VVZ API is not affiliated with ETH Zurich. Data might be outdated or incorrect. Please view the official ETHZ Vorlesungsverzeichnis for binding information.
Abstract
This PhD block course lets you experiment with AI tools for generating texts, images, audio, and code for scientific use, from presentations to publications. Engage in problem-based learning while experts guide hands-on workshops on AI use with research data, scientific illustrations, and model customization in GPTs and bots.
Objective
• Hands-on training on using generative AI in a responsible, efficient and professional way in daily scientific work. • Competencies to deal with limitations, potential pitfalls and benefits of AI-based tools in relation to research integrity and ethics. • Reflect on manipulation techniques of present and past scientific visualizations techniques. • Get introduced into synthetic audiovisual data such as deepfakes. • Use visualization techniques in co-setup with AI-tools to create visual abstracts, thumbnails or graphical abstracts for peer-reviewed publication. • Practice collaborative workflows with genAI. • Get familiar with technologies that should help to protect personal or copyrighted data when using AI and apply them to own data.
Content
Ethical Prompting Lab Outputs generated with AI often amplify social biases, stereotypes or inaccuracies through their filtering of data: for example, favoring certain groups or ideas, perpetuating stereotypes, or making incorrect assumptions based on learned patterns. Through ethical prompting these shortcuts can be partly overcome together with system-based measures. The day will be structured by short theoretical input on prompt engineering with a special focus on following the guidance of the PROMPT AI Framework for Informed AI Use : Precise, Responsible, Personalized, Culturally Sensitive, Explainable, Conversational, Troubleshooting. We will discuss the different outputs of genAI. LLMs Customization / AI-supported scientific workflows Participants will explore workflows that can simplify or automate certain steps in scientific work. Especially we will look into individual chat bots and customized GPTs. We will understand how RAG workflows (PDF to text, Image to vector) are working as they are the underlying principle of individual and controlled chat bots. We are aiming also to test GPTs for their correctness or biases. Responsible image creation / scientific visualization Participants are introduced into the potential and fallacies of image-based AI tools respecting originality, transparency, copyright. The applications of synthetic audiovisual media such as “deepfakes” in the realm of academic work will also be critically reflected. How to create reproducible and accurate scientific visualizations from scientific data using e.g. stable diffusion models? What are the potentials and limitations of using diffusion models for scientific visualizations? How to prompt for original ideas and how to change the images through human expertise and own style by using creative but controlled prompting? Students will practice sketch-to-image, collage-to-image, inpainting methods to explore potential use of these tools for their own projects. Protecting your privacy Awareness what you give away when you share (your) data in AI-based environments. What is the private, sensitive or sensible information that in different AI-based tools and instances users are leaking? We exchange on techniques and processes to avoid leakage of private or non-copyrighted but original and personally owned material, from privacy-protecting measures in institutional environments to techniques that protect images and information to be used by gen AI to new forms of depersonalizing or protecting of data with watermarks to make the origin and the copyrights transparent. In this session, we will also create awareness and sensitivity for the avoidance and protection from deep fakes from personalized data.
General Information
- Language
- English
- Levels
- DR
- Frequency
- Semesterly recurring
Examination
- Type
- ungraded semester performance
Course Components
| Type | Title | Time & Place | Hours |
|---|---|---|---|
| lecture with exercise |
PhD-Student Experimenting Lab: Explore the responsible use of AI in generating scientific content
Does not take place this semester.
|
No time listed | 32 h semesterly |
Offered In
-
Doctorate Environmental Systems Sciences (More Information at: )
-