Artificial Intelligence 2, DVA265

Term: Spring 2024 (VT24)

Location: Eskilstuna
Programme: Bachelor of Science (BSc) in Applied Artificial Intelligence

Examiner and lecturer: Baran Cürüklü, [email protected], 073-9607453 (visiting

address: U1-079 Högskoleplan 1, Rosenhill, Västerås)
Lecturer in ethics: Maria Ehn [email protected]
Laboratory assistant: Elmeri Syrjänen [email protected]
Document history
Version Date Contributing Contribution
person
0.1 01.03.2024 B. Cürüklü Initial version of the document is
finalised.
0.2 04.03.2024 Maria Ehn Ethics added

2
 1. Course schedule
This is the link to the schedule: The schedule. Please, take a look at think link to access
correct information regarding our activities in the course.

2. Reading instructions
We start with a few chapters (Chapter 1 that you may have read during the Artificial
Intelligence 1 (AI1) course.

So why do I recommend you to read them again? Well, in this course our ambition is to look
into intelligent agents and we will try to advance our knowledge in designs of intelligent
agents that interact with the environment, with each other, and hence create complex
interactions.

Table 1. Reading instructions

Chapter Lecture Notes
Ch. 2 We will design advanced agents in this course,
Intelligent probably more advanced than you have done in the
Agents 1 AI1 course. Please, read this chapter (again) with that
ambition, and this about what you would like to do
every assignment.

3
 Ch. 4 Search Sect. 4.1.4 Evolutionary algorithms (page 133-137).
in Complex However, you will need to relate population-based
2
Environments algorithms to single state algorithms such as Hill-
climbing and Simulated annealing.
Ch. 17 Read and try to understand the concepts. We will
Multiagent blend this chapter with what we did in LAB3 to
Decision define a problem in LAB4.
Making 3
Ch. 11 Focus on Sect. 11.1 Definition of Classical Planning
Automated and block-world problem.
planning
Ch. 7 Logical Read all. Compare with what is Chapter 2
Agents Intelligent Agents.
Ch. 8 First- Again, read all, please.
Order Logic
4
Ch. 9 Read all, but focus on Sects. 9.2 Unification, 9.3
Inference in Forward chaining, and 9.4 Backward chaining. (In
First-Order this chapter you will need the theory from the
Logic Chapters 7 & 8.)
Ch. 10 Critical to understand all AI
Knowledge
representation
5
Ch. 12 Read all, but focus on Bayes’ Rule
Quantifying
Uncertainty
Neuroscience, Own material. The ambition is to go through some of
Computational the theories in neuroscience, and cognitive
neuroscience, 6 neuroscience, from a computational perspective and
Cognitive see how they have contributed to AI.
neuroscience
On-line lecture Look through the short films.
Ethics – material Module 1 is a basic introduction to ethics in AI.
Fairness – Module 2 is the focus of ethics in this course and the
Equality in assignment. Read all article sections listed under this
AI-based module.
Agents Module 3 presents the assignment (see section 3.2.4)
Question time in Teacher available for questions related to upcoming
Zoom seminar (via Zoom)
Seminar Presentation of assignment (6 groups per occasion)

The last part of the course is about ethics (including gender equality) related to AI-based
conversational agent. It contains:

(I) On-line course material

Module1 Basic concepts (ethics-fairness-equality) and related literature
Recommended reading
- European Commission: High-Level Expert Group on AI presented Ethics Guidelines
for Trustworthy Artificial Intelligence. Ethics Guidelines for Trustworthy AI.
Available: https://digital-strategy.ec.europa.eu/en/library/ethics-guidelines-
trustworthy-ai [accessed Feb 19th 2024]. Ethical principles described on pp. 11-13

4
- Jobin, A., Ienca, M., Vayena, E.: The global landscape of ai ethics guidelines. Nature
Machine Intelligence pp. 1–11 (2019). Ethical principles described on pp. 7-13
Module2Ethical concerns particularly relevant for agent-based systems (specific focus:
conversational agents)
Reading instructions
- At least Sections 3 and 4 in
Ruane, E., Birhane, A., & Ventresque, A. (2019, December). Conversational AI: Social
and Ethical Considerations. In AICS(pp. 104-115). https://ceur-ws.org/Vol-
2563/aics_12.pdf
- pp 285-287 in
Luxton DD. Ethical implications of conversational agents in global public health. Bull
World Health Organ. 2020 Apr 1;98(4):285-287. doi: 10.2471/BLT.19.237636. Epub
2020 Jan 27. PMID: 32284654; PMCID: PMC7133471.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7133471/
- pp 106-116 in
West, M., Kraut, R., Chew, H.E.: I’d blush if I could: closing gender divides in digital
skills through education (2019). https://unesdoc.unesco.org/ark:/48223/pf0000367416
Module3 The assignment
Reading instructions: Same as in module 2. See (see section 3.2.4)

