Certificate Course: Keynotes, Sessions and Speaker
Please find detailed information on all four days in the Keynote and Session descriptions below.
06 Nov `26, Kick-Off
Keynote | 06 Nov ´26, Kick-Off
Responsible AI - A Philosophical Perspective
Abstract: In this talk, I give an overview over philosophical basics on responsibility. I distinguish different notions of responsibility, including moral, leagal, and causal responsibility, and foward-looking and backward-looking responsibility. I then focus on the standard philosophical notion of moral responsibility, which is tied to notions of blameworthiness and praiseworthiness. The second focus of the talk concerns issues of responsibility in AI contexts. I discuss whether AI systems meet the conditions for moral responsibility (they do not) and how humans relying on AI support can maintain their responsibility. Finally, I investigate whether the narrow philosophical concept of moral responsibility is too narrow a concept for thinking about responsible AI, and if so, how the concept might reasonably be broadened.
Eva Schmidt
TU Dortmund / Lamarr Institute
Bio: Eva Schmidt is Professor of Theoretical Philosophy at TU Dortmund. She works in epistemology, the philosophy of action, and philosophy of mind. Her main interests are epistemic reasons and reasons for action; explainable artificial intelligence; and the epistemology and nature of perception. She is a PI and the Values Chair of the Lamarr Institute for Machine Learning and Artificial Intelligence.
Lecture | 06 Nov ´26, Kick-Off
Data-Centric Responsible AI: Automated Validation & Debugging
Abstract: Responsible AI depends not only on model design, but on the data and pipelines that shape model behavior and potentially introduce technical bias. This lecture introduces data-centric methods for building more reliable, accountable, and auditable AI systems. We will examine scalable data quality verification, production data validation for machine learning, interactive tools that suggest improvements to brittle data preparation code, and Shapley-value-based approaches for valuing and debugging training data. Across these topics, students will see how responsible AI becomes an operational practice: detecting harmful data issues early, understanding which data points matter, improving pipelines before deployment, and making model outcomes more traceable from data to decision.
Sebastian Schelter
TU Berlin / BIFOLD
Bio: Sebastian Schelter is a Full Professor at the Berlin Institute on the Foundations of Learning and Data (BIFOLD) and Technische Universität Berlin. His research focuses on the intersection of data engineering and machine learning, with the goal of lowering the technical bar to efficiently and responsibly work with data. The research of his group is accompanied by efficient and scalable open source implementations, many of which are applied in real world use cases, for example in the Amazon Web Services cloud and in large European e-commerce platforms. In the past, he has been an assistant professor at the University of Amsterdam, a faculty fellow at New York University, a senior applied scientist at Amazon Research and a research intern at Twitter and IBM Almaden in California. His research contributions have been recognized with an ACM SIGMOD Systems Award, an ACM SIGMOD Best Demo Runner Up Award, and a Best Paper Runner Up Award from the Table Representation Learning workshop at NeurIPS.
Lecture | 06 Nov ´26, Kick-Off
Identifying and Mitigating Bias in LLM Training Data: From EU AI Act Requirements to Practical Approaches
Abstract: Under the EU AI Act, providers and deployers of AI systems are legally obligated to identify and address bias against protected groups throughout the AI lifecycle in order to prevent unfair outcomes. This obligation is especially challenging for large language models (LLMs), whose textual training data is known to encode multifaceted biases, including harmful language and skewed demographic representations. Despite growing awareness of these risks, practical guidance on how to systematically detect and mitigate such biases remains scarce. This seminar offers a structured introduction to bias in LLMs, bridging regulatory, conceptual, and technical perspectives. We begin by establishing a joint taxonomy of bias with a focus on textual training data, before turning to the relevant requirements of the EU AI Act and the phases in the LLM lifecycle where bias can arise. As a concrete example for technical approaches to address these requirements, we will explore a comprehensive, configurable data bias detection and mitigation pipeline developed at Fraunhofer IAIS, which identifies and reduces bias in LLM training and fine-tuning data across various sensitive attributes. We further discuss the often complex effects that preprocessing interventions can have on downstream model behavior, and the practical challenges that remain when addressing bias in real-world LLM applications.
