Generative AI contested statements line-up

Recommended for: All levels; works particularly well as the first AI exercise in a course

Background for the instructor

Students arrive in a course with widely different prior beliefs about AI: some regard it as reliably accurate, others as fundamentally untrustworthy, and many have not examined either view critically. Making these prior beliefs visible and discussable before any formal instruction is a well-supported pedagogical step. Research on conceptual change shows that learning is more effective when prior conceptions are first made explicit and then challenged with evidence (Chi, 2008, in Sigel & Renninger, Handbook of Child Psychology, Vol. 2, Wiley).

The physical movement element is not merely energising. Requiring students to commit to a side of the room before the discussion prevents them from staying in a comfortable middle position, which tends to reduce the quality of subsequent discussion. It also makes the distribution of views within the group immediately legible, which itself becomes data for discussion.

The statements below are taken from the original presentation and can be used as supplied. Instructors are encouraged to add one or two statements drawn from their own discipline.

Suggested statements

• Generative AI uses up-to-date data
  • LLMs have content with fixed end dates.
  • Chat applications can enhance the information flow with up-to-date information by performing web searches.
• Using generative AI has a negative impact on the environment.
  • This is true in principle. But it is also true for a lot of other activities and processes.
  • We refer to the AI-Literacy Companion for students for more ideas. For example: What costs more water: eating a hamburger or a 1000 generative AI chats
• Generative AI will rival human creativity in the future.
  • This is really open for discussion.
• Generative AI will affect my field of work for good.
  • Most likely for all professions. The question is how and to what extent.

Procedure

1. Designate two sides of the room as “agree” and “disagree.” If the room layout does not permit this, use a show of hands or a polling tool, though physical movement is preferable.
2. Read one statement aloud. Students move to their chosen side without discussion. Sitting in the middle is not permitted.
3. Ask two or three students from each side to state their reasoning in one or two sentences. Record key arguments briefly on a whiteboard or slide.
4. After collecting arguments from both sides, present one piece of concrete evidence that bears on the statement (for example, for the statement about up-to-date data: a brief explanation of training data cutoff dates and retrieval-augmented generation). Note whether the evidence supports or contradicts the majority view.
5. Repeat with a second statement. You do not need to use all statements in one session.
6. Close with the discussion questions below.

Discussion questions

• Did any of the arguments from the other side surprise you? What would it take to change your position?
• For each statement: what kind of evidence would settle the question? Is that evidence easy to find?
• How did your initial position compare to the evidence presented? What does the gap between initial belief and evidence suggest about how you typically encounter claims about AI?

Assessment opportunity

This exercise works best as a formative activity without a grade attached. If a written component is desired, students can write a brief exit ticket (three to five sentences) in which they identify one statement where their position shifted or was confirmed by evidence, and explain what kind of evidence was most persuasive.

Critical reflection on this exercise

The main risk of this exercise is that it becomes a simple opinion poll rather than a critical thinking activity. This happens if the instructor does not insist on evidence during the discussion phase. The statements must also be genuinely contestable: statements that are obviously true or false will produce low-quality discussion. Before the session, instructors should verify the factual status of each statement so they can correct misinformation without reinforcing it. The statement about up-to-date data, for example, requires care: the correct answer is nuanced (training cutoffs exist, but some tools also use live retrieval), and oversimplifying it in either direction would itself be a form of misinformation.