Societal Sessions at Obertauern 2026
This page outlines the themes and discussion questions for the Societal Sessions at Obertauern 2026, part of the Machine Learning workshop of the German Academic Scholarship Foundation. These sessions explore the social and ethical dimensions of AI and machine learning — regulation, healthcare, misinformation, and the future of work. Each theme is designed as a participant-led discussion with guiding questions and interdisciplinary perspectives.
Your group’s job is not to lecture — it is to catalyse a 60-minute conversation that draws on the full disciplinary range of the room. The advice for running a session is the same for every group, so we give it once below. Each topic then adds only its own background, discussion points, and a concrete case to open with.
📚 Topics Covered
- The EU AI Act – Comparing Regulation in the EU, China, and the US
- AI in Healthcare & Medicine – Ethical and Regulatory Concerns
- AI for Fake News Generation and Detection
- Impact of AI and Automation on the Future of Work
Running Your Session
Every group has the same 60 minutes and the same goal: get the room talking. A structure that works well:
- ~10 min — framing: One or two group members set up the topic — what’s at stake and the two or three fault lines you want the room to focus on. Keep it tight; the room needs time to talk, not just listen.
- ~40 min — discussion: Drive the conversation with your discussion points. Don’t try to cover everything — pick the two or three the group finds genuinely contentious and let those run.
- ~10 min — synthesis: Name the tensions that surfaced and what stayed unresolved, rather than smoothing it over. These topics rarely end in consensus — that’s fine.
A few things that make these sessions work, whichever topic you have:
- Choose your angle. You are strongly encouraged to focus on specific case studies or sub-questions you find relevant rather than surveying the whole field — engagement comes from specificity.
- Open with a concrete case, not “what do you think?” A specific scenario (we suggest one under each topic) gives everyone something to argue about straight away.
- Draw in the whole room. These topics reward disciplinary range — law, medicine, philosophy, economics, and the sciences each pull in different directions. Put those perspectives into dialogue rather than letting one dominate.
- Push for specifics. When the discussion drifts to “is this good or bad overall?”, pull it back: which provision, for whom, compared to what alternative? And resist “the technology is inevitable” fatalism — ask who makes the choices and what alternatives exist.
- Ground it in solid sources. Build your framing on reputable, current material — a plain-language summary from an established institute, a regulator or standards body, or a well-documented case — rather than the first search result. We’re happy to point you toward good sources for your topic if you ask.
1. The EU AI Act – Comparing Regulation in the EU, China, and the US
The EU Artificial Intelligence Act is the world’s first comprehensive legal framework for AI. It entered into force in August 2024 and has been rolling out in phases: prohibitions on the highest-risk practices (social scoring, real-time biometric surveillance, emotion recognition in workplaces and schools) applied from February 2025; rules for general-purpose AI models from August 2025; and requirements for high-risk systems — in healthcare, employment, education, law enforcement, and critical infrastructure — come into full effect in August 2026. By the time of this school, the Act is no longer a policy proposal but an active regulatory reality that companies are adapting to right now.
How does this approach compare to China’s more state-directed framework or the United States’ sector-by-sector, largely voluntary model? This topic invites analysis of how different regions are shaping AI’s future through regulation. You might explore enforcement structures, cross-border compliance challenges, or how regulatory choices reflect different political and cultural values.
Potential Discussion Points:
- Risk-Based Classification: The AI Act categorises systems as unacceptable, high-risk, limited risk, or minimal risk. What falls into each category, and why? Where are the contested boundaries?
- Prohibited Practices: What is banned outright — and why were those specific uses deemed unacceptable risks rather than just tightly regulated?
- High-Risk Systems: What requirements apply — risk management, human oversight, transparency, data governance, cybersecurity — and how do organisations comply in practice?
- General-Purpose AI: How does the Act approach foundation models and LLMs, which don’t fit neatly into the risk categories designed for narrow applications?
- Business and Innovation Impact: How does the regulation affect companies inside and outside the EU (extraterritorial reach)? What is the cost of compliance, and what are the penalties for non-compliance?
- Global Influence: Is the EU AI Act shaping regulation elsewhere — a “Brussels Effect” for AI? How are the US and China responding?
- Fundamental Rights: How does the Act interact with existing rights (privacy, non-discrimination)? Where might it fall short?
Opening case: “A European employer uses an AI tool to score candidates’ recorded video interviews for ‘communication skills’ and ‘enthusiasm.’ The vendor is based in the US and insists the tool is merely ‘decision support.’ Is this prohibited, high-risk, or limited-risk under the Act — and if the vendor sits outside the EU, who actually enforces the answer?”
2. AI in Healthcare & Medicine – Ethical and Regulatory Concerns
AI is rapidly transforming healthcare — from clinical diagnostics to drug discovery to hospital logistics. The potential is substantial: faster diagnoses, more personalised treatments, better prediction of patient deterioration. But as algorithms become embedded in medical decision-making, they raise critical questions: How do we ensure patient safety when the model is a black box? Who is responsible when an AI-driven diagnosis is wrong? How do we address bias baked into training data?
This topic invites exploration of how AI is reshaping medicine, and where ethics, regulation, and societal values come into tension with clinical innovation. You could focus on a specific application domain, a regulatory controversy, or a broader question about how healthcare AI should be governed.