Rebekka Görge
TU Dortmund / Lamarr Institute
Bio: Rebekka Görge is a Senior Data Scientist at the Fraunhofer Institute for Intelligent Analysis and Information Systems (IAIS) in Germany and a member of the Lamarr Institute for Machine Learning and Artificial Intelligence. Her research focuses on trustworthy artificial intelligence, with particular emphasis on detecting and mitigating bias in large language models grounded in sociolinguistic foundations. As project lead of several research and industrial projects—including "Mission KI" (2024–2025), a large-scale initiative on trustworthy AI in Germany—she develops new approaches for the evaluation and auditing of AI systems. Beyond her project work, she advises public authorities on AI trustworthiness and serves as an active trainer within the Fraunhofer Big Data Alliance, teaching data science and trustworthy AI.
13 Nov `26, Governance
Lecture | 13 Nov `26, Governance
Beyond Rigid Specification: Responsible AI through Justifiability
Abstract, short: Responsible AI differs fundamentally from related concepts such as AI Ethics, AI Safety, and AI Governance. Where AI Ethics articulates normative principles, AI Safety pursues technical robustness, and AI Governance builds institutional structures, Responsible AI demands the integration of insights from all these fields across an AI system's entire lifecycle—from design through deployment to post-market monitoring—and its socio-technical context of use, in a way that enables proper and effective attributions of responsibility and accountability. This integrative ambition confronts a core tension: normative uncertainty—the absence of a normative ground truth—meets implementation pressure. Values conflict, trade-offs are ubiquitous, technological capabilities evolve rapidly, and abstract principles must translate into concrete design choices. Specifications that are too concrete become rigid and quickly outdated; principles that remain too abstract fail to guide actual development and deployment.
More Details
We propose to address this tension through justifiability rather than standardization. Drawing on practical reasoning (philosophy), public justification (democratic theory), and proceduralization (law), we argue for meta-standards: specifications of how decisions about AI systems are to be made and justified, rather than of what those decisions must be. On this view, a decision is justifiable if it results from a reasoning process that meets explicit procedural conditions—making it acceptable, in that it satisfies what can reasonably be demanded of decision-makers under uncertainty, and defensible, in that diligent adherence to the process counts in its favour when decisions are later contested. The result is dynamic, context-sensitive, and contestable governance that allows ongoing normative deliberation while maintaining accountability and transparency.
Our running case is an AI-based system that ranks university applicants—high-risk under the EU AI Act. Its design forces trade-offs between accuracy, privacy, and competing notions of fairness, and thereby between fundamental rights: non-discrimination, data protection, access to education, academic freedom. The Act responds with layered requirements—data governance (Art. 10), transparency (Art. 13), accuracy and robustness (Art. 15)—with human oversight (Art. 14) on top as the catch-all safeguard. Yet these layers are deeply interdependent, and the same trade-offs cut across them. Whether sensitive attributes may be processed at all (Art. 10) determines which biases can even be detected—privacy pulling against fairness. Mitigating detected bias may cost accuracy (Art. 15). And what "effective" oversight (Art. 14) concretely demands—what overseers must be able to notice, understand, and correct—depends both on the transparency the system provides (Art. 13) and on every design decision taken upstream. The Act stipulates each requirement individually but says little about how they are to be reconciled in the concrete case; no purely technical criterion settles that question. Complying thus ultimately means taking—and being able to justify—a stance on competing rights. This is what procedural-oriented meta-standards are for: they do not prescribe the resolution, but specify how a defensible one is reached, documented, and defended.
You will gain theoretical foundations and practical orientation for developing, deploying, regulating, and evaluating AI systems under normative uncertainty. Throughout, we make the case that effective Responsible AI demands integrated expertise across philosophy, law, computer science, and human factors—not in sequence, but in genuine collaboration. No background in philosophy or law is required; just bring a case from your own research or practice.
UNIT STRUCTURE
Part 1 — Responsible AI and the Case for Justifiability (Kevin Baum): Responsible AI as an integrative concept—distinguishing it from AI Ethics, Safety, and Governance; philosophical foundations of responsibility and accountability; the core problem of normative uncertainty under implementation pressure; and justifiability as an alternative to rigid standardization.