(II) Teacher available for questions related to upcoming seminar (via Zoom)

(III) Seminar with presentations of assignment

5
 3. Assignments
You will solve the assignments in groups of two (2) students, or if you prefer alone, meaning
by yourself. Remember that LAB1-4 is your examination, so please minimize you interaction
with your class mates regarding the specific details of your solutions, simply stick to your
own group!

LAB2 and LAB4 provides you options regarding the problem to be solved (see below,
please).

The assignments aim at covering symbolic and non-symbolic AI. Symbolic Ai is based on, or
derived from logic, whereas non-symbolic AI is about numerical calculations (some would
say number crunching).

Note that symbolic versus non-symbolic is about how to represent the world, or the problem.
So, the question of representing a problem to fit an algorithm is important. Knowledge
representation is a key activity in AI, and actually in all problem solving. Simply put, you
don’t use a hammer for everything when you renovate your kitchen!

In the course schedule we have 5 supervision meetings for supporting you with the
assignments. These are 8th, 11th, 18th, and 25th of April, followed by 6th of May.

As you can see below we will start with numerical AI, that is LAB 3-4, after that we will look
at symbolic AI in LAB1-2.

Table 2. Overview of the assignments, which also corresponds to the course examination.
Type of Credits Scheduled Note Deadline for
representation (hp) the ‘LAB’
LAB1 0,5 We start with LAB3- *Tuesday 23rd
Symbolic
LAB2 2,5 April 8, 11, 18, 4 (numerical AI) and May
LAB3 2,0 25, and May 6 continue with LAB1- *Tuesday 2nd of
Numerical 2 (symbolic AI).
LAB4 2,5 May
*You will be able to do present your results online (Zoom/Teams), in addition to the
schedules sessions. Try to see these deadlines as hard so that they can guide.

3.1. EA and optimisation (LAB 3)

Implement the most simple and classic genetic algorithm. Implement selection, crossover,
mutation of your choice. The genes will take binary values, so they can be either ‘0’ or ‘1’. In
a sense these individuals represent yes/no answers and this is very much symbolic
representation as well.

In this test number of genes of the individuals’ chromosome should be 50:

Individual _1: [0 0 1 … 1]. <- you see only 4 out of 50 genes in the chromosome
Individual _2: [0 1 1 … 0]
….
Individual _20: [1 1 1 … 0]

Define now an objective function to “Maximise the sum of all genes”, well this means that
the fitness function will try to find an individual with the chromosome consist of only ‘1’

6
 This is how the best individual should look like [1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1].

Q1: Do you need crossover and mutation?

Q2: What probabilities for crossover and mutation seems to be working well?
Q3: Test your algorithm with different population sizes, from 10 to 100 individuals. Which
population works well, and why? How does the performance of the algorithm changes with
the population size, with respect to (1) time to convergence, (2) memory requirements?

function evalution-program
t ¬ 0;
init Pop(t); /*init the first generation of individuals (solutions).*/
eval Pop(t); /* compute fitness-values of them. */

while (not termination-condition) do

t ¬ t+1; /* generation counter. */
select Pop(t) from Pop(t-1); /* parents to next generation */
alter Pop(t); /* generate new individuals. */
eval Pop(t);
end

Figure 1. The pseudo code of a genetic Evolutionary algorithm.

