Explainable AI

FaculteitFaculty of Science

NiveauMaster

Studiejaar2026-2027

Beschrijving

Course goals

After completing the course, the student

1) understands the need for the use of Explainable AI and the importance of explainability for AI trustworthiness and ethical alignment. (assessed in project 1 and exam).

2) can explain the different types of explanations, and the difference between interpretability and explainability (assessed in the exam).

3) can justify the selection of ML explainability and explainability-by-design methods by critically evaluating their theoretical foundations, underlying algorithms, and inherent strengths and limitations. (assessed in both projects and exam).

4) is able to implement Explainable AI methods and critically evaluate them based on the specific data and problem being addressed .
(assessed in both projects).

5) is prepared to do research on the topic of Explainable AI, including designing and implementing appropriate methods and algorithms, conducting user studies or experiments to collect and analyse quantitative and qualitative data, visualising and interpreting results, and effectively communicating findings through scientific writing and oral presentation. (assessed in both projects).

Assessment

exam (40% of the final mark)
two projects (each 30% of the final mark)

To pass the course, your final grade must be at least 5.5, with a minimum grade of 5.0 for the exam and 4.0 or higher for each project.

Students with a final grade between 4.0 and 5.5, and project grades of 4.0 or higher, are eligible for a retake covering both theoretical and project-related topics.

Content

In recent years, there has been a significant surge in the capabilities and uses of Artificial Intelligence (AI) and Machine Learning (ML). AI and ML are more and more supporting us in making decisions that affect all of us, and are increasingly used, in domains ranging from finance to healthcare, by both experts and non-expert humans.
Even though ML models are usually seen as “black boxes”, there are several methods to interpret and/or explain the decisions of a model. Explainable AI (XAI) methodologies help to get a better understanding of the ML models as well as the data, and to build AI and ML systems whose internal mechanisms are easy to understand and ensure their trustworthiness, ethical alignment, and conformity with societal norms and values.

In this course, we provide a comprehensive overview of the state-of-the-art XAI methods and of the critical aspects of developing AI systems that are not only explainable but also transparent and responsible within social settings.
The course covers a wide spectrum of topics, including methods to explain state-of-the-art ML models, human-centric and explainable-by-design approaches, and ethical and societal implications of AI and ML development.

Course overview in brief:

introduction to Explainable AI
Machine Learning explainability (e.g., saliency maps, example-based, concept-based, attention-based, etcetera)
explainability-by-design and social explainability (e.g., explainable agents, justifiability, trust and values, etcetera)
assessment of explanations (e.g, quantitative metrics, human criteria)

Course form
Lectures, projects.

Reviews0 reviews

Nog geen reviews voor dit vak. Wees de eerste!

Heb jij dit vak gevolgd?

Deel je ervaring met toekomstige studenten. Inloggen met je Universiteit Utrecht mailadres duurt één minuut.

Schrijf een review