Part 2 — The EU AI Act's Safety Architecture (Anne Lauber-Rönsberg): How the Act's layered requirements interact, with human oversight as the underspecified safety net; legal proceduralization—regulating the how rather than the what—as a response to regulatory uncertainty; and the tension between concrete compliance and adaptive flexibility—developed along the running case of AI-based university admission.
Part 3 — Justifiability Applied: Human Oversight and AI Agents (joint): How the framework addresses the Act's interpretive challenges; what a justifiable decision process must contain—comprehensive investigation, proportionate weighing, contestable documentation, and good faith; the special case of AI agents and the relocation of oversight responsibilities; and connections to broader normative theory.
Part 4 — Your Turn (interactive): Breakout sessions in which participants apply justifiability thinking to their own research domain or practice (no case of your own? pick one of ours: an AI-based insulin pump, a factory patrol robot, or an automated recruitment system), followed by a moderated tandem panel discussion on limits, open challenges, and alternative approaches.
Anne Lauber-Rönsberg
TU Dresden / Scads.AI
Bio: Prof. Dr. Anne Lauber-Rönsberg is Professor of Civil Law, Intellectual Property Law (in particular Copyright), Media and Data Protection Law at TU Dresden and Managing Director of the Institute for International Law, Intellectual Property and Technology Law. She is a Principal Investigator at the Center for Scalable Data Analytics and Artificial Intelligence (ScaDS.AI) Dresden/Leipzig, where her research addresses the legal framework of AI. Her work focuses on AI regulation and governance, data protection, and intellectual property in the digital transformation; she is a member of Plattform Lernende Systeme and the German Council for Scientific Information Infrastructures (RfII).
Kevin Baum
German Research Center for Artificial Intelligence (DFKI)
Bio: Dr. Kevin Baum is a philosopher and computer scientist. He leads the Responsible AI and Machine Ethics (RAIME) research group at the German Research Center for Artificial Intelligence (DFKI) and co-leads CERTAIN, the Center for European Research in Trusted AI. His research focuses on AI ethics, human oversight of AI, machine ethics, and the justifiability of algorithmic decision-making; he regularly advises policymakers on AI governance. He is also a Research Associate at the Oxford Internet Institute, Senior Scientist at the Hamburg University of Technology, and co-founder of the nonprofit Algoright e.V.
Lecture | 13 Nov `26, Governance
Operationalisierung der KI-Verordnung
Abstract: tba
Maximilian Poretschkin
Fraunhofer IAIS / Lamarr Institute
Bio: Dr. Maximilian Poretschkin leads the team “AI assurance and certification” at Fraunhofer IAIS, where he develops testing procedures, methods, and tools for AI systems. He also heads the working group “AI testing and certification” as part of the DIN standardization roadmap and is involved in the development of various standards in the field of trustworthy AI. Dr. Poretschkin studied physics in Bonn and Amsterdam and, after completing his Ph.D. in mathematical physics, was a postdoc at the University of Pennsylvania in Philadelphia. Following a stint at the strategy consulting firm Bain & Company, he has been conducting research at Fraunhofer IAIS since 2018.
Lecture | 13 Nov `26, Governance
From Nebulous Clouds to Solid Ground: Legal Responsibility When AI Leaves the Data Center
Abstract: Contemporary AI usage is predominantly defined by reliance on cloud-based interfaces provided by large corporations, where the legality of training data and model weights remains in limbo amidst ongoing copyright litigation. This talk contrasts this nebulous legal landscape with the implications of executing models locally on user-owned hardware. We map and evaluate existing legal obligations across three key actors: the (open-weight) model creator, the cloud AI provider, and the local end-user.
We examine how liability and legal scrutiny shift regarding copyright infringement, data protection (GDPR), and safety compliance when inference moves from the cloud to the edge. Specifically, we evaluate whether local deployment mitigates regulatory exposure or merely redistributes it among the parties involved. By dissecting the distribution of legal responsibility, this session complements the course's focus on operationalizing the EU AI Act and defining responsible AI as a systemic concept, clarifying where accountability resides when the infrastructure of AI deployment changes.