3.1.1. Hints for the implementation

Remember this pseudo code from the 2nd lecture (Fig.1), and the other slides on this topic.
What you will do is the following:

1. Implement a roulette wheel selection algorithm.

2. The fitness (or objective) function is very simple. The goal is to get all genes = 1 for
the best solution. Thus, start with just adding all genes in an individual’s chromosome
to get the total sum.
a. Afterwards, try to improve the fitness function. Can you do that?
3. The whole current population will be replaced after selection à crossover à
mutation. Thus, if you have 50 individuals in the population, that population number
will never change.
4. It is OK if one individual is selected several times. Actually, the roulette wheel
algorithm will result in this behaviour. It is also OK if a pair of individuals are the
same individual (this is not a good thing, however, remember that testing this will
mean more computations so we ignore it).
5. As in the slides implement a crossover algorithm that with a probability of 60%
swaps the genes.
a. Afterwards, you will need to test other values, e.g., 10%, 20%, 40%, and
80% just to see if you can speed up the search.
6. Mutation is again very simple. With a probability of 3% flip ‘0’ à ‘1’, or ‘1’ à ‘0’.
a. Afterwards, change the mutation value to, very low = 1%, or much higher =
5%, or 10%. What happens with the performance (in convergence time)?

7
 7. Creating the new population:
a. When a pair of parents are modified (with crossover and mutation) 2 new
offspring are created. Thus, you have 4 individuals in total. Rank them based
on their fitness values and add the best 2 into the new population.

With the alternatives test on the fitness function (Nr. 2), crossover (Nr. 5), and mutation (Nr.
6) you have tested different algorithms, for this simple problem. What are your
conclusions?

3.2. EA and artificial agents for planning (LAB4)

3.2.1. Background

The ambition is to start from the algorithm in LAB3 and scale it up and implement a platform
for interactions between agents and at the same time solve a planning problem (see Sect.
3.2.1). Remember the EA-operators that you have implemented in LAB3: (i) selection, (ii)
crossover and (iii) mutation.

(i) Selection: This operation is about competition between the agents. They compete
to have the possibility to produce their own offspring (children).
(ii) Crossover: This operator is about exchange of information between the agents.
Thus, this is where agents interact with each other.
(iii) Mutation: In this step you can think about random changes in an agent status (or
know

In Sect. 3.2.1. you have the framework for the problem you will solve. You need to discuss
the details with your group member (and with your class mates, at this stage you all can talk
to each other freely) how you will implement the problem. Get back to me with your ideas
before starting the implementation.

This assignment will help you to think about all the intricates of designing a simple, yet
powerful, MAS world.

Learning outcome 8 is on “carrying out an ethical and gender equality analysis of an agent-
based AI system”. In Sect. 3.2.4 the assignment on ethics and gender equality in agents is
presented.

8
 3.2.2. The Multi-agent system (MAS) world

Builder agents:

Construction
material agent

Figure 2. An illustration of the MAS world that you will implement. There is one (1)
Construction material agent. This agent sells all the material that is needed for building a
house. The builder agents build houses, which they sell, and by ding that earn money. The
agent that has built the most (or earned the most) will win the competition.

In the MAS world there are 2 different types of agents:

• 1 Construction material agent
• Builder agents.

The builder agents compete with each other to build the maximum number of houses. They
can start building maximum 2 houses at the same time. All the building material will be
purchased from the construction material agent.

All houses consist of one floor, and one garret (swe: vindsrum). The MAS world house is, of
course, an oversimplification (Table 3. has all the details).

Table 3. Requirements (the components) for building a house in the MAS world.
Output The components
The house 1 floor + 1 garret
1 floor 4 bed rooms, 2 bath rooms, one living room
1 bed room 2 windows + 1 door + 1 wall module
1 bath room 1 door + 1 toilet-seat + 1 tab + shower cabin + 1 wall
module
1 living room 1 door + 3 windows + 1 wall module
1 hall 1 outside-door + 1 window + 1 wall module
1 garret 3 windows + 1 door + 1 wall module

As you can see you need to fulfil certain requirement for building a house, thus building a
house can be formulated as a planning problem. Also, a floor (and a garret) requires
components also. This is not a hard planning problem, however, it is still a planning problem.

