Introduction to Artificial Intelligence
© G. Lakemeyer
G. Lakemeyer
Winter Term 2024/25
Introduction to Artificial Intelligence
Instructor: Gerhard Lakemeyer
Office Hour: Th 1-2PM, or by appointment.
Email: lakemeyer@kbsg.rwth-aachen.de
© G. Lakemeyer
Teaching Assistants:
Matthew Lynn hlynn@kbsg.rwth-aachen.dei
Qihui Feng hfeng@kbsg.rwth-aachen.dei
AI/WS-2024/25 2 / 29
Textbook
Stuart Russell und Peter Norvig
© G. Lakemeyer
Artificial Intelligence
— A Modern Approach —
Pearson, 4th Ed., 2020
(3rd Edition ok)
AI/WS-2024/25 3 / 29
Bachelor/Master Informatik and others
Course can be used for both Bachelor and Master Informatik in
the area Data and Information Management
Same for Master Software Systems Engineering.
For other Master programs, please, check whether or how the
course can be used.
© G. Lakemeyer
Requirements:
Written exam (120 Minutes, dates: Feb. 06 and March 18, 2024).
Homework assignments are optional but strongly recommended.
Prepare solutions in groups of 2–4 students.
Solutions can be written in English or German.
AI/WS-2024/25 4 / 29
What is AI (1)
. . . or: when is an artifact “artificially” intelligent?
1. Systems that act like humans:
Turing Test (1950)
Interrogator asks questions by typing them in.
© G. Lakemeyer
Answers are displayed on a screen.
Deduce from the answers whether answers come from a machine or a
person.
AI/WS-2024/25 5 / 29
What is AI (1)
. . . or: when is an artifact “artificially” intelligent?
1. Systems that act like humans:
Turing Test (1950)
Interrogator asks questions by typing them in.
© G. Lakemeyer
Answers are displayed on a screen.
Deduce from the answers whether answers come from a machine or a
person.
very difficult; needs at least:
Natural language processing
Knowledge Representation
Automated Reasoning
Machine Learning
Loebner Prisa
CUntil 2019)
AI/WS-2024/25 5 / 29
What is AI? (2)
2. Systems that think like humans
Attempts to understand human thought processes.
Not just the solution to a problem counts, but the way it is achieved.
© G. Lakemeyer
Research topic of Cognitive Science and Psychology
AI/WS-2024/25 6 / 29
What is AI? (2)
2. Systems that think like humans
Attempts to understand human thought processes.
Not just the solution to a problem counts, but the way it is achieved.
© G. Lakemeyer
Research topic of Cognitive Science and Psychology
Not dealt with in this course.
AI/WS-2024/25 6 / 29
What is AI? (3)
3. Systems that think rationally
Example: Aristotelian Syllogisms:
Socrates is a man.
© G. Lakemeyer
All men are mortal.
Therefore: Socrates is mortal.
Research topic of Philosophy, Logic and Mathematics.
Largely the foundation of Knowledge Representation
AI/WS-2024/25 7 / 29
What is AI? (4)
4. Systems that act rationally. more general than 3.
rational thought is often, but not always necessary
Example: Reflexes for self-protection.
Central idea:
© G. Lakemeyer
Agents which perceive their environment and act in a reasonable way.
The goal is to select actions with the highest utility.
Optimality is not always possible because of limited resources.
AI/WS-2024/25 8 / 29
What is AI? (4)
4. Systems that act rationally. more general than 3.
rational thought is often, but not always necessary
Example: Reflexes for self-protection.
Central idea:
© G. Lakemeyer
Agents which perceive their environment and act in a reasonable way.
The goal is to select actions with the highest utility.
Optimality is not always possible because of limited resources.
AI as “Acting rationally” is the foundation of this course!
Questions we will try to answer:
What are the components which allow an artifact to act rationally?
