Generative AI is a branch of artificial intelligence that focuses on generating new content – ​​for example, in the form of text, images, audio, or video. Unlike traditional models that classify or predict, generative models create entirely new data based on what has been learned previously.

Current generative models, such as GPT-4, Claude 3, and Gemini 1.5, are capable of producing natural language, images, program code, audio, and even video. These models are increasingly used in teaching, research, and content creation.

Example

We will take as an example text generation using the GPT-4 model, one of the most well-known generative models.

Situation: You have writer’s block, but you want to write the beginning of a short story with a hero who finds a magical item.

Prompt: Write a short story beginning about a hero who finds a magical object.

Text created by generative artificial intelligence:

“Far away at the foot of the northern mountains, in a quiet village called Kaarnala, lived a young blacksmith’s apprentice named Eljas. He lived a modest life until one evening, as a storm raged in the sky, he found an old, moss-covered chest at the foot of a fallen spruce behind his workshop. Inside the chest lay a strange object – a small, glowing compass whose needle never pointed north. When Eljas touched it, the world around him seemed to stop – and a whisper echoed in his ears: “You are the chosen one.”“

How does generative artificial intelligence work?

1. Training phase
   The model is trained on a huge amount of data (e.g. billions of text snippets or images). This phase develops the model’s ability to recognize structures in language and other content.
2. Generation (inference)
   When the user gives the model input (e.g. the beginning of a sentence), the model predicts its continuation based on learned patterns and probabilities.
3. Fine-tuning (or prompting)
   The user can refine the output by changing the input or using additional instructions (e.g. “write in formal style”).

Multimodality

Multimodality refers to the ability of an AI to process multiple data formats – such as text, images, audio, and video – simultaneously. For example, ChatGPT or Gemini can understand images and respond to them using text.

Multimodal systems enable:

• Guiding the user based on visual or voice messages
• Combined image and text analysis (e.g. image interpretation)
• Identifying and producing video or audio content

Ethical and responsible use

The use of generative artificial intelligence requires awareness and responsibility, especially in teaching and research.

The most important things to consider:

• Transparency: The use of artificial intelligence must be clearly disclosed (e.g. in student work).
• Accountability & Human oversight and source criticism: Content produced by artificial intelligence may be incorrect or biased – the user is responsible for the end result.
• Data Protection: Do not enter personal data or confidential material into open AI services.
• Fairness and equality: Ethics and fairness must be taken into account when using generative artificial intelligence.

Legislation: EU AI Act

The EU Artificial Intelligence Regulation (AI Act), which came into force in 2024, specifically obliges public administrations (such as higher education institutions):

• To mark when content is generated by artificial intelligence.
• Assess risks before deployment.
• Document usage and training.

More information: EU AI Act – summary

Recommendations for teaching

• Clear instructions in the course: what artificial intelligence can be used for, how its use should be documented
• The use of artificial intelligence must be described, e.g. “Artificial intelligence was used for text structuring and language checking (GPT-4, 5.5.2025)”

Links and additional information

• EU Artificial Intelligence Regulation (EU AI Act)
• Arene’s recommendations on the use of artificial intelligence for universities of applied sciences
• OpenAI – GPT-4 demonstration
• Google Gemini
• Best Practices for Responsible Use of AI in Education

 ChatGPT and Copilot were used to help design and compile the instructions.