9
Table 4. The cost of the components
The component The price (SEK)*
1 door 2500,- Inside door for the rooms
and the toilet, etc.
1 outside-door 8500,-
1 window 3450,- All windows are the same.
1 wall-module 75000,- A 4-wall module making a
room
1 toilet-seat 2995,-
1 tab 2350,-
1 shower cabin 8300,-
* The prices are realistic and come from https://www.bauhaus.se/

3.2.3. Completing the problem definition and finalising the implementation

There are a few details that are remaining in the problem definition. Discuss them with your
group member and finalise the design:

1. The construction material agent cannot have unlimited material in any given time.
How should a realistic agent behave?
2. When 2 builder agents compete to purchase, there must be a way for the construction
material agent to choose between them. what should that criterion be?
3. You need to think about how often a builder agent tries to swap material with other
agents.
4. There are probably other details as well…

3.2.4. Ethics and gender equality in agents

This assignment is related to the course’s learning outcome 8 (i.e.,” after completing the
course, the student shall be able to carry out an ethical and gender equality analysis of an
agent-based AI system”). The task is focused on AI-based conversational agent.

Content and realization:

The assignment includes:

(I) Reading in three articles on ethical concerns related to AI-based conversational

agent (individual work):
- at least Sections 3 and 4 in
Ruane, E., Birhane, A., & Ventresque, A. (2019, December). Conversational AI: Social
and Ethical Considerations. In AICS(pp. 104-115). https://ceur-ws.org/Vol-
2563/aics_12.pdf
- pp 285-287 in
Luxton DD. Ethical implications of conversational agents in global public health. Bull
World Health Organ. 2020 Apr 1;98(4):285-287. doi: 10.2471/BLT.19.237636. Epub
2020 Jan 27. PMID: 32284654; PMCID: PMC7133471.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7133471/
- pp 106-116 in
West, M., Kraut, R., Chew, H.E.: I’d blush if I could: closing gender divides in digital
skills through education (2019). https://unesdoc.unesco.org/ark:/48223/pf0000367416
- You are welcome to also add references from own searches.

10
 (II) Analyzing a conversational agent related to ethics and gender equality based on
aspects from articles (work in groups of three student):
1. Select an existing AI-based conversational agent (text- or voice based)
2. Analyze the system in relation to ethics and gender equality by
a) Selecting 4-6 aspects related to ethics (including gender equality) from the articles that
you find relevant to analyze for your system
b) Exploring/interacting with the system and discussing in the group
The analysis should consider risks for individual users (including men and women, also
from vulnerable groups) and risks on a societal level. Each identified risk should be
briefly described based on the articles and based on your interactions with the system.

(III) Preparing a 10 min presentation where you demonstrate

1. The system
2. The aspects related to ethics (including gender equality) that your analysis is based on
3. The results of your analysis illustrated by your interactions with the system.
Your interactions with the system can be presented “live” or via screen shots on PPT: s.
Please note that all group members need to actively contribute to the analysis and the
presentation (both the preparations and the demonstration at a seminar described below).

The teacher is available for questions related to upcoming seminar (via Zoom) on May
7th at 14.00-15.00.

Examination:
Will take place in a seminar where the group present their work for the class. Active
participation in the group presentation and discussion is required for grade “pass” (G) on the
assignment. Student that are absenr from or insufficiently active in the seminar will receive an
individual written assignment from the teacher (after the seminar).
Seminar occasions (both including presentations from 6 groups) are:
Monday May 13th (at 10-12) or Thursday May 16th (at 10-12). Both occasions will be in
Eskilstuna.
Please note that all groups need to sign up for one of seminar slot in Canvas (“first come,
first serve”) before May 7th.

3.3. Introduction to logic programming (LAB1)

This assignment is about using logic to represent information. This is called symbolic AI.
Why is it symbolic? Well, logic is about symbols and manipulation with symbols. Symbolic
AI is the most classical form om AI. The Prolog language have been around for a long time,
and it is an excellent tool in representing information in propositional (swe: satslogik) and
predicate (swe: predikat) logic.

You will use the web implementation of the SWI Prolog: https://swish.swi-
prolog.org/p/STRIPS%20Block%20World.swinb You can open new Program editor using the
‘+’ sign. This is community, so there are excellent examples/tutorial, and you can run
Prolog from your browser!

11
Figure 3. The SWISH page. At top right you can see a picture of me. By one click I connected
my Google account/email to SWISH so I have an account now for saving files, interacting
with others, etc.

There several other tutorials, etc. available. Some good examples are here:

1. A very good tutorial on lists:

https://www.cpp.edu/~jrfisher/www/prolog_tutorial/2_7.html
2. Tutorial on simple programs:
https://www.tutorialspoint.com/prolog/prolog_basic_programs.htm

Write this simple program (see the blue textbox below) that consist of one predicate called
medlem/2. This predicate is actually implemented already in all Prolog systems as the
predicate member/2.

medlem(X,[X|R]).
medlem(X,[Y|R]) :- medlem(X,R).

