Dr. Tarek R. Besold (Eindhoven University of Technology, Netherlands and Sony AI Lab, Barcelona)
The term “trustworthy AI” has become increasingly popular over the last years – from academic research and publications, lawmakers and regulators, to the press and the general public. In this course we have a look at what dimensions are involved in making an AI system trustworthy, what are the scientific underpinnings and the current state of the art in the corresponding fields of academic and industry R&D, and what are the regulatory and market mechanisms which aim to make sure that end products indeed meet a certain (minimum) level of trustworthiness. We will see that the concept of “trust” is multifaceted and spans across several criteria some of which may well be in conflict with each other, requiring (ideally: conscious) tradeoffs between those mutually not fully compatible aspects.
We will visit the foundations of the explainability and interpretability of AI systems, of privacy-preservation, of fairness/bias mitigation, of security, and of safety and discuss these from technological, regulatory and societal perspectives. After successful completion of the course the participants will be able to take conscious decisions about tradeoffs between different sub-dimensions of trustworthiness of AI systems, can contextualize the question of trustworthy AI in the relevant European regulatory discourse, and understand the mechanisms which are being put in place to assure that consumers do not have to worry about the (un)trustworthiness of AI systems.
Week 1 (3 sessions)
Tuesday, 23/05, 16:30-18:30
- Introduction to Trustworthy AI – Mapping the Landscape
Wednesday, 24/05, 16:30-18:30
Thursday, 25/05, 08:30-10:30
- Explainable/Interpretable AI
Week 2 (2 sessions)
Tuesday, 30/05, 16:30-18:30
- Fairness/Bias (continued)
- Safety (Functional Safety, Human Oversight)
Wednesday, 31/05, 16:30-18:30
- Regulation & Testing/Certification
- The EU AI Act
- The Trustworthy AI Framework of the EU HLEG
- AI Standardization
Exam instructions and modality.
In the final evaluation we are having a look at the intersection between explainable AI and privacy-preservation or explainable AI and cyber security.
Select one paper from the list of papers by putting your student ID in one of the free fields to the right of the paper title and category. https://docs.google.com/spreadsheets/d/1WFPvPaGQrXQanb8O8T1WPF_vpjyiAyicyreDc9t1pY4/edit?usp=sharing
Please note: Each paper can at most be chosen by four people; once there are four IDs then that paper is "completely occupied" and you have to pick one of the remaining papers with open fields.
Write a report about your chosen paper (suggested length is not shorter than 3 pages, one column, single-line spacing, 10pt font).
The report should summarize the key technical contributions of the paper in such a way that any computer scientist (i.e., also if not being an expert in XAI and privacy-preservation/cyber security) can follow the description.
Additionally, you should give your view on the paper, discussing the potential relevance of the results, advantages/disadvantages, etc.
Finally, the report should feature a section in which you put the paper and its contribution into the wider context of trustworthy AI, discussing potential trade-offs the presented technique(s) require as regards different dimensions of trustworthiness (e.g., thinking of the HLEG Trustworthy AI criteria and/or the trade-offs between different aspects of trustworthiness we also mentioned in the lecture).
The assignment is individual, i.e., whilst you can -- of course! -- discuss with fellow students your report must be written be you alone.
Please send your reports to email@example.com and to firstname.lastname@example.org.
The deadline for submissions is Wednesday, 28/06, 23:59 CEST.
At the top of your submission, please include your name, student ID and what master's programme you are in, as well as whether you are taking this module to gain credits for other training activities (i.e., OTHER) or for the whole ATCS exam (i.e., EXAM).
Recommended table of content:
- Overview of the paper
- Trustworthy AI context
- Personal evaluation (advantages/disadvantages, etc.)
The following papers are recommended readings for everyone:
- S Ali, T Abuhmed, S El-Sappagh, K Muhammad, JM Alonso-Moral, R Confalonieri, R Guidotti, J Del Ser, N Díaz-Rodríguez, F Herrera. Explainable Artificial Intelligence (XAI): What we know and what is left to attain Trustworthy Artificial Intelligence. Information Fusion, 101805, 2023. https://www.sciencedirect.com/science/article/pii/S1566253523001148
- Arrieta, A. B., Díaz-Rodríguez, N., Del Ser, J., Bennetot, A., Tabik, S., Barbado, A., García, S., Gil-López, S., Molina, D., Benjamins, R., et al. (2020). Explainable artificial intelligence (xai): Concepts, taxonomies, opportunities and challenges toward responsible ai. Information Fusion, 2020. https://www.sciencedirect.com/science/article/pii/S1566253519308103