Fairness, Accountability, Confidentiality and Transparency in AI

Course manual 2025/2026

Course content

Artificial Intelligence and data science are increasingly being used to power decision making in settings that are highly consequential for people and society. Algorithmic decision making promises great benefits across sectors and domains. At the same time, there are many high-profile stories about the potential harms of algorithms, for individuals, population groups, and society at large. With great power comes great responsibility. What does “responsibility” mean from a technical perspective? What are technical solutions available to those who develop and deploy AI, machine learning and data science to help mitigate the potential risks of algorithms?

The concepts we develop in Fairness, Accountability, Confidentiality, Transparency in Artificial Intelligence (FACT-AI) are aimed at students who have joined or are likely to join the developer community. They should help them to articulate what concepts like “responsibility” mean from a technical perspective, and to make informed decisions when assessing and addressing the potential risks of algorithms. 

This course will provide an overview of recent algorithmic approaches to improve fairness, transparency and privacy in AI. We will also discuss how such methods can be combined to improve accountability in algorithmic systems. We will discuss the technical challenges of implementing fairness, transparency and privacy techniques, as well as the challenges of anticipating how effective, in the long-run, such interventions are. We focus on both supervised learning and more recent systems such as LLMs and Agentic systems. Students will also discuss the environmental impacts of AI models and the  societal challenges associated with recent generative AI methods. 

This course will have a strong practical component: students will be exposed to state-of-the-art approaches and replicate/reproduce/extend a recently published paper in the area. 

Objectives

• The student is able to explain the main types of algorithmic harm, in general and in terms of concrete AI applications
• The student is able to justify the major notions of fairness, accountability, confidentiality, and transparency that have been proposed in the literature, with their strengths and weaknesses
• The student is able to describe with state-of-the-art algorithmic approaches to fairness, confidentiality and transparency
• The student is able is able to improve fairness, transparency or confidentiality in AI by evaluating a recently proposed algorithmic approach
• The student is able to evaluate and discuss the environmental impacts of current AI models

Teaching methods

• Lecture
• Seminar
• Supervision/feedback meeting
• Working independently on e.g. a project or thesis
• Presentation/symposium

Lectures and seminars: There are six lecture blocks for this course, one for each topic (Privacy, Transparency, Fairness, Accountability, Generative AI and Environment). Each lecture block will consist of (i) a general lecture on the topic, (ii) guest lecture by an invited speaker on the same topic.

Supervision/feedback meetings: You will have two onsite Practicums per week where you can ask your TA questions about your paper. You need to go to the Practicums that correspond to the paper you are reproducing. Each group will get 2x20 onsite supervision meetings each week. The final TA and supervision meetings times (which will depend on the paper you will be assigned to) will be communicated on January 5. Please check DataNose for specific times and rooms.

Quizzes: The contents taught in the lectures will be exercised and evaluated in a set of quizzes. These quizzes will be done through Canvas and will consist of approximately 15 multiple choice questions to be answered in 15 minutes.

Presentation/symposium: The final part of the project is a 10 minute presentation on your findings. This should be a summary of your written report and will take place during the last week of the course, on 30 January 2026 (exact times to be scheduled, any time from 9:00 to 15:00). Presentations will happen onsite at Science Park. You are expected to also attend the sessions of your colleagues and actively participate in the Q&A.

Learning activities

Attendance

This programme does not have requirements concerning attendance (OER part B).

Assessment

Assignments

The lack of reproducibility has been an ongoing issue in academic research. The goal of the FACT-AI course assignment is to expose students to state-of-the-art research in the FACT field(s) while assessing the reproducibility of existing work by reimplementing an algorithm, replicating and/or extending the experiments from the corresponding paper, and detailing findings in a report.

In this project you will implement an existing FACT algorithm in groups of 4. 

The assignment will have an impact of 85% in your final grade. You will be supported by TAs.

There will be two meetings with TAs per week. 

Groups should be formed until maximum Monday January 5 (14:59).

Fraud and plagiarism

The 'Regulations governing fraud and plagiarism for UvA students' applies to this course. This will be monitored carefully. Upon suspicion of fraud or plagiarism the Examinations Board of the programme will be informed. For the 'Regulations governing fraud and plagiarism for UvA students' see: www.student.uva.nl

Course structure

Contact information

Coordinator

• Fernando Pascoal Dos Santos