AI/WS-2024/25 8 / 29
AI at RWTH Aachen University
Prof. Wil van der Aalst: Process Mining
Prof. Stefan Decker: Web Science, Semantic Web
Prof. Holger Hoos: Machine Learning, Automated Reasoning, Optimization
Prof. Hector Geffner: Automated Planning, Machine Learning
© G. Lakemeyer
Prof. Gerhard Lakemeyer: Knowledge Representation, Cognitive Robotics
Prof. Bastian Leibe: Computer Vision
Prof. Christopher Morris: Machine Learning on Graphs
Prof. Sebastian Trimpe (Mechanical Engineering): Machine Learning
See also RWTH AI Center: ai.rwth-aachen.de
AI/WS-2024/25 9 / 29
Sources for Research Literature
Conferences
International Joint Conference on Artificial Intelligence (IJCAI)
Conference of the Association for the Advancement of AI (AAAI)
European Conference on AI (ECAI)
German AI Conference (KI)
© G. Lakemeyer
Journals (among many others)
Artificial Intelligence, Elsevier
Journal of Artificial Intelligence Research (Open Access!)
Other resources
Internet (Most authors post pre-prints on the web)
Google Scholar is a good starting point.
AI/WS-2024/25 10 / 29
A Very Brief History of AI
Important ideas predating AI:
Aristotle (384–322 B.C.)
Logical syllogisms.
© G. Lakemeyer
Leibniz (1646–1716),
Idea of the “Calculus Philosophicus,” tried to formalize all of human
thought.
Boole and Frege, founders of modern logic.
AI/WS-2024/25 11 / 29
History of AI (2)
The early days of AI: 1940–1956
1943 Neural Nets (NN), McCulloch + Pitts
1949 Learning in NN, Hebb
since 1950 First chess programs: Shannon, Turing
© G. Lakemeyer
1956 Dartmouth Conference
McCarthy, Minsky, Shannon, Rochester,
More, Samuel, Solomonoff, Selfridge,
Newell, Simon.
Giving it a name: Artificial Intelligence.
AI/WS-2024/25 12 / 29
History of AI (3)
First enthusiasm: 1952–1969
since 1952 Samuel: a checker program that could learn,
eventually could beat Samuel!
since 1955 Newell + Simon: Logic Theorist,
proved theorems from Principia Mathematica
© G. Lakemeyer
1957 Chomsky: Theory of natural language processing
1958 McCarthy: Lisp
1962 Rosenblatt: Proved that perceptrons (simple NN’s)
could learn effectively if function representable.
1969 Green: Planning using logic.
SRI: First mobile robot called Shakey
AI/WS-2024/25 13 / 29
History of AI (4)
Setbacks: 1966–1974
1966 abrupt end of funding for machine translation
!
1969 Minsky + Papert:
© G. Lakemeyer
Perceptrons able to learn almost nothing!
1973 Lighthill Report (UK)
“AI is just too hard, hence will never work.”
English Russian - >
English "
in : "The Spirit is
willing , but the flesh i weak .
out : "The Vodka is good ,
but the meat is rotten " .
AI/WS-2024/25 14 / 29
History of AI (5)
Knowledge-Based Systems (then: expert systems): 1969–1985
1969 Buchanan et al.: DENDRAL
Recognizing molecular structures
1975 Shortliffe: MYCIN
Diagnosing infectious diseases
© G. Lakemeyer
(often better than specialist MD’s!)
1982 J. McDermott: R1 (XCON)
Configuration of Computer Systems (VAX)
Beginning of the Commercialization of AI
Special AI hardware: Lisp Machines
(dead-end because of cheaper and powerful workstations, PCs)
AI/WS-2024/25 15 / 29
History of AI (6)
The return of NN’s: 1982–
since 1982 Learning methods for multi-layered feed-forward nets
(back-propagation algorithm)
1982 Hopfield: Associative Memory
© G. Lakemeyer
1986 Hinton, Sejnowski:
NN’s learn how to read written text aloud.
