Learning Goals
This exercise is designed to help you make your existing problem-solving practices and your use of AI tools in this context more deliberate, systematic, and reflective.
It will support you in:
- sharpening structured problem solving with and without AI
- using AI as cognitive support by applying the Six Support Approaches of AI appropriately to different tasks
- strengthening prompting and metaprompting strategies to actively steer the quality, direction, and depth of AI contributions
- reflecting on how AI influences your reasoning, and making explicit decisions about where human expertise, judgment, and responsibility must remain central
Ongoing Reflection After Each Step
To support these learning goals, briefly discuss the following after each step:
- What did we learn that transfers to other tasks and to critical AI use?
- How can we apply this in our day-to-day research practice?
- Where is caution required (e.g., data security, good scientific practice, ethics, responsible authorship)?
1. Understanding the Method “Four Steps to a Solution”
1.1 Read the Method (5 Minutes)
Read: “Four Steps to a Solution”
1.2 Walk Through One Example Without AI (15 Minutes)
As a group, select one real problem from the research or professional work of one group member and apply the method to this shared example — without using AI.
Choose a problem that:
- is genuinely complex (the solution is not obvious), and
- is small enough to work on meaningfully within 15 minutes.
The goal is to understand and try the method—not produce the perfect solution.
Work briskly. Move on once the point of each step is clear.
Stop after 15 minutes even if you are not finished.
2. Develop a Guide: “Four Steps to a Solution With AI”
2.1 Human-in-the-Loop Workflow (1–2 Hours)
Design a workflow showing how AI can support the method. The goal is to:
- increase the quality of problem solving and/or
- reach a useful outcome faster.
Work through the four steps sequentially and decide, for each step:
- Which of the Six Support Approaches of AI is most helpful here—and why?
- Which prompts do we need to get reliable, research-appropriate outputs?
You are encouraged to use AI while developing the guide, for example:
- Metaprompting – for example, by using the Prompt Generation Prompt.
- Brainstorming – generating ideas and alternative perspectives.
- Reflection – analyzing your own thinking and AI’s responses.
- Structuring – organizing your guide and refining your argumentation.
Working mode: Iterate on your prompts and formulations until the outputs for each step meet your own standards for quality.
When helpful, briefly agree as a group on what “sufficient quality” should mean in your specific context.
Important: Save interim results outside the chat in a non-AI tool (e.g., a document, Notion, Etherpad) to prevent loss and to support traceability.
2.2 Quality Check and Iterative Improvement (30–45 Minutes)
Re-run your original example (the one you used in 1.2) —this time using your AI-supported workflow.
Then reflect:
- What is better than the no-AI run (e.g., clarity, completeness, quality of options)?
- What is easier or faster?
- Where did issues or risks appear (e.g., questionable assumptions, overly confident statements, invented details, misfit suggestions)?
- What needs to be changed in the guide to remove/reduce these issues? (Iterate and improve.)
2.3 Deliverables
Create a guide that is as concise as possible while still covering all essential elements:
- explains your workflow clearly to an outsider (e.g., a colleague)
- walks the reader through the process step by step
- provides ready-to-use prompt examples for all key steps in the workflow
3. Optional: Full Automation / Zero-Shot (20–30 Minutes)
In this exercise, you try a contrasting approach: instead of steering the process step by step (human in the loop),
you try to accomplish as much as possible with one carefully designed prompt (full automation).
3.1 Task
Create a zero-shot prompt (e.g., for an agentic workflow) that produces a comparable guide from a single input—i.e., one prompt that generates the full result in one run.
Consider:
- What context, constraints, and required structure must the prompt include?
- Use AI to explore what such a prompt should contain and how it should be organized.
- Use metaprompting to improve the prompt and record it in clean, readable markup.
3.2 Comparison and Reflection
- Compare the quality of the automatically generated guide with your human-in-the-loop guide.
- Reflect on the advantages, limitations, and associated risks (e.g., quality, time, control, transparency).
