behaviors of animals can be influenced by
climate change, and by understanding their
methods of communication with the help of AI,
scientists can understand how these changes affect
their communication, migration patterns and overall
ecosystems, enabling better strategies for
conservation and climate adaptation."
• "If AI is able to assist us in understanding bird
sounds in the future, what insights might we gain
from birds, and what requests might they make of
us?"

Image Source: [Link]

Definitions of AI

• A widely accepted AI textbook defines intelligence

as the study of systems that perceive their
environment and take actions to maximize their
chances of achieving specific goals.
• This perspective treats intelligence as problem-
solving, where both the complexity of the task and
the system’s performance serve as direct measures
of intelligence, avoiding philosophical debates.
• Despite the copyright, some universities host the
book on their university websites.
Image Source: Wikipedia
The Most Popular Textbook (> 50 000 citations)

[Link]
Definitions of AI
Definitions of AI

• Google, a major AI developer, defines intelligence as

a system's ability to synthesize information,
aligning with how biological intelligence is
understood.
• However, some argue that AI remains a unclear and
contested term, with disagreements over whether
classical algorithms qualify as AI.
• During the AI boom of the early 2020s, many
companies used "AI" as a marketing term, even
when their technology did not meaningfully
incorporate artificial intelligence.
Image Source: [Link]
The Most Popular AI Textbook
What are the key impacts
of AI on people's daily
lives?
What is the impact of AI on
modern education?