2006 Hinton:
-
Deep Learning
2024 Nobel Trise.
AI/WS-2024/25 16 / 29
Stanley wins Grand Challenge in 2005
© G. Lakemeyer
Driving autonomously through the Mojave Desert for 200km
In 2007 cars drive 100km autonomously in a (mockup) city
Today autonomous cars are tested routinely on regular roads
AI/WS-2024/25 17 / 29
Watson wins Jeopardy! in 2011
© G. Lakemeyer
Answers questions about various categories
(movies, history, science, etc.)
Extracts knowledge from the internet
Generates (many) possible answers
Ranks possible answers according to confidence
Answers when confidence reaches a threshold
AI/WS-2024/25 18 / 29
AlphaGo conquers Go in 2016
© G. Lakemeyer
Google DeepMind’s AlphaGo wins 4:1 against Lee Sedol
Uses deep reinforcement learning and self play
Outperformed by AlphaGo Zero, uses no expert knowledge
AI/WS-2024/25 19 / 29
AlphaFold 2 predicts protein structures in 2020
© G. Lakemeyer
DeepMind’s AlphaFold predicts the 3D structure of proteins from
amino-acid sequences
Uses deep learning with physical and geometric constraints as additional
input
Achieves about 90% accuracy, which is considered a solution
AI/WS-2024/25
2024 Nobel Trite in Chemists 20 / 29
Large Language Models (LLMs)
© G. Lakemeyer
consist of very large neural networks
(> 100 billion connections)
are trained with hundreds of Gigabytes of text from the internet
are based on the idea of statistical sentence completion
generate answers to (arbitrary) questions
are able to generate computer programs
AI/WS-2024/25 21 / 29
Large Language Models (LLMs)
© G. Lakemeyer
consist of very large neural networks
(> 100 billion connections)
are trained with hundreds of Gigabytes of text from the internet
are based on the idea of statistical sentence completion
generate answers to (arbitrary) questions
are able to generate computer programs
Caveat:
do not always tell the truth
have a tendency to hallucinate
AI/WS-2024/25 21 / 29
Topics Covered in the Course
Introduction
Agent Architectures
Problem Solving as Search
Constraint Satisfaction
© G. Lakemeyer
Knowledge Representation
Planning
Reasoning under Uncertainty
Learning
Robotics
Philosophical Aspects
AI/WS-2024/25 22 / 29
Problem Solving as Search
Example: The 8-Queens Problem
© G. Lakemeyer
AI/WS-2024/25 23 / 29
Constraint Satisfaction
Example: The 3-coloring Problem
Northern
Territory
© G. Lakemeyer
Queensland
Western
Australia
South
Australia New
South
Wales
Victoria
Tasmania
AI/WS-2024/25 24 / 29
Knowledge Representation
Example: logical representations
A proud mother is someone whose children are all doctors:
© G. Lakemeyer
8xProudMother (x ) ⌘
9y .MotherOf (x , y ) ^ 8z .MotherOf (x , z ) Doctor (z ).
AI/WS-2024/25 25 / 29
Planning
B C A C
)
© G. Lakemeyer
Primitive Actions:
pickup(x ), puton(x , y )
Possible successful action sequence:
pickup(B); puton(B, C); pickup(A); puton(A, B)
Problem: How to plan action sequences in general such that a given initial
state is transformed into a goal state.
AI/WS-2024/25 26 / 29
Reasoning under Uncertainty
Example: Causal (Bayesian) Nets
P(E)
P(B)
Burglary Earthquake .002
.001
© G. Lakemeyer
B E P(A)
T T .95
Alarm T F .94
F T .29
F F .001
A P(J)
A P(M)
JohnCalls T .90
F .05 MaryCalls T .70
F .01
AI/WS-2024/25 27 / 29
Learning
Example: Neural Nets
Output units Oi
Wj,i
© G. Lakemeyer
Hidden units a j
Wk,j
Input units Ik
AI/WS-2024/25 28 / 29
Robotics
© G. Lakemeyer
AI/WS-2024/25 29 / 29
