Source criticism and responsibility in AI use
Understand why AI can generate incorrect information with high confidence, verify claims against external sources, and identify appropriate use cases for AI support.
Prerequisites
Everyday explanation
A generative AI is trained on large amounts of text to predict the next likely word. It does not understand facts — it generates text that resembles patterns it has seen before. Therefore, it can formulate statements that sound authoritative but lack factual basis.
This phenomenon is sometimes called hallucination. It does not mean the AI is lying intentionally, but rather that it lacks internal knowledge of whether the answer is correct.
Intuition
Three principles for safe use:
- Verify facts. Dates, names, and figures should always be checked against an independent source, not the AI.
- Be sceptical of details. Specific source references and quotes may be fabricated. The more precise the answer seems on an unusual topic, the more important verification becomes.
- Be transparent. In work and study environments, it must be clear when AI has been used. Submitting AI-generated text as your own work is a breach of integrity, similar to plagiarism.
AI is a tool for analysis, structuring, and idea generation, not a primary source for facts.
Mastery means
- States two reasons why an AI response might be incorrect
- Describes a method for verifying a specific claim
Sign in to do the exercises and build your mastery up.
Sources
Leads to
Part of the goals (21)
- AI for beginners
- Builder — collect data, train a model and test AI critically
- AI, ethics and society
- Language models in practice
- Data: collect, clean, document
- Interpreting a language model
- AI safety in practice
- Responsible AI in practice
- An AI service in operation
- Build a RAG system you can trust
- Build an agent you can trust
- Build an AI service that survives production
- Build a memory system for an agent
- Multimodal systems
- AI in production
- Fine-tune and run your own models
- Build a voice interface
- Frontier Lab — an independent research project
- Evals in practice
- Reproduce a paper
- Deep reinforcement learning