New GSI department guideline for the use of AI in scientific and student work adopted

AI Charter

What AI is and what it can do. And what it cannot do.

AI-supported tools are becoming increasingly prevalent in research and teaching, offering assistance with language editing, literature discovery, translation, and error correction in computer code. However, it is important to use these tools consciously, as they may perpetuate biases against minorities and reinforce existing stereotypes. Additionally, AI-supported tools often operate in the cloud and rely on user input for model training, raising concerns about data privacy. Therefore, it is essential to understand the advantages and disadvantages of using these tools in order to maintain control over one's own work process.

Understanding the internal functioning is particularly crucial for large language models (LLMs), which have gained widespread popularity lately. LLMs are developed using substantial amounts of text data, allowing the model to identify patterns in text. Knowing these patterns, the model can produce human-like answers to user queries as these answers mirror the structure of the existing text data on which the model was trained. By recreating patterns in training data, these language models resemble ‘stochastic parrots’ (Bender 2021). Because of their design, these AI models are well suited if the main purpose is to generate text that should recreate certain ways of writing (e.g., a poem on AI in the style of Schiller). Yet, LLMs also resemble ‘stochastic parrots’ in that they do not understand the meaning of the text they produce. LLMs have no sense of logic or truth.

This combination of generating text that appears credible but is not grounded in a comprehension of the world necessitates caution when employing these tools to create scientific text. You should not use LLMs to generate text that you lack the expertise to verify yourself.

AI in Student Work

The principles of independence and transparency apply to the use of all types of tools in student work. These principles also apply to the use of AI-supported tools.

Assignments must be written independently so that the person writing the assignment bears full responsibility for the content of the work. In addition, all tools used in the creation of the paper must be listed.

When using AI-supported tools, the name and function of the tool must be stated. When using AI-supported language models with a chat function (e.g. Chat-GPT, Bard, Llama), the complete chat history must be archived and presented upon request.

Template: Declaration of Originality

"I hereby declare that I have written this seminar paper on my own, have not used any sources or aids other than those specified, and have marked the passages taken from them as such. All aids used are listed in the table below. Regarding the use of Large Language Models (LLM, e.g. Chat-GPT), I have archived the entire LLM interaction history with reference to the submitted work and can present it upon request. This seminar paper has not been presented in this or a similar form in any other course."