Prof. Dr. Thomas Steiner
The Art of Intelligent AI Use
From talking to AI to thinking with AI
Course premise
You begin this course with an intentionally simple instruction: "Write a paper on the economic value of beauty in tourism." The topic stays stable while your relationship with AI changes. You first learn to communicate with a language model, then to use it as a cognitive partner to extend, structure, and examine your thinking.
This course is not built around the promise of a perfect prompt. It introduces prompt design as one element within a larger practice of intellectual design: framing, dialogue, decomposition, evidence, critique, judgement, responsibility, and reflection. The aim is not to make the tool appear intelligent. The aim is to make your use of the tool more intelligent.
Baseline prompt
Write a paper on the economic value of beauty in tourism.
Do not only ask AI for an answer. Design the conditions, operations, and judgements through which a worthwhile answer can emerge.
Learning outcomes
- You design contextual prompts that distinguish situation, perspective, purpose, operation, and deliverable.
- You use conversational structures, personas, meta-prompts, feedback, and memory deliberately rather than accidentally.
- You use AI to question, decompose, map, traverse, test, and reflect on complex problems.
- You externalise methods in portable agents as Word documents and orchestrate several tools or agent specifications.
- You evaluate logical validity, evidence, uncertainty, framing, manipulation, sustainability, and delegation.
Course architecture
Technical distinctions
| Context window | The conversation material currently available to the model. It is finite and system-dependent. |
|---|---|
| In-context adaptation | Changes in response caused by instructions and feedback in the available conversation, not necessarily by model retraining. |
| Cross-conversation memory | Optional saved or retrieved personal context made available in a new conversation; availability and control vary. |
| Structured reasoning output | An inspectable decomposition requested from the model. It is not complete access to hidden internal computation. |
| Agent as a Word document | A portable instruction artefact defining objectives, boundaries, process, information, resources, and completion criteria. |
| Tool orchestration | Dividing cognitive work among a general model, specialised tools, evidence sources, and human judgement. |
The art of intelligent AI use begins for you with better prompts, but cannot stop there. It grows through better conversations, more inspectable thinking, stronger evidence practices, and explicit human judgement.
Use AI not merely to produce more, but to question better, think more clearly, and decide more responsibly.