This predicate defines the relationship between two entities, where the second is a list. What
these 2 lines say is as follows:

• ‘X’ is the first element of the list, which means that it is a member.
• ‘X’ is not the first member of the list, thus it is perhaps somewhere in the rest
of the list. Thus, the interpreter removes the first element and tries to find ‘X’
is in the rest of the list called ‘R’.

Q1: Ask now the following questions to the Prolog interpreter:

12
 medlem(2,[1,2,3]).

What is the answer?

Even this simple predicate can be used in many different ways. This shows the strength of
logic programming and Prolog. See in Q2 and beyond how we use this predicate in different
ways.

Q2: Imagine that the list consists of band names. Now ask the following 2 questions. Can you
say in plain language what these questions mean?

medlem(abba,[roxette, kiss, abba]).

What is the answer?

Note that we use the same predicate for both data types.

medlem(justin_bieber,[roxette, kiss, abba]).

What is the answer?

Q3: You can even mix different types:

medlem(”van halen”,[1, roxette, 4, ”motley crue”, 3, ”van halen”]).

What is the answer?

Q4: We can make the list a bit more complex. The members of the list are band names and
the members of the band. Note that we cannot have predicates inside other predicates (at least
not as we have done below). Let us not ask this question where abba() is the first element.

member(abba(M1, M2, M3,M4), [roxette(marie, per), van_halen(david,

eddie, alex, anthony),abba(agneta, anni-frid, bjorn, benny)]).

What is this question about? What is the answer?

Q5: Benny is a member of the band ABBA. Now ask this question where only Benny is
mentioned.

13
 abba(benny, X, Y, Z).

Did it answered by matching X, Y, Z with the other 3 members or did it say ’No’? Why?

Q5 demonstrates unification, which is Prolog’s way of doing pattern matching. This means
that you need to know the order of the arguments, and that Prolog does unification by
comparing the arguments one by one.

Q6: Write a program that returns the difference between the largest and the smallest element
in the list. Start from the predicate ’largest_element’ which is written below. Use
only preicatas that you have written.

largest_element(X, [X]).
largest_element(X, [X|Rest]):- largest_element(Y, Rest), X >=Y.
largest_element(N, [X|Rest]):- largest_element(N, Rest), N > X.

Q7: Now you will test a few more Prolog predicates on your own. These are already
implemented. Try to combine them to create one program. Perhaps you can find something
interesting from the AI1 course. Some of the important predicates in Prolog are: concat/3,
length/2, reverse/2, sum/2, mean/2, etc.

During the LAB3 presentation you will need to explain the program that you have decided to
implement.

Q8: Peano numbers are simply a way to represent natural numbers, [0, 1, 2, 3, …] as
functions starting form zero. Define now +, -, /, x for 2 Peano numbers.

Natural
number
0 0
1 f(0)
2 f(f(0))
3 f(f(f(0)))
… …

During the LAB3 presentation you will need to explain how +, -, /, x are implemented.

3.4. Relationships and expert systems in Prolog (LAB2)

Let us now design an expert system. Usually, expert systems are based on forward chaining, however,
you can choose to implement an expert system similar to how Prolog works, i.e., using the Prolog’s

14
backward chaining mechanism. This implementation decision (backward, instead of forward chaining)
is yours.

My problem suggestion: In backward chaining implement the classical problem of “The Bird
Identification System” described in Chapter 2 of the book Building Expert Systems in Prolog, by
Dennis Merritt (the book as a PDF is in the course’s Canvas page under Kursinformation).

Read mainly Sect. 2.1, and parts of Sect. 2.2 for simple interaction. Study also the Figure 2.1. It helps
to visualise the problem domain. You don’t need to implement any user interface, or any menu. Just
make sure that you can interact with the program at a basic level.

Q1: Add now one more bird category at the level of ‘laysun_albatross’, ‘black_footed_albatross’, and
‘rumpeter_swan’. You can choose a Swedish/European bird, or any other bird from the American
continent.

Remember that you task is to implement an expert system, think about the current expert system and
decide if you can add a Swedish, or European, bird easily, or if the current implementation only allows
adding American birds.

