MODULE 4 — CLOSING

Applications, limits and critical use

Understanding AI is only half of it; the other half is knowing when to trust it. This final module turns technical understanding into responsible practice: evaluating answers, sources and risks, caring for privacy and bias, and closing with a simple method for critical use.

Module principle

Fluency is not truth. Speed is not validation. Good AI needs a good human in the loop.

Risky claims detector

Paste claims, one per line. It does not verify truth — it flags risk signals to train critical reading.

Language models always tell the truth.

high risk

Absolute or anthropomorphic claim. Requires strong verification.

Attention is the same as consciousness.

high risk

Absolute or anthropomorphic claim. Requires strong verification.

RAG can help an AI use external documents.

moderate

Plausible claim, but depends on context and evidence.

GPUs are useful because they accelerate parallel operations.

check

Separate fact, interpretation and source before using.

A fluent answer can still contain a factual error.

moderate

Plausible claim, but depends on context and evidence.

Source evaluator

Adjust the criteria and see how authority, date, traceability and bias risk change the confidence score.

Peer-reviewed article published this year

Academic article

95% confidence

Recency: 5Authority: 5Traceability: 5Bias risk: 2

Viral post with no identified author

Social media

35% confidence

Recency: 4Authority: 1Traceability: 1Bias risk: 5

Official document from a public institution

Official source

90% confidence

Recency: 4Authority: 5Traceability: 5Bias risk: 2

Old technical blog, but with good references

Technical blog

60% confidence

Recency: 2Authority: 3Traceability: 4Bias risk: 3

Responsible use planner

Choose a use case and adjust sensitivity, impact and accuracy need. This helps decide the level of care.

Use caseData sensitivity: 2Decision impact: 2Factual accuracy need: 4

Estimated risk

53%

Moderate care

AI can help a lot, but keep fact-checking, note limits and human review.

Critical use method checklist

A simple list before taking an AI answer into research, report, class or decision.

Checklist complete38%

Lesson complete35 minUso crítico

4.1

Using AI as a critical researcher

From curiosity to method: ask, verify, compare and decide

▸ Lesson objective

Learn to use AI as an investigation tool without turning textual fluency into automatic authority.

AI as a thinking partner, not an oracle

After understanding history, language machines, embeddings, attention and RAG, we reach the most important point for a non-specialist researcher: how to use well without being too dazzled.

AI can help formulate hypotheses, explain concepts, organise reading, compare ideas, review texts and suggest directions. But it should not be treated as the final source of truth.

The critical use cycle

Good use follows a cycle: ask clearly, receive an answer, identify verifiable claims, check sources, compare with other references and only then decide what to keep.

The common mistake is to stop at the second step, because the answer looks beautiful. A beautiful answer is an excellent start. But in research, beauty without verification is just a well-dressed lecture.

Use AI to amplify, not to outsource

AI is strong at expanding possibilities: suggesting angles, summarising arguments, listing risks, proposing structures and explaining topics at different levels.

The researcher remains responsible for curation, critique, framing, validation and intellectual authorship. AI can help think; it should not replace thinking.

[ practice ]

Close the cycle

Think of a real question you would ask AI today. Before accepting the answer, list: which of its claims would you need to verify, and where would you check each one?

See expected answer

A good exercise separates what is verifiable fact (dates, numbers, citations) from what is interpretation, and names a source for each fact. If you cannot say where to check, the answer is still a hypothesis — not a conclusion.

✓ What you take away

Using AI critically means treating it as an investigation instrument: useful, fast and fallible. The value lies in the cycle of questioning, verification and human decision.

Living glossary

Terms we will encounter in this module.

Hallucination

A plausible but incorrect, invented or insufficiently supported answer.

Traceability

The ability to verify where a statement or conclusion came from.

Primary source

Material closest to the original fact, such as an official document, database or original article.

Secondary source

Material that interprets, summarises or comments on primary sources.

Bias

A systematic tendency that can distort responses, recommendations or interpretations.

Sensitive data

Personal, confidential or delicate information that requires special care.

Critical use

Use that combines exploration, checking, comparison and conscious decision-making.

Human validation

Review and decision made by a responsible person, especially in high-risk contexts.

Applications, limits and critical use

Module principle

Fluency is not truth. Speed is not validation. Good AI needs a good human in the loop.