GenKI@UHB

The University of Bremen provides access to various generative AI (genAI) models that comply with data protection regulations via the Academic Cloud of the GWDG (Gesellschaft für wissenschaftliche Datenverarbeitung /Society for Academic Data-Processing). GWDG is a joint enterprise of the University of G?ttingen and the Max Planck Society (www.gwdg.de). These genAI services are referred to as ChatAI and their use is voluntary. Access is granted via a web interface using the University of Bremen account login data (federated login). Several different open-source models can be used, some of which are hosted internally. Externally hosted commercial models such as ChatGPT from OpenAI can also be used. The advantages are as follows:

• GWDG processes all data in compliance with the GDPR (General Data Protection Regulation).

?     GWDG is contractually obligated to protect user data. Login data is only used by GWDG to verify user authentication and authorization; this is contractually secured.

?     Even with externally hosted tools, GWDG is not required to disclose personal user data to external service providers (note: The entire input content (prompts) must, however, be passed on in an unfiltered manner).

?     Data entered will not be used to train AI models.

?     Input is not stored on GWDG servers, but exclusively locally in the browser used. With the external model (ChatGPT), however, Microsoft reserves the right to store the data for up to 30 days in order to prevent misuse.

academiccloud.de/services/chatai/

The University of Bremen provides secure AI-based systems using the GWDG platform, but this does not release you from your personal responsibility for the content you enter, your need to critically review the output, or your resulting use thereof.

The use of external generative AI-based systems (with the exception of GWDG’s ChatAI) is not recommended; research projects may be an exception to this rule. The City of Bremen is preparing to launch an administration-specific AI-based system (LLMoin), which will also be made available to the University of Bremen’s administrative staff.

This guide will be updated on an ongoing basis to reflect changing technical, legal, and ethical requirements.

• Reflective Practice

Always carefully consider the use of generative AI. What are your goals and what results do you expect? Do you know how the selected model works as well as its strengths and limitations?  Do you know what you are entering and how you are allowed to use the results?

Always check and verify the output against primary sources, subject-matter expertise, or peer review.

• Critical Questioning

Do not pass on unverified AI outputs – take full responsibility for what you share and carefully evaluate the output. Is the information plausible, correct, and ethical? Are there any possible biases, discrimination, or factual errors?

• Data Protection Compliance

Do not enter any personal data into the genAI systems (without legal authorization or consent). Check to see if results include personal data that is not allowed to be shared.

• Information Security

Do not enter confidential information into AI systems. Note the classification and sensitivity of your information.

• Copyright Compliance

Be careful not to enter or publish any copyrighted content. Verify all outputs for potential copyright violations and determine who has the rights to AI-generated content. When in doubt, do not publish AI-generated content.

• Ethical Considerations

Critically assess outputs for biases, factual inaccuracies, or misrepresentations; label content generated with AI and, if necessary, document your AI usage process comprehensibly.

• Prohibited Practices

Generative AI must not be used to create profiles, for automated grading, evaluations, generating plagiarized content, or providing misleading / non-transparent information.

• Always check against the Checklist for Legally Compliant Use of Generative AI Services #LINK to the checklist

Checklist genKI

• Central point of contact for inquiries and for the GenKI@UHB project,

www.uni-bremen.de/en/digital-transformation/projects/genkiuhb

Email: genkiprotect me ?!uni-bremenprotect me ?!.de

• genAI in Teaching and Learning

Administrative Unit 13

Link to the recommendations for use for teaching and learning

ZMML

Link to ZMML information on the use of artificial intelligence

• Data protection and information security

Link to the DSB and ISB information and service portal (in German only)

Generative (GenKI) Artificial Intelligence (AI) 

AI applications that are trained with data to automatically generate content such as texts, images, sounds, videos, or program code when prompted.

Large Language Models (LLMs)

Text-generating systems (e.g., ChatGPT, Llama, Gemini, Claude) that are trained with large amounts of data. They have no understanding of the content, but calculate the statistically most likely next character or word sequence. Erroneous, fabricated, or factually incorrect outputs ("hallucinations") are possible.

Hallucinations

AI provides convincingly formulated output that is factually incorrect. The content and argumentation seem coherent, but the AI system completely fabricated the information.

Bias / Distortions

Bias or misrepresentation embedded in training data or model design can carry over to AI outputs and lead to discrimination (e.g., unequal treatment of groups of people).