Potential Discussion Points:
- Concrete Applications:
- Diagnostics: AI interpreting medical images (radiology, pathology, dermatology) — where is performance now, and where do errors cluster?
- Drug Discovery: AI accelerating molecular screening and clinical trial design.
- Predictive Analytics: Forecasting deterioration, readmissions, or disease outbreaks.
- Surgical Robotics: Precision-assisted procedures.
- Administrative Automation: Scheduling, coding, documentation — lower stakes but high volume.
- The Black Box Problem: How do we build trust in models whose reasoning isn’t transparent? What does “explainability” actually require in a clinical setting?
- Bias and Fairness: Models trained on unrepresentative data amplify existing health disparities. What does equitable AI in healthcare actually require?
- Data Privacy: Patient data is sensitive and regulated (GDPR, national health data laws). How do we train effective models while protecting privacy?
- Accountability and Liability: When an AI recommendation contributes to patient harm, who is responsible — the developer, the hospital, the clinician who followed it?
- Human-AI Collaboration: Will AI replace clinicians or augment them? What happens to diagnostic skills if clinicians increasingly defer to models?
- Regulatory Approval: How do health regulators evaluate AI medical devices? What standards apply, and are they adequate for continuously updated models?
Opening case: “A hospital deploys an AI model that predicts which patients will deteriorate and should be prioritised for intensive care. It was trained on data from a wealthier region and systematically under-flags patients from the hospital’s poorer catchment area — but its recommendations are logged as the official basis for bed allocation. Who is accountable for the patients it overlooks, and should the hospital have deployed it at all without local validation?”
3. AI for Fake News Generation and Detection
Generative AI has dramatically lowered the barrier to producing convincing fake content — fabricated articles, synthetic images, deepfake videos, voice clones. This makes it harder to distinguish authentic from fabricated information, and easier to manufacture persuasive disinformation at scale. Recent election cycles have provided concrete evidence of how these tools are used — and of the limits of both human and automated detection.
This raises profound questions: How do we protect public discourse from automated manipulation? What are the responsibilities of platforms, governments, and individuals? And is the technical arms race between generation and detection winnable?
Potential Discussion Points:
- The Changed Landscape: How has generative AI shifted the economics of disinformation? What does it now cost to produce convincing fake content at scale, compared to five years ago?
- Deepfakes and Synthetic Media: What specific threats do deepfake videos and voice clones pose — to individuals, public figures, elections, and institutional trust?
- The Detection Challenge: What approaches exist for detecting AI-generated content — technical signatures, watermarking, content provenance (e.g., C2PA standards)? How reliable are they, and what is the arms-race dynamic?
- Societal and Psychological Impact: How does widespread synthetic media affect trust in authentic evidence? What are the downstream effects on public discourse and democratic participation?
- Platform and Regulatory Responses: What obligations do platforms have? What does the EU’s Digital Services Act require? Are voluntary commitments sufficient?
- Media Literacy: What can individuals do? How should education systems respond?
- Ethical Tensions: How do we balance free expression with the need to suppress harmful disinformation? Who decides what counts as harmful?
Opening case: “Two days before an election, a voice clone of a party leader appears in a robocall telling supporters the vote has been moved to a different day; platforms take six hours to act. Separately, a detection tool wrongly flags a genuine campaign video as AI-generated, and it gets throttled. Which failure is worse — the fake that spread or the real content that was suppressed — and who should be trusted to make that call in real time?”
4. Impact of AI and Automation on the Future of Work
AI and automation are restructuring work in ways that are increasingly concrete — no longer hypothetical. Large language models have made routine cognitive tasks (drafting, summarising, translating, coding) significantly faster or cheaper. Robotic automation continues to transform logistics and manufacturing. The central questions are shifting from “will this happen?” to “how do we respond?”
But the picture is not simply one of displacement. Some roles are being augmented rather than replaced; new kinds of work are emerging; and the distributional effects — who benefits and who loses — are uneven across sectors, geographies, and skill levels.
Potential Discussion Points:
- Displacement vs. Creation vs. Transformation: Which types of work are most exposed to automation right now? Which are being augmented? What genuinely new roles are emerging, and who can access them?
- The Cognitive Task Shift: LLMs have extended automation into white-collar and knowledge work in ways earlier waves of automation did not. What does this mean for professions like law, medicine, journalism, and academia?
- Skills and Education: What capabilities become more valuable as AI handles routine tasks — and are educational systems adapting fast enough? What does lifelong learning actually look like in practice?
- Inequality: Are the gains from AI productivity broadly shared, or do they accrue primarily to capital and high-skilled workers? How might automation exacerbate or reshape existing inequalities?
- Policy Responses: What are the arguments for and against policy interventions — retraining programmes, universal basic income, shorter working weeks, stronger labour protections, taxes on automation?
- Workplace Surveillance and Autonomy: How is AI being used to monitor workers — productivity tracking, performance evaluation, algorithmic management? What are the implications for autonomy and dignity at work?
Opening case: “A logistics company rolls out an AI system that assigns tasks, sets the pace, and rates each worker’s ‘productivity’ in real time. Output rises 15% and two supervisor roles are cut. Workers say the pace is relentless and the ratings are opaque; management calls it efficiency. Is anything actually wrong here — and if so, is the problem the job losses, the surveillance, or who captured the 15%?”