Science
A machine can imitate a scientific claim. It cannot know if it is true. Can you?
Jump straight to the exercise →
Core
A scientific fact can sound settled even when it rests on weak, outdated or retracted research. Here you learn to trace a claim back through its research history instead of trusting surface consensus.
Before you start
Before you start: the human anchor is about your thinking, not spelling, presentation or a perfect result. You may use dictation or speech-to-text if that suits you better, and it is fine to make mistakes. Choose an example you feel comfortable sharing. In the algorithmic pivot, you may use any AI tool; a free account is enough and you do not need a subscription.Human Anchor
Write the first part of the project without AI. The goal is to create a human benchmark that can later be compared with algorithmic analysis. The text should be personal, situated, and precise enough so that AI cannot simply replace it with a generic summary.
Choose a scientific claim, a natural phenomenon, an animal species, or an astronomical topic that you are interested in. Write down what you think is known, where that knowledge comes from, and what evidence would change your mind.
Scope
- 1,000 words minimum, unless the teacher decides on a shorter test version.
- No AI, no automated rewriting, and no machine translation.
- Use your own examples, your own words, and clearly state uncertainty where it exists.
Include This
- What you know or believe before the AI is introduced to the work.
- What experience, memory, physical observation, or local knowledge shapes your perspective.
- Which three questions you want the AI to challenge in the next step.
Assessment
The teacher does not assess whether the text is perfect. They assess whether the human voice, the student's own judgment, and real context are clear enough to hold up against algorithmic output later in the project.
Algorithmic Pivot
Have an AI trace the origin of one scientific claim, name key studies, assess whether the results have been replicated, and point out retracted or controversial papers that might have contaminated the model's knowledge base.
Fyrirspurn til gervigreindar
This is a ready-made frame to use in Step 2. It disciplines the AI by giving it a role, a task, constraints and verification questions, instead of a free-form chat. Copy it, try it, and adapt it to your own project.
- Role
- You are a critical assistant in the Tækniseigla category Science. You should expand the student's analysis without taking the responsibility away from them.
- Task
- Trace a scientific claim back to research and assess its strength.
- Constraints
- Distinguish between primary research, review articles, and popular coverage.
- Look specifically for retractions, corrections, or failed replications.
- Mark the confidence level of claims instead of presenting everything as fact.
- Verification questions
- 3 verification questions are included in the copied prompt. You answer them in step 3.
Fiduciary Anchor
The accountability document is the final gate of the project. Here you show that you can use AI without hiding behind it. All claims that survive in the final product require verification, a clear disclaimer, or conscious rejection.
Show how an LLM can produce a scientific consensus from outdated or weak data. In the accountability document, compare the AI output with actual papers, databases, or review studies.
1. AI Process
Record which tool was used, which prompts were tested, what changed between versions, and which parts of the output first seemed convincing.
2. Verification
Use at least three independent pillars depending on the nature of the project: a primary source, official data, an academic source, your own calculation, physical testing, expert evaluation, or comparison with human experience.
3. Errors and Blind Spots
Answer these questions:
- Which source is the actual origin of the claim?
- Has the result been replicated?
- How does the answer change if a retracted or weak paper is removed?
4. Rejection
Name at least two points that you reject from the AI output. Explain whether they were incorrect, unproven, ethically weak, too general, or simply not in line with your human anchor.
5. Accountability Statement
Conclude with a short statement in the first person: what you take responsibility for, what you do not know yet, and what decision you would never delegate to AI.
Now it is your turn. The exercise is the bridge between step 2 and step 3.
Go to exerciseFor teachers, competency line: Substrate judgment
That is the whole module: anchor, pivot, accountability. The work itself happens off this page.