Applications of artificial intelligence, critical thinking and human collaboration in research, learning and work. An open experimentation space where members take the initiative, correct their mistakes and choose their partners; the lab does not run their project for them.
Goal: Learn to build a small project that includes a real problem, evidence sources, output checking and ethical reflection. Members choose their own context and record their sources and the changes they made; no rank or points are given merely for taking part.
How to work in this lab
Free-initiative method: start in a way that suits your circumstances, record your assumptions, mistakes and corrections, then decide for yourself whether to involve a knowledge circle or invite participants.
After each activity, write down what changed in your choices after review, and set an optional next step.
Attach the date of the experiment, its sources and limits, and the versions of your work, showing mistakes and corrections and distinguishing individual work from partners’ contributions.
FUTURIA provides no binding individual service and awards no automatic grade for completing an activity.
Identify a problem in learning or services and phrase a specific question that names the beneficiaries, the time frame and an outcome that can be checked. Avoid turning assumptions into facts, and define what you can measure without collecting personal data.
Applied extension — testing hypotheses: choose three independent claims and write a verifiable answer for each. Set two explicit comparison criteria (accuracy of information and limits of confidence) and compare the results with dated primary sources. Do not enter people's names or private data, and record any differences between outputs.
Applied activity
Create a problem card: the context, the question, three alternative hypotheses and two measurable indicators.
Expected evidence
A dated problem card with traceable public sources.
Self-check questions
What is the difference between a general question and a testable one?
Answer: A testable question names the beneficiaries, the time frame and an outcome that can be checked; a general question offers no way to judge its result.
Why write three alternative hypotheses instead of one?
Answer: Alternatives stop us from treating our first assumption as a fact and push us to look for evidence that distinguishes between them.
How can you measure the outcome without collecting personal data?
Answer: Use general, aggregated or published indicators and avoid names or anything that identifies individuals.
2. Checking results and bias
Rephrase the same request in three ways and compare the answers. Check the cited references in their original sources, and distinguish between facts, inferences and undocumented claims. Discuss how language and context affect the result.
Applied extension — a risk and bias matrix: test the answer across different contexts and languages, and identify who could be harmed if a user relied on it without scrutiny. Separate probability from evidence and keep a human as the decision-maker; record corrective actions and the limits that remain unknown.
Applied activity
Build a table that compares an evidence-backed answer, a contradictory answer and a claim that still needs verification.
Expected evidence
A table of evidence, caveats and corrective actions.
Self-check questions
What is the benefit of rephrasing the same request three ways?
Answer: It shows whether the answer stays stable or changes with the wording, which indicates how reliable it is.
What distinguishes a fact, an inference and an undocumented claim?
Answer: A fact is backed by a verifiable source, an inference is an interpretation built on evidence, and an undocumented claim has no source.
Who makes the final decision when an AI tool is used?
Answer: The human: the tool suggests, while the user verifies, decides and takes responsibility.
3. The small prototype and the human in the loop
Design a simple procedure that helps a learner choose study sources or organise a learning path, leaving the final approval to a person and documenting the limits of the model.
Applied extension — human-centred design: build a small prototype that helps users choose without deciding for them. Compare an option without AI to an AI-supported option, and test whether users can contest the output, understand it and opt out of the service.
Applied activity
Design a three-step interaction model and test it with two hypothetical scenarios, without real data.
Expected evidence
A workflow diagram, test results and a note on the limits of use.
Self-check questions
What does "human in the loop" mean?
Answer: A person keeps a role in reviewing the system's outputs and approving or rejecting them before they have any effect.
Why compare an AI-supported solution with an option without AI?
Answer: To find out whether AI adds real value or whether a simpler alternative is enough.
Name three properties to test in the prototype.
Answer: Whether outputs can be contested, whether they can be explained, and whether users can opt out of the service.
4. Presenting results in the forum
Write a presentation that covers the question, the method, the evidence and what did not work. Ask for critique from another discipline and include the changes in the next version.
Applied extension — auditing results after a change: repeat the experiment after changing only one element and show its effect on the result. Document the model version used, the test date, any failure and the conditions needed to reproduce the work. Do not claim accuracy or impact you have not measured.
Applied activity
Publish a short report with an open question for multilingual discussion.
Expected evidence
A documented version of the report and a change log.
Self-check questions
What are the essential elements of a results presentation?
Answer: The question, the method, the evidence, what did not work and the later changes.
Why change only one element when repeating an experiment?
Answer: So that any change in the result can be attributed to that element alone.
What information is needed to reproduce the experiment?
Answer: The model or tool version, the test date and the conditions and settings used.
An educational and research lab: it does not award academic qualifications or professional licences, and it does not replace specialist supervision for hazardous or regulated work. Members make their own decisions and abide by the law and the rights of others; an exercise must not be turned into a high-risk professional procedure without specialist supervision.