Q2: Test the predicate trace to see how Prolog behaves in Q1 in LAB1 and also the examples in
LAB 2. Spend some time on this task, please. Try to learn how Prolog behaves.

Q3: Test the meta predicates fidnall, bagoff, and setof, described in the SWI-Prolog
https://www.swi-prolog.org/pldoc/man?section=allsolutions As you know, it is not possible to have
predicates as inputs (terms) to other predicates, however, these meta predicates are exception to that
rule.

Q4: For the problem in Q1, use the predicate trace to demonstrate how unification works in Prolog.
Write a report on ½ A4 of text. Include screen dumps, as figures, to help you with the text. Imagine
that receiver of this report is a 1st year computer science student.

15
 4. Overview of the examination

*** Examination (svenska) ***

LAB1, en programmeringsuppgift som ska demonstreras, 0,5 hp, examinerar lärandemål 1 och 7, betyg
Underkänd (U) eller Godkänd (G).

LAB2, en programmeringsuppgift som ska demonstreras, och presenteras med en rapport, 2,5 hp,
examinerar lärandemål 1, 3, 4, 6 och 7, betyg Underkänd (U) eller Godkänd (G).

LAB3, en programmeringsuppgift som ska demonstreras, och presenteras med en rapport, 2 hp,
examinerar lärandemål 2, 6 och 7, betyg Underkänd (U) eller Godkänd (G).

LAB4, en programmeringsuppgift som ska demonstreras, och presenteras med en rapport, 2,5 hp,
examinerar lärandemål 2, 5, 6, 7 och 8, betyg Underkänd (U) eller Godkänd (G).

För slutbetyg Godkänd (G) krävs betyget Godkänd (G) i alla fyra laborationerna.

*** Examination (English) ***

LAB1, an assignment that is demonstrated to the teacher, 0.5 credits, examines the
learning outcomes 1 and 7, marks Fail (U) or Pass (G).

LAB2, an assignment that is presented with a report and a demonstration to the teacher, 2.5
credits, examines the learning outcomes 1, 3, 4, 6 and 7, marks Fail (U) or Pass (G).

LAB3, an assignment that is presented with a report and a demonstration to the teacher, 2
credits, examines the learning outcomes 2, 6 and 7, marks Fail (U) or Pass (G).

LAB4, an assignment that is presented with a report and a demonstration to the teacher, 2.5
credits, examines the learning outcomes 2, 5, 6, 7 and 8, marks Fail (U) or Pass (G).

For final grade Pass (G) the mark Pass (G) is required in all four laborations.

4.1. Learning outcomes

Lärandemål (på svenska)

1. analysera och definiera sats samt predikatlogik för implementation av agentmodeller,

2. förklara och tillämpa populationsbaserade agentmodeller med utgångpunkt från
evolutionära
algoritmer,
3. analysera och definiera de mer grundläggande metoderna inom oskarp logik, osäkerhet, och
resonemang,
4. förklara och tillämpa expertsystem för att lösa ett domänspecifikt problem,
5. förklara och tillämpa planering genom populationsbaserade agentmodeller,

16
6. förklara hur olika representationer av ett problem med avseende på ökad prestanda kan
bedömas, där representationen sker i form av antigen (i) logik, m.a.o. symboliskt, (ii) numerik
eller (iii) i kombination av dessa,
7. analysera och definiera ett givet problem, samt bestämma om det kan lösas med en av de
tekniker som ingår i denna kurs samt
8. genomföra en etik- och jämställdhetsanalys av ett agentbaserat AI_system.

Learning outcomes (in English)

After completing the course, the student shall be able to:

1. analyse and define propositional and predicate logic; and demonstrate how these theories
can be used in logic programming for solving problems and representation of agent models,
2. explain and apply population-based agent models, such as evolutionary algorithms and its
variations,
3. analyse and define the most representative methods in uncertainty, fuzzy logic, and
reasoning,
4. explain and apply expert systems for solving domain specific problems,
5. explain and apply planning through population-based agent models,
6. explain the means for knowledge representation with respect to performance, especially in
the context of different representation paradigms, such as (i) logic, that is symbolic, (ii)
numeric, and (iii) the combination of both,
7. analyse and define a given problem with the ambition of deciding if it can be addressed by
the methods covered in this course and also
8. carry out an ethical and gender equality analysis of an agent-based AI system.

17