Michael Tiemann
Tübingen AI Center
Bio: Dr. Michael Tiemann received his PhD from the Max Planck Institute for Intelligent Systems and the University of Tübingen. Following eight years as a research scientist in industry, he is currently a Postdoctoral Researcher at the Tübingen AI Center. There, he contributes to the Minerva project, providing HPC expertise to the AI community, including guidance on regulations for ethical and responsible AI. He is also the co-founder of DenkBox GmbH, a company specializing in on-premise AI appliances for the local execution of open-weight models. This talk explores the intersection of these two occupations, though Dr. Tiemann notes that his legal insights are self-taught rather than formally qualified.
20 Nov `26, Fairness and privacy in practice: Concepts, trade-offs and challenges
Lecture | 20 Nov ´26, Fairness and privacy in practice: Concepts, trade-offs and challenges
Algorithmic Fairness
Abstract: Prediction algorithms fueled by Machine Learning (ML) and Artificial Intelligence (AI) are increasingly used to guide decisions in various domains, including public health, criminal justice, credit scoring and profiling of job seekers. By supporting or even substituting human judgment, algorithmic decision-making (ADM) promises to increase the efficiency and effectiveness of institutional processes. At the same time, however, a number of infamous ADM applications demonstrate that algorithms may also foster discrimination and amplify existing disparities present in the model's training data. This course unit gives a comprehensive overview of the burgeoning field of algorithmic fairness (fairML), from its inception to recent advances. This includes discussing different types of biases in data, a systematic overview of fairness notions, metrics, and bias mitigation techniques as well as an outlook on how fairness considerations map to data science practices more broadly. A key aspect of the course will be situating fairness concepts within the broader, multi-layered structure of algorithmic decision-making systems and their various objectives, paired with a focus on ADM examples from the European context.
Christoph Kern
LMU / Munich Center for Machine Learning (MCML)
Bio: Christoph Kern is Junior Professor of Social Data Science and Statistical Learning at the Ludwig-Maximilians-University of Munich and Project Director at the Mannheim Centre for European Social Research (MZES). He received his PhD in social science (Dr. rer. pol.) from the University of Duisburg-Essen in 2016. Before joining LMU Munich, he was a Post-Doctoral Researcher at the Professorship for Statistics and Methodology at the University of Mannheim and Research Assistant Professor at the Joint Program in Survey Methodology (JPSM) at the University of Maryland. His work focuses on the reliable use of machine learning methods and new data sources in social science, survey research, and algorithmic fairness.
Jan Simson
LMU / Munich Center for Machine Learning (MCML)
Bio: Dr. Jan Simson is a Postdoctoral Researcher at the Department of Statistics at the University of Mannheim / the LMU Munich, with affiliations at the Konrad Zuse School for Excellence in Reliable AI and the Munich Center for Machine Learning (MCML).
Lecture | 20 Nov ´26, Fairness and privacy in practice: Concepts, trade-offs and challenges
Privacy, Utility and Trust in Synthetic Biomedical Data
Abstract: Synthetic data generation is increasingly positioned as a technical solution to privacy in biomedical AI, yet the assumption that it eliminates risk deserves critical investigation. This lecture uses synthetic bulk RNA-sequencing (RNA-seq) data generation as a concrete case study to examine trade-offs that arise broadly across privacy-preserving generative models. We will draw on real-world challenge design experience (Health Privacy Challenge) and community feedback to understand how comprehensive benchmarking serves as a foundational step toward building trustworthy AI systems.
Hakime Öztürk
EMBL, Heidelberg
Bio: Hakime Öztürk is a research scientist at the European Molecular Biology Laboratory (EMBL), where she currently focuses on evaluating trustworthy machine learning methods for healthcare applications as a part of European Lighthouse on Safe and Secure AI. She completed her postdoctoral research at the German Cancer Research Center (DKFZ), working with machine learning models to represent cancer and population cohorts. Her My PhD thesis at Bogazici University was on NLP-based representation learning for drug-target interaction modeling, earning the Best PhD Dissertation Award in 2020.
