nl en

Modern Game AI Algorithms


Admission requirements

Not applicable.


In the last years, Game AI has left the nerd interest area and is now known as incubator for many of the most important AI algorithms of our time. Almost the same methods are employed for beating the world champion in Go as well as solving very complex Chemistry problems for the first time in an automated fashion. We have recently witnessed programs beating human professionals in StarCraft and MOBA games and it is one part of the course to learn what kind of algorithms were used to do it.
Game playing is one pillar of modern Game AI, the others are procedurally generating content (PCG) and modelling players, and of course all these 3 can be combined in various ways. Of course, these methods have ties also to other fields (e.g., behavior trees originated in robotics) and are not constrained to game uses. This renders the course a good introduction also to general AI.
Modelling players is connected to analysing game data, which can have different goals, e.g. to balance, to modify difficulty, to adapt content, or simply understand how the set of players is composed or what they do.

Course objectives

  • Getting an overview over and a fundamental understanding of the algorithms that drive Game AI and thus also AI progress, e.g. Monte Carlo Tree Search, Deep (Reinforcement) Learning, and behavior trees;

  • Learn how to employ these methods especially in the context of Procedural Content Generation and General Video Game Playing;

  • Learn how to use human feedback for generation and adaptation processes;

  • Obtain practical research experience, learn how to learn from experiments - be warned: the new methods as e.g. DQN are very powerful, but they require a lot of experience to be applied successfully.


The most recent timetable can be found at the Computer Science (MSc) student website.

Mode of instruction

Lectures and practical sessions. The practicals (labs) are for starting work on the assignments and getting feedback, I highly recommend to use that in this way, if you defer everything and try to find out on your own later, you will need much more time.

Course load

Total hours of study: 168 hrs. (= 6 EC)
Lectures: 26:00 hrs.
Practicals: 26:00 hrs.
Assignments: 116:00 hrs.

Assessment method

  • Individual assignments

  • Group work assignments

  • No exam

The grade will be based solely on the assignment performance, we will have around 3-4 assignments of increasing complexity, at least one of these is individual.

Reading list


  • You have to sign up for courses and exams (including retakes) in uSis. Check this link for information about how to register for courses.


Lecturer: dr. M. Preuss


While the course touches on many algorithms, it is rather practically oriented. The most important questions are a) how does it work, and b) how can it be used in practice. Due to the availability of suitable software, group assignments shall mostly use Java or Python. Have a look at IEEE CoG 2021 Competitions to get an idea of what we deal with.