Stochastic Parrots and Beyond
Normative Dimensions of Generative AI
Wednesday, 4 November 2026, 09:00–17:00. Universitätsbibliothek Basel, GRACE room 306, 3rd floor.
About the course
Generative AI is used increasingly in research, yet its environmental costs, training data, labor conditions, discriminatory representations, commercial incentives, and governance structures are often considered separately. This workshop combines four critical lenses with practical risk-analysis tools. Participants investigate the design, data, training, deployment, evaluation, and governance of text- and image-generating systems; then use Panarchy and Ecocycle Planning to identify feasible changes in their own research environments. The aim is a defensible, transparent approach to responsible generative AI use rather than a universal list of permitted and forbidden tools.
Editions
One set of teaching materials serves every edition. The practical facts differ for each edition.
- Autumn semester 2026 — Confirmed. Wednesday, 4 November 2026, 09:00–17:00.
- Spring semester 2027 — Planned. Date: To be confirmed.
This site is a course companion. It takes no registrations and awards no credits. The Graduate Center (GRACE) of the University of Basel does both. For registration, credits, and the binding course rules, go to the official course listing of the University of Basel. Each edition page links it.
Start here
If you take part in a course edition
- Open the page of your edition above. It gives the date, the venue, and the rules.
- Read the required text before the course day: Bender et al., On the Dangers of Stochastic Parrots (2021).
- Write a short description of one actual or planned use of generative AI in your research. The assignments page explains it.
- Bring a laptop.
Do not bring confidential or personal research data to the exercises. No exercise needs a commercial AI account.
If you teach or study on your own
- The syllabus shows the learning path and the timetable.
- The four lens pages each hold concepts, evidence, a documented case, and an activity.
- The facilitator guide gives the Panarchy and Ecocycle session minute by minute.
- The handouts are printable and editable.
- The reuse guide explains how to adapt the site for your own course.
Four lenses
| Lens | Guiding question | Output |
|---|---|---|
| 1. Stochastic Parrots: Scale and Environment | Which environmental and data choices are hidden by a model’s apparent performance? | Material-and-provenance map |
| 2. Representation and Pygmalion | Whose perspectives become normal and whose are displaced? | Representation and harm pathway |
| 3. Lifecycle and Labor | Who does the work, carries the risk, or has no recourse? | Lifecycle risk matrix |
| 4. Political Economy | Which incentives preserve an unsafe or inequitable practice? | Stakeholder and incentive map |
After the lenses, the group uses Panarchy to see how changes at different levels interact. A short adaptation of Ecocycle Planning then sorts the most promising changes.
Learning outcomes
After the course, participants should be able to:
- Analyze environmental, social, cultural, and institutional issues in generative AI through four distinct analytical lenses.
- Identify normative risks at design → data → training → deployment → evaluation → governance stages, including affected people and potential mitigation.
- Apply risk maps, documentation and disclosure heuristics, evaluation questions, and governance checks to their own research.
- Formulate and defend a context-specific position on responsible research use of generative AI, informed by scholarship and governance frameworks (including the EU AI Act as one example).
License
Original course content is available under CC BY 4.0. Material adapted from Liberating Structures is available under CC BY-SA 4.0. Code is available under AGPL 3.0. Cite these materials as described in CITATION.cff.