Lecture | 20 Nov ´26, Fairness and privacy in practice: Concepts, trade-offs and challenges
Privacy and Security Challenges in Agentic AI
Abstract: Agentic AI systems - autonomous agents that access external tools, data sources, and collaborate with other agents - introduce novel privacy and security challenges beyond traditional AI applications. This talk examines how agent autonomy, tool invocation, and multi-agent collaboration create new attack surfaces and privacy risks. We explore critical vulnerabilities including unauthorized data access through tool misuse, information leakage across agent interactions, prompt injection attacks that manipulate agent behaviour, and challenges in maintaining data governance when agents operate across distributed systems. Drawing on recent developments in standardized protocols like Anthrophic's Model Context Protocol (MCP), we discuss practical mitigation strategies including authentication mechanisms, access controls, audit trails, and privacy-preserving architectures. Participants will gain technical insights and frameworks for addressing privacy and security risks when deploying agentic AI in organizational contexts.
Annika Hannemann
Swiss Centre for Responsible AI
Bio: Anika Hannemann is the Head of Research at the Swiss Centre for Responsible AI, an interdisciplinary initiative by all cantonal universities of Zürich, and a senior researcher at Zürich University of Applied Sciences. A computer scientist by training, she brings experience from both academia and industry. In 2025, she earned her doctorate at Leipzig University with research on privacy-preserving machine learning for distributed systems. Her current work focuses on Responsible AI, with a particular emphasis on privacy and security in AI systems. Passionate about connecting research and practice, she is keen on bridging the gap between academia and industry.
27 Nov `26, From explainable to ethical AI
Keynote | 27 Nov `26, From explainable to ethical AI
From Explainable to Trustworthy AI
Abstract: Traditional engineered systems, such as an airplane’s wing or a bridge’s foundation, are built from modular, transparent components, each with a well-defined and independently verifiable function. In contrast, modern AI models are developed end-to-end through data-driven optimization. While this often yields powerful capabilities, it also produces opaque systems whose internal logic is difficult to interpret or validate. This talk introduces new methods that bring engineering-style inspection to AI models, enabling a deeper understanding of their internal representations and behaviors. In large language models, for example, we can identify specialized attention heads: in-context heads, which interpret instructions and retrieve relevant contextual information through retrieval-augmentation, and parametric heads, which encode relational knowledge about entities. This fine-grained insight not only illuminates how LLMs reason but also supports practical tools for detecting and mitigating hallucinations, advancing the development of safer and more trustworthy AI.
Sebastian Schelter
Fraunhofer Heinrich Hertz Institute (HHI) / BIFOLD
Bio: Wojciech Samek is a professor in the department of Electrical Engineering and Computer Science at the Technical University of Berlin and is jointly heading the department of Artificial Intelligence and the Explainable AI Group at Fraunhofer Heinrich Hertz Institute (HHI), Berlin, Germany. He studied computer science at Humboldt University of Berlin from 2004 to 2010, was visiting researcher at NASA Ames Research Center, CA, USA, and received the Ph.D. degree in machine learning from the Technische Universität Berlin in 2014. He is associated faculty at the ELLIS Unit Berlin and the DFG Graduate School BIOQIC, and member of the scientific advisory board of IDEAS NCBR. Furthermore, he is a senior editor of IEEE TNNLS, an editorial board member of Pattern Recognition, and an elected member of the IEEE MLSP Technical Committee. He is recipient of multiple best paper awards, including the 2020 Pattern Recognition Best Paper Award, and part of the expert group developing the ISO/IEC MPEG-17 NNC standard. He is the leading editor of the Springer book “Explainable AI: Interpreting, Explaining and Visualizing Deep Learning” and organizer of various special sessions, workshops and tutorials on topics such as explainable AI, neural network compression, and federated learning. He has co-authored more than 150 peer-reviewed journal and conference papers; some of them listed by Thomson Reuters as “Highly Cited Papers” (i.e., top 1%) in the field of Engineering.
