← AI for Professionals: Using LLMs at Work
Lesson
Lesson 2: AI literacy, the EU AI Act, bias and transparency
Explain workplace AI literacy as a duty (including the EU AI Act Article 4 example) and identify ethics issues: data bias, transparency/disclosure, and copyright/attribution.
AI literacy as a professional duty
AI literacy as a professional duty
AI literacy means understanding what AI systems can and cannot do, and using them responsibly. In the workplace this translates to knowing LLM strengths — drafting, summarising, rephrasing, explaining, brainstorming, generating routine text or code — and their weaknesses: they can confidently state false facts (hallucinations), have a training-data cutoff, cannot do reliable arithmetic without a calculator tool, and cannot make legally binding decisions. A literate professional verifies AI output before acting on it and takes personal responsibility for the result.
The EU AI Act introduced a specific AI-literacy obligation in Article 4. As of 2 February 2025, all providers and deployers of AI systems — across all risk tiers, not just high-risk ones — must take reasonable steps to ensure their staff and contractors have a sufficient level of AI literacy. Enforcement is not expected before 3 August 2026. Ongoing EU Omnibus discussions may also soften the obligation from 'ensure' to 'support the development of' literacy, so the practical standard may evolve. This is educational context, not legal advice.
Three ethics issues arise constantly in professional AI use. Data bias: models reflect biases present in their training data, which can produce outputs that favour certain demographics or perspectives — always consider who might be disadvantaged. Transparency and disclosure: when AI substantially produces a document or decision, stakeholders often have a right to know; check your organisation's policy and any applicable regulation. Copyright and attribution: AI-generated text may echo passages from training data and cannot reliably cite sources it invented — verify and attribute factual claims independently.
The responsible pattern is: know the tool's limits, disclose AI involvement where required, check outputs for bias and accuracy, and give credit where it is due.
Lesson notes
AI literacy as a professional duty
AI literacy means understanding what AI systems can and cannot do, and using them responsibly. In the workplace this translates to knowing LLM strengths — drafting, summarising, rephrasing, explaining, brainstorming, generating routine text or code — and their weaknesses: they can confidently state false facts (hallucinations), have a training-data cutoff, cannot do reliable arithmetic without a calculator tool, and cannot make legally binding decisions. A literate professional verifies AI output before acting on it and takes personal responsibility for the result.
The EU AI Act introduced a specific AI-literacy obligation in Article 4. As of 2 February 2025, all providers and deployers of AI systems — across all risk tiers, not just high-risk ones — must take reasonable steps to ensure their staff and contractors have a sufficient level of AI literacy. Enforcement is not expected before 3 August 2026. Ongoing EU Omnibus discussions may also soften the obligation from 'ensure' to 'support the development of' literacy, so the practical standard may evolve. This is educational context, not legal advice.
Three ethics issues arise constantly in professional AI use. Data bias: models reflect biases present in their training data, which can produce outputs that favour certain demographics or perspectives — always consider who might be disadvantaged. Transparency and disclosure: when AI substantially produces a document or decision, stakeholders often have a right to know; check your organisation's policy and any applicable regulation. Copyright and attribution: AI-generated text may echo passages from training data and cannot reliably cite sources it invented — verify and attribute factual claims independently.
The responsible pattern is: know the tool's limits, disclose AI involvement where required, check outputs for bias and accuracy, and give credit where it is due.