Lecture | 27 Nov `26, From explainable to ethical AI
Responsible AI in Protein Design: Navigating Innovation and Risk Mitigation
Abstract: The field of protein design has seen significant advancements in recent years representing chances in design of protein medicines and biomaterials. Beyond improvements in laboratory workflows, computational capabilities have increased drastically, particularly with the integration of Artificial Intelligence (AI) which allows for the design of proteins de novo without relation to natural proteins. However, these rapid technological shifts bring inherent risks and limitations. Our research focuses on responsible AI in applied protein design centering on both the computational tools and the scientists using these. Specifically, we analyze trust, perception, regulation and safety. We have assessed this with a benchmark analysis of toxicity in de novo protein design using AI-based tools. We aim to highlight the criticality of these issues; in a field where the design of novel neurotoxins or selective toxins is possible, the consequences of such actions must be rigorously addressed. Our findings suggest a lack of awareness and specialized knowledge regarding risk mitigation within the scientific community. Furthermore, the responsibility of scientists to communicate these risks to the general public remains unfulfilled. In this quickly moving field between chances and risks, we want to use this platform to discuss responsible usage of biodesign tools. We believe that our work is contributing significantly in this format to close knowledge gaps and fostering necessary awareness.
Clara Schoeder
Leipzig University / Scads.AI
Bio: Prof. Dr. Clara Schoeder is Junior Professor for the Development of Novel Immunotherapeutic Drugs at Leipzig University, Faculty of Medicine, a position she has held since August 2023. She also serves as Junior Research Group Leader at the Institute of Drug Discovery at Leipzig University, Faculty of Medicine, since 2021. She studied Pharmacy at the University of Kiel from 2007 to 2012 and became a licensed pharmacist in 2013. She received her PhD in Pharmacy from the University of Bonn in 2017, where she worked in the laboratory of Christa E. Müller as part of the GRK1873 research training group. She then completed a postdoctoral fellowship from 2018 to 2021 in the laboratories of Jens Meiler and James E. Crowe, Jr. at Vanderbilt University, Nashville. Her work focuses on the development of novel immunotherapeutic drugs.
Hermann Diebel-Fischer
TU Dresden / Scads.AI
Bio: Dr. Hermann Diebel-Fischer is a Research Associate at the Center for Scalable Data Analytics and Artificial Intelligence (ScaDS.AI), where he has led the team's work in the field of ethics since April 2021. He studied Protestant theology and English/American studies, completing a first state examination for teaching, and also studied Economics, graduating with a Bachelor of Science. He defended his dissertation in March 2018 and subsequently held a postdoctoral position at the Faculty of Theology at the University of Rostock from April 2018 to March 2021. From October 2020 to September 2023, he was a Young Academy Fellow and Associate Young Academy Fellow at the Academy of Sciences and Humanities in Hamburg. His work focuses on questions of ethics at the intersection of data analytics and artificial intelligence.
Anne Schmieder
TU Dresden / Scads.AI
Bio: Anne Schmieder is a doctoral candidate and member of ScaDS.AI in the department of Responsible AI, a position she has held since October 2025. She is also a scientific member of the Schoeder Lab at the Institute of Drug Discovery, Leipzig University, since April 2025. She studied Business Informatics at the University of Leipzig from 2019 to 2022, graduating with a Bachelor of Science, and went on to study Bioinformatics at the University of Leipzig from 2022 to 2024. She has participated in two Rosetta workshops, held in Leipzig and Davis, and has presented at the Biosafety conference "Generative AI for Biological Protein Design" as well as a security congress in Munich.
Lecture | 27 Nov `26, From explainable to ethical AI
Ethics as an added value in responsible AI development
Abstract: Ethics in AI development is often treated primarily as a compliance requirement. In contrast, we propose framing ethics as a strategic source of value creation. When embedded systematically into development processes, ethical reflection can improve system quality, increase robustness, and strengthen trustworthiness. During the session, we aim to engage participants in examining and learning how ethics can be eGectively operationalized as a driver of added value in AI systems.
Mihai Maftei
German Research Center for Artificial Intelligence (DFKI)
Bio: Mihai Maftei is a researcher on responsible AI development within the Ethics Team at the German Research Center for Artificial Intelligence (DFKI) and a member of the steering committee of the Ethical and Trustworthy Artificial and Machine Intelligence BDVA Taskforce (etami-BDVA). He is also an EU Expert Evaluator for AI-funded research through the European Research Executive Agency at the European Commission.
Hartmut Hilpert
German Research Center for Artificial Intelligence (DFKI)
Bio: Hartmut Hilpert is a research assistant in the Ethics Team at the German Research Center for Artificial Intelligence (DFKI) in Kaiserslautern and a member of the Special Interest Group on Ethical, Legal and Social Aspects of